2.1 Defining FinTech through Core Financial Functions

FinTech has become a catch-all term for innovations that leverage digital tools to deliver financial services more efficiently, inclusively, and transparently. In a seminal paper, Philippon (Reference Philippon2016) defines FinTech as covering

While this definition is prescient and informative, to understand the institutional context in which FinTech emerged, one must lay the groundwork by answering foundational questions from the past. First, what core functions does the financial system provide, and why are these functions central to economic activity? Second, what allocative or productive inefficiencies justify governments and regulators intervening in the supply and demand dynamics of financial services?

Economic theory identifies six core capabilities supplied by well-functioning financial systems, for example, see detailed explanations in Freixas and Rochet (Reference Freixas and Rochet2008). First, financial markets enable the clearing and settling of payments to facilitate commerce. From ancient barter systems to modern online payments, finance ensures that buyers and sellers can transact with confidence, minimizing counterparty risk (i.e., the probability that one party in a financial transaction will default on their contractual obligation, resulting in a loss for the other party) and transaction costs (i.e., all expenses incurred in the process of exchange beyond the price of the good or service itself such as search costs, bargaining costs, and enforcement costs associated with contract fulfillment).

Second, financial markets provide mechanisms for pooling funds to undertake large-scale projects. This pooling function makes it possible to assemble capital from diverse sources, ranging from individual savers to corporate treasuries to institutional investors. This pooled capital can then be channeled into ventures too large or complex for any single entity. Moreover, by subdividing ownership (i.e., through securitization processes such as those associated with issuing equity), financiers allow investors to diversify across multiple businesses, thus managing their exposure to risk from any single default or loss.

A third function of the financial system, particularly that of intermediaries, is transforming assets to facilitate better resource transfer across time or maturity, region or denomination, and industries. Finance allows households to save today for future consumption (e.g., via retirement savings), permits firms to invest capital in high-return projects (e.g., via corporate venture capital), and facilitates reallocating tangible and intangible assets when economic conditions shift. This intermediation balances the liquidity needs of firms and households while fostering a more dynamic economy. A fundamental feature of this resource transfer and reallocation is that the underlying assets have been transformed to better meet the demands of various actors.

Fourth, and related to the transformation of assets, is the transformation of risk-return profiles and the ability to manage risk. Financial instruments and markets offer many ways to manage risk. Through insurance contracts, derivatives, or loan syndication, parties can price, hedge, and distribute different types of risk. This bundling and decomposition of risk-and-return profiles helps financial and nonfinancial actors remain resilient to changing circumstances.

Fifth, financial markets contribute to price formation and informativeness. Stock prices, interest rates, and other market signals aggregate vast amounts of decentralized information about underlying economic prospects. When these prices reflect fundamental values, capital flows to its most productive uses. However, where informational frictions persist, the market may misprice assets, leading to instability.

Sixth, financial systems provide ways to mitigate incentive alignment problems when there is information asymmetry or agency costs. For example, banks have always had a role to play in managing the imperfect information provided by borrowers, such as the meaningful effort exerted by loan officers to gather soft details. Beyond asymmetric information, principal-agent relationships permeate finance, from shareholders delegating to empire-building managers to retail investors needing advice from honest, unconflicted financial advisors. Through novel contract and security design features, financial service providers help to align incentives.

The financial system provides a foundation for economic activity by fulfilling these six functions. Historically, banks have primarily been responsible for delivering these six functions. Banks were in this privileged position because their business model of taking in deposits and making loans enabled them to provide unique services like payments to the general public that were not as feasible for other business models without the advent of digital tools.

Summarizing these core capabilities enables a broad perspective of FinTech innovation. Namely, any innovation meant to improve or build upon these core financial functions is part of the “FinTech” landscape. From an innovation perspective, Schumpeter (Reference Schumpeter1934) distinguished between five types of innovations, which include (i) new products, (ii) new methods of production, (iii) new sources of supply, (iv) exploitation of new markets, and (v) new business models. What is striking about FinTech startups is that they often combine more than one type of innovation, especially outside of the conventional FinTech realm. For example, blockchain provides a new method of production, which, when combined with smart contracts, enables new business models that arguably have a competitive advantage because they can offer new, customized products or include those traditionally excluded from financial markets.

The transformative potential of FinTech, therefore, should not solely be judged by a single technical merit (e.g., better credit prediction from AI) but rather by its combined capacity to enhance the six core functions of any financial system. It is also true that this powerful interconnectedness among the innovations means that new core capabilities are emerging as part of the system. Given these developments, in the subsequent subsections, I will describe the core technologies underlying FinTech (i.e., AI, blockchain, smart contracts, digital assets, etc.) and their emergent combination (AI × crypto).

This functional perspective also helps explain why a future with potentially novel regulatory solutions may be the appropriate response in the FinTech era. Another unique facet in this era of FinTech innovation is that the innovations have not been limited to new entrants trying to work within the system. Instead, the entrants are trying to build a whole new financial infrastructure, typically on blockchains, rather than improving access to or making the existing traditional system more efficient. By creating a whole new infrastructure, especially one that is open-source and allows anyone to build upon it, the barriers to entry in financial services, such as high fixed costs from compliance, have been dramatically lowered. This, in turn, encourages even more entry and competition, ushering in meaningful change to the financial system. This innovation is typically associated with efficiency gains, yet novel and unpredictable risks, potentially necessitating a new definition of systemic risk (Adrian and Brunnermeier, Reference Adrian and Brunnermeier2016).

Before I introduce the specific technologies underlying this era of FinTech and data on their prevalence, I want to put their development in historical context, making it easier to see where traditional regulatory solutions for innovation are likely to face challenges.

2.2 Financial Innovation in Historical Context

It is well established that young, high-growth entrepreneurial firms are essential for innovation and economic growth. Indeed, innovation often serves as a competitive advantage for new entrants to gain market share and generate profits in established industries. In addition, regulatory barriers are known to prevent entry or distort prices in such a way as to inhibit entrepreneurship. Historically, the banking industry and the financial system generally have not had as many entrepreneurial entrants as other industrial sectors. Thus, it’s essential to understand how financial innovations come about and how much this current wave of innovations differs from prior ones.

Focusing on parallels and divergences in the pattern of financial innovation reveals that in past eras, a single type of innovation in the Schumpeter sense often emerged from incumbents to achieve efficiency gains in core financial capabilities, which contrasts starkly with this era of FinTech innovation. Yet both this era and previous eras of financial innovation seem catalyzed by either cost reductions associated with technological progress or in response to institutional change. Typically, a given technological advance in business methods reduced costs and drove efficiency gains in financial services, and these more efficient practices spread throughout the industry, influencing the overall efficiency. Shifts in the regulatory landscapes only occurred after increased adoption (Tufano, Reference Tufano2003; Lerner and Tufano, Reference Lerner and Tufano2012). In these cases, financial innovation emerged as a solution to frictions common in financial services, such as high levels of asymmetric information, transaction costs, or tax constraints.

As an example, consider how early financial innovation enabled faster transactions. There is no new product, but rather, there is a new method to produce it. The automated teller machine (ATM) exemplifies this type of early financial innovation, as it significantly transformed retail banking by automating transaction processing and providing convenience to customers. While ATMs reduced transaction costs and improved service accessibility, they reinforced banks’ roles rather than displacing them. Next, consider new products, such as a unique way to structure a security (e.g., tax-advantaged municipal bonds) or introducing credit cards. Again, like with ATMs, these products improved service accessibility. In these instances, no new core financial service was offered, and the underlying infrastructure (e.g., banks and intermediaries) remained the same.

Also, typically, these innovations were not the type of invention for which banks would seek IP protection. Thus, while credit cards transformed payment methods, they did not significantly alter the financial infrastructure. Notably, Bank of America issued the first credit card, but there was a belief that early credit cards may not have been considered novel enough for patent protection. Moreover, given the high cost of entry in the financial services industry, most financial institutions relied on other competitive advantages like their existing customer relationships, brand recognition, and distribution networks to fend off competitors. In fact, it was only after the 1998 State Street Bank decision that the financial services industry started to pursue IP protection regularly (Hall, Reference Hall2009).

From a historical perspective, the current wave of innovation is different for several reasons, but only a few reasons lead to regulatory differences. First, the locus of innovation has moved from banks being the innovators to technology-focused entities and FinTech startups being the innovators (Lerner et al., Reference Lerner, Seru, Short and Sun2024). This transformation marks a change whereby traditional banks have ceded innovative capabilities to tech-driven newcomers. Part of this may be that this cycle of financial innovation has focused less on business-centric applications and the business method improvements incumbents typically pursued and more on consumer-centric ones, where the user design experience is paramount.

To understand why this distinction alone does not necessitate regulatory differences, consider neobanks. These digital-only startups embody a single type of innovation, that of business model innovation. Neobanks essentially provide a more user-friendly interface to traditional banking services while still operating within the established financial system. Neobanks often use a bank-partnership model. The partner bank holds the deposits and issues the debit cards, while the neobank provides the customer-facing app, technology infrastructure, and customer service. Unlike traditional banks that rely on interest income, neobanks generate revenue through subscription fees for premium services, referral fees for third-party financial products, and fee income from specific services like expedited transfers. In this sense, it is clear that the regulatory scrutiny should focus on where bad actors are likely to engage, such as deceptive marketing practices and the exploitation of behavioral biases. These are all areas where regulators have experience, and the approach parallels easily. Yet most FinTech innovations now encompass multiple innovations in the Schumpeter sense.

Another reason this wave of innovation is different is that information itself has become the central commodity being transacted, not merely a facilitator of a financial exchange. The tension between more efficient information-gathering capabilities and strategic selection of information sets represents an economic trade-off that prior regulatory frameworks have not adequately addressed. When firms can selectively harvest consumer data at near-zero marginal cost, the resulting information asymmetry creates not merely a private advantage but a potential market failure. Similarly, when blockchain-based smart contracts need information external to the smart contract to execute the agreement (e.g., current exchange prices, which oracles can provide), understanding new risks associated with accessing that information (i.e., biased or manipulated representation) is critical. As is explored in Section 3.8, this information issue is central to the emerging FinTech technologies, including AI and blockchain.

Yet another feature that makes this wave of innovation so different is that the innovators are trying to create an entirely new financial infrastructure rather than simply improving access to or reducing friction in the existing system. In this sense, the current wave of financial innovation, which, at times, is trying to replace traditional intermediaries with automated protocols or smart contracts does not have natural regulatory parallels as the markets may provide similar core capabilities, but the infrastructure used to generate the capability doesn’t necessarily have the same centralized bottlenecks like a small number of brokers that current regulatory can focus on.

Perhaps, more importantly, and uniquely, this achievement of creating a new infrastructure has been done in a much more open-source way than any past innovation. This is making it a lot easier for anyone to build something new. This also enables customization and composability of financial primitives at an unprecedented scale. When such primitives interconnect in a Lego-like fashion, it allows for new core capabilities to be offered by the financial system (e.g., see my discussion of DAEs and entrepreneurship as a novel service in Section 3.7). By lowering the barriers to entry and innovation, the new financial system will likely be more efficient along some dimensions (e.g., less concentrated, so less rent extraction), yet, at the same time, the ease of entry will attract bad actors and speculators.

Along with technological innovation, institutional forces serve as powerful catalysts for financial innovation by establishing frameworks that shape market behavior and economic activity. An example of institution-driven financial innovation is the development of index funds in the 1970s. They emerged as a direct response to institutional constraints in the investment management industry. Before their creation, the prevailing institutional structure favored active management despite mounting evidence that most active managers failed to outperform market indices after fees. Thus, there was pent-up demand for a cost-effective investment vehicle. Recognizing this inefficiency, Vanguard’s founder created the first publicly traded index fund.

This example, however, reinforces the ideas we saw when technological progress rather than institutional changes catalyzed innovation. Early periods of FinTech innovation were characterized by a single type of innovation (typically product or process innovation). They occurred in response to persistent inefficiencies or failure to serve market participants optimally. However, they never created a new financial system, nor did they combine multiple types of innovation in Schumpeter’s sense.

In contrast, recent FinTech innovations introduce novel institutional arrangements and reshape how trust, governance, and control are managed in financial transactions. Unlike previous institutional structures, blockchain technology can reduce reliance on traditional intermediaries. Yet, these decentralized structures pose new regulatory challenges, from definitional questions such as what defines custody to deeper concerns related to accountability, risk management, and financial stability.

Combining AI with blockchain further compounds these complexities by increasing the potential for economic distortions. AI-driven predictive technologies can amplify systemic risks, create market manipulation vulnerabilities, and exacerbate informational asymmetries in these new financial markets. These distortions present challenges distinct from historical precedents, necessitating entirely new regulatory approaches.

Thus, while previous literature has acknowledged the complexity introduced by financial innovation, including its potential for substantial externalities, it does not speak to the specific distortions introduced when integrating multiple types of innovation in the Schumpeterian sense and which AI × crypto applications exemplify. As I argue in this Element, AI and blockchain are integrating, and therefore, it is helpful to view the regulation from an integrated approach. When combined, the economic distortions these technologies create can differ from individual ones, although we can learn from the individual ones. In this sense, the sum of the parts equals more than the parts individually because of these synergies, and new solutions are required to address the unique risks associated with this powerful combination.

2.3 Domains of FinTech Innovation

Below is a high-level overview of the major areas of FinTech that will form the backdrop for this Element and the dataset described below. The classification reflects common market segments and the interplay of emerging technologies with traditional financial services (TradFi).

First, there is conventional FinTech, which focuses on enhancing existing services rather than reinventing them from scratch. For example, peer-to-peer (P2P) lending platforms connect individual borrowers and lenders directly, reducing overhead costs and streamlining loan approvals. Similarly, Buy Now, Pay Later (BNPL) solutions aim to disrupt traditional consumer credit markets by offering short-term installment financing, often at the point of sale. Each of these developments addresses long-standing friction in cost and speed within mainstream finance and is the most similar to the ATM or credit card examples from history. Conventional FinTech shares the most parallels with earlier financial innovations (e.g., ATMs, credit cards) in that they tend to be process or product-oriented and are not recreating new infrastructure outside the traditional financial system, nor are they creating a new financial function.

A second category focuses on applying AI to core financial functions through automation, personalization, and predictive capabilities. AI-powered robo-advisors now provide personalized investment recommendations at a fraction of traditional management fees, democratizing wealth management services. In lending, AI can analyze real-time, big data beyond traditional credit scores to assess risk more accurately, expanding access to credit for underserved populations. Fraud detection systems leverage AI to identify suspicious patterns in real-time, dramatically reducing financial crimes while minimizing false positives that previously created friction for legitimate customers. Meanwhile, LLMs have transformed customer service through intelligent chatbots that handle routine inquiries instantly, allowing human agents to focus on more complex tasks. While AI promises efficiency gains and improved accuracy, it also raises concerns about transparency, data privacy, and potential biases embedded in its models.

A third category focuses on the blockchain and Web3 technologies that underpin cryptocurrencies like Bitcoin and Ethereum, but their use cases extend beyond digital currency. By providing a decentralized, tamper-resistant ledger, blockchain can expedite settlement times, reduce transaction fees, and enable automated smart contracts, which are self-executing code that enforces contract terms without central intermediaries. Digital assets have blossomed beyond simple payment tokens to encompass stablecoins and many other types, each with potentially distinct regulatory considerations (Cong and Xiao, Reference Cong, Xiao, Pompella and Matousek2021).

DeFi furthers blockchain’s potential, constructing financial products like lending, trading, and insurance on open protocols without traditional intermediaries. In DeFi, governance may lie with DAOs, whose rules and operations are encoded in software and collectively decided upon by token holders (Appel and Grennan, Reference Appel and Grennan2023). Finally, Web3 components like NFTs introduce digital scarcity into art, collectibles, and real estate. Their unique identifiers mean each token can represent a distinct piece of digital (or physical) property. Soul-bound tokens (SBTs) build on the scarcity introduced with NFTs by serving as identity-bound, nontransferable assets, which are valuable for credentialing or reputation systems.

Finally, an emerging frontier is the fusion of AI with blockchain technologies, which has colloquially been called “AI × Crypto.” Early ventures in this domain apply cryptoeconomic frameworks to accelerate AI development. Although major centralized AI providers have dominated the field through economies of scale, new projects are developing decentralized alternatives by experimenting with novel model architectures and incentive platforms. Broadly, four areas define this space: (1) decentralized compute networks offering pooled GPU resources for training and inference as highlighted by Cong et al. (Reference Cong, Tang, Wang and Zhao2023g); (2) coordination platforms incentivizing AI model development or enabling specialized inference approaches; (3) DAEs, which are AI agents and related tools and applications that can closely replicate traditional workflows and engage in entrepreneurial endeavors; and (4) decentralized general intelligence (DGI) which aims to produce an application layer similar to what has been put forth by OpenAI and Anthropic, namely, a productized AI solution for consumer or enterprise use.

2.4 Data on FinTech Innovation

To evaluate trends in FinTech innovation, I assembled data on FinTech startup entry from four major providers: Messari, Dove Metrics, Crunchbase, and Preqin – each of which has some advantages and caveats in terms of comprehensive FinTech data on funding, founding dates, and company-level details. For instance, Messari, which later acquired Dove Metrics, has the most comprehensive data on crypto startups, including unofficial announcements of funding via Twitter. Crunchbase and Preqin are more traditional sources of data for startups. Preqin has some of the most comprehensive fundraising data, but this gives it a bias toward firms that receive at least some VC funding, whereas Crunchbase typically has details even on young startups that may not have raised meaningful capital yet. By harmonizing across these sources, I aim to provide a more complete view of both traditional venture-funded startups (the standard route for conventional FinTechs), and those financed through more novel mechanisms, such as token-based financing. Online Appendix A details the initial aggregation.

To classify the startup’s business description into one of the FinTech domains, I employ a multistep classification procedure. The four categories or domains include traditional FinTechs, AI, Blockchain, and AI × Crypto. This process began with targeted keyword searches (e.g., blockchain, artificial intelligence, stablecoin, etc.) and was subsequently refined using large-language-model (LLM) filters tailored to identify nuanced business model references. Finally, I conduct manual checks for borderline or ambiguous cases, resulting in the final dataset. This multi-step approach balances scalability and rigor, enabling a detailed analysis of domain-level patterns while maintaining a high degree of classification accuracy. For the exact keywords and prompts used to classify, please see Online Appendix A.

2.5 Summary Statistics

Figure 1 provides a high-level view of how total disclosed FinTech fundraising has evolved worldwide and in the U.S. from 2012 through 2024. Globally, the total amount of capital raised that has been publicly disclosed has grown markedly over this period across all four subcategories of FinTech innovation: traditional FinTech, AI, crypto, and combined AI × crypto. Yet the relative contribution of each category diverges over time. In the early 2010s, funding was dominated by more conventional FinTech business models focused on payments, peer-to-peer lending, and robo-advising, while AI- and crypto-focused ventures initially raised only modest amounts. By the late 2010s and into the early 2020s, however, AI and crypto reached parity or surpassed traditional FinTech startups in terms of total funding, signaling an inflection point in investor preferences. Although starting from a small base, the AI × crypto segment displays especially rapid growth, reflecting demand for solutions that integrate these technologies. Notably, although U.S. fundraising volume is proportionally smaller than global totals, it echoes the same upward trajectory and category shifts, highlighting the prominent role of U.S.-based startups in shaping FinTech innovation despite meaningful regulatory uncertainty and related challenges.

Figure 1

Total disclosed FinTech fundraising (2012–2024). The figures depict the total disclosed FinTech fundraising globally and in the United States from 2012 through 2024. The business models of the funded startups are categorized into Traditional FinTech, AI, Crypto, and AI × Crypto.

Figure 1 Long description

This figure contains two bar charts showing the total amount of disclosed fundraising by any FinTech startup from 2012 to 2024, measured in billions of 2012 U.S. dollars. The top chart displays worldwide fundraising with values ranging from less than $10 billion in 2012 to peaks above $300 billion in 2021. Four categories of startups are shown in different colors: Traditional Fintech (tan), AI (light blue), Crypto (medium blue), and AI × Crypto (dark blue). The data show lower levels of fundraising from 2012 to 2015, gradual growth from 2016 to 2019, dramatic increases from 2020 to 2021, with 2021 reaching the highest peak, followed by declining but still substantial levels from 2022 to 2024 (doubling levels from 2016 to 2019). The bottom chart shows U.S. fundraising with the same categories and time periods, but lower overall values ranging from near zero in 2012 to approximately $60 billion at the 2024 peak. The pattern is similar to the worldwide data from 2012 to 2019, but departs in 2020. The U.S. shows its highest fundraising levels in 2024 rather than 2021, particularly driven by increases in the AI and AI × Crypto categories. In contrast, Traditional FinTech is the second-largest category worldwide in 2023. Both bar charts use stacked bars where each segment represents the contribution of each funding category to the total disclosed fundraising for that year.

Turning from the aggregate funding trends documented in Figures 2 to the specific startup counts, Table 1 provides a breakdown of the number of unique startups funded each year in the U.S. and globally between 2012 and 2024. Again, the counts are disaggregated by the four subcategories of FinTech innovation. In addition, the columns are further broken out by global versus U.S. totals. The data reveals several trends. First, traditional FinTech and AI startups show the largest absolute increases over time, reflecting the continued maturation of these technologies worldwide. Second, crypto-focused ventures exhibit a marked rise from modest beginnings in the early 2010s to increasingly substantial numbers by the mid-2020s, coinciding with the broadening of crypto into DeFi and other applications. Importantly, from a regulatory perspective, the AI × crypto category, although starting from a relatively low base, displays a steep upward trajectory both within the U.S. and globally. This latter observation highlights the growing convergence of AI-driven and blockchain-based innovations in the FinTech sector. This trend, along with other parallels between the branches of FinTech, requires a more integrated regulatory framework rather than separate oversight. Finally, the U.S. numbers, in particular, indicate that policymakers must be prepared to handle the compound challenges posed by AI × crypto startups, given the rapidly expanding presence of these startups and their revealed preference for U.S. locations.

Figure 2

Total disclosed FinTech fundraising by stage (2012–2024). The figures depict the total disclosed FinTech fundraising by stage of startup (seed funding, early-stage, or late-stage funding) for the global sample from 2012 through 2024. The business models of the funded startups are categorized into Traditional FinTech, AI, Crypto, and AI × Crypto.

Figure 2 Long description

This figure contains three line charts showing different stages of startup funding from 2012 to 2024, with the total amount of global fundraising measured in billions of 2012 U.S. dollars displayed on the y-axis and funding year on the x-axis. Each chart displays four categories of startups in different colors and patterns: Traditional Fintech (solid tan), AI (long-dash, light blue), Crypto (dotted, medium blue), and AI × Crypto (dashed dark blue). The top chart shows seed funding with values ranging from near 0 to $20 billion. All categories initially show lower levels of seed fundraising, followed by steady growth through 2019, and then sizable increases from 2020. Crypto funding shows the most dramatic increase, peaking at around 18 billion in 2022 before declining slightly. Traditional FinTech also peaks in 2022 at approximately $12 billion, while AI and AI × Crypto peak in 2024, reaching about $18 and $4 billion, respectively. The middle chart displays Early-Stage VC Funding with values from $0 to $80 billion. The pattern is similar but more pronounced, with AI showing explosive growth, reaching approximately $70 billion by 2024. The bottom chart shows Late-Stage VC Funding with values from $0 to $100 billion. This stage shows the most dramatic variations, with AI reaching nearly $100 billion in 2021. Traditional FinTech and Crypto both echo this volatile pattern, while AI × Crypto remains relatively modest, not exceeding $10 billion throughout the period. All three charts show a general pattern of low activity in early years, significant growth starting around 2016, peaks around 2021 to 2022, and varying degrees of decline or stabilization in the later years.

Next, I explore a more granular view of how funding flows vary by startup maturity. Figures 2 and 3 break down publicly disclosed FinTech fundraising into seed, early-stage VC, and late-stage VC categories. These figures build upon the startup counts provided in Table 1 by clarifying not only how many ventures are funded in each subsector but also how heavily they are financed at different points in their life cycles, which helps assess the sector’s trajectory and the potential regulatory touchpoints.

Figure 3

U.S. FinTech funding by stage of startup (2012–2024). The figures depict the total disclosed FinTech fundraising by stage of startup (seed funding, early-stage, or late-stage funding) for the U.S. sample from 2012 through 2024. The business models of the funded startups are categorized into Traditional FinTech, AI, Crypto, and AI × Crypto.

Figure 3 Long description

This figure contains three line charts showing different stages of startup funding from 2012 to 2024, with the total amount of U.S. fundraising measured in billions of 2012 U.S. dollars displayed on the y-axis and funding year on the x-axis. Each chart displays four categories of startups in different colors and patterns: Traditional Fintech (solid tan), AI (long-dash, light blue), Crypto (dotted, medium blue), and AI × Crypto (dashed dark blue). The top chart shows seed funding with values ranging from near $0 to $8 billion. All categories initially show lower levels of seed fundraising, followed by steady growth through 2019, and then sizable increases beginning in 2020. Crypto and AI each received about $5 billion in seed capital in 2024, with AI continuing to surge and Crypto exhibiting more volatility, having previously peaked at around $7.5 billion in 2022 before declining to under $4.5 billion in 2023 to rebound in 2024. Traditional FinTech also peaks in 2022 at approximately $4 billion, but does not rebound and continues to decline to less than $2 billion in 2024. AI × Crypto peaks in 2024, reaching about $2 billion. The middle chart displays Early-Stage VC Funding with values from $0 to $20 billion. The pattern is similar but more pronounced, with AI, Crypto, and AI × crypto all showing explosive growth and reaching new heights in 2024 while Traditional FinTech declines. The bottom chart shows Late-Stage VC Funding with values from $0 to $10 billion. Like the previous charts, Traditional FinTech diverges while all others are heading toward new peaks, although late-stage crypto projects are not raising funds nearly as much as early-stage ones.

Figure 2 focuses on global trends and reveals that, although seed-stage funding for traditional FinTech dominated early in the sample, AI startups commanded an increasing share of early-stage and late-stage funding by the late 2010s. Crypto investments, while initially small, gained momentum in the early 2020s, with AI × Crypto showing growth during the seed and early-stage rounds after 2020. For example, in 2023–2024, AI × Crypto exhibits near-parity with traditional AI in certain stages, suggesting enthusiasm for combined AI × Crypto solutions.

Figure 3 shows an analogous breakdown for U.S. startups, tracing similar patterns but with some notable differences in the relative magnitudes across subsectors. Seed funding for U.S. AI ventures, for instance, accelerated significantly around 2021 and overtook Traditional FinTech by 2022. Crypto and AI × Crypto remains smaller in the seed stage than globally, but has surged more prominently in early- and late-stage rounds, indicating that U.S. investors place a heightened emphasis on scaling ventures in these emerging technologies.

When comparing global versus U.S. trends, the principal divergence lies in the timing and intensity of early-stage investments in AI × Crypto, with dramatic spikes in recent years in the U.S. compared to more moderate increases globally. This suggests that the U.S. continues to play a dominant role in nurturing advanced FinTech innovation and startups, despite regulatory uncertainty. A second trend in the figures is that investor attention is shifting from smaller seed bets in toward more substantial later-stage commitments for narrowly focused innovations (e.g., either AI or crypto in isolation), reinforcing the need to develop cohesive regulatory structures that can adapt to the scaling FinTech ecosystem.

Building on the stage-specific insights provided by Figures 3 and 4, Table 2 provides a complementary perspective by detailing average funding round sizes (in millions of 2012 U.S. dollars) for each FinTech subcategory across seed, early-stage VC, late-stage VC, and token sale rounds. While overall funding amounts in seed rounds have grown notably over time (particularly for AI-focused ventures), early-stage and late-stage VC categories account for the largest share of capital, which makes sense since much of the ecosystem hinges on scaling established startups rather than seeding new entrants. Late-stage AI deals, in particular, stand out from 2018 onward, indicative of investor confidence in the potential for AI solutions to mature and expand. By contrast, Crypto and AI × crypto rounds, while experiencing phases of dramatic growth, display a more volatile pattern, potentially reflecting the still-evolving regulatory uncertainties they face.

Figure 4

Heatmap of FinTech firms globally by subcategory. The figures depict the total unique FinTech firms by country. The firm’s business model has been categorized into traditional FinTech, artificial intelligence (AI), crypto, and AI × crypto. The green figure represents traditional FinTech (upper left). The red figure AI (lower left). The blue figure crypto (upper right), and the purple figure AI × crypto (lower right). The darker the color, the greater the concentration of startups. The color gradient moves from lightest to darkest with breaks at 1, 10, 20, 50, 200, 1000, 2000, 5000, and 30000 firms.

Figure 4 Long description

This figure shows four world map heatmaps displaying the global distribution of FinTech firms by subcategory. The maps are arranged in a grid, each representing a different FinTech category, where color intensity indicates the concentration of startups by country. In the top left is Traditional FinTech, represented by a concentration map with green shading, where the darkest green indicates the highest firm density. The United States, parts of Europe (particularly the UK and Germany), China, and India have the darkest green coloring, indicating the highest concentrations of traditional FinTech firms. Other regions show varying lighter shades of green. The top-right map uses blue shading to represent the concentration of AI firms. Similar to the traditional FinTech map, the United States, China, parts of Europe, and India are colored the darkest blue. The pattern closely mirrors the traditional FinTech distribution but with some regional variations in intensity. The bottom left map displays Crypto startup concentration using red shading. The United States is depicted in a very dark red, indicating a high concentration. Parts of Europe, particularly the UK, also exhibit a dark red color. Other regions, including parts of Asia and some European countries, show moderate to light red shading. Finally, the lower right map shows the category of AI × crypto using purple shading. This map shows a more limited geographic concentration, with the United States being the dominant region. At the same time, most of the world remains very light purple or white, indicating lower firm density in this hybrid category. The color gradient moves from lightest to darkest with breaks at 1, 10, 20, 50, 200, 1000, 2000, 5000, and 30000 firms. The figure illustrates the total number of unique FinTech firms by country, categorized by business model type.

The inclusion of token sale funding highlights another important way FinTech startups can raise funds for their business. Although ICOs achieved notoriety between 2017 and 2018 due to a slew of fraudulent offerings before falling out of favor, particularly in U.S. markets, due to intensifying regulatory scrutiny (Lyandres and Rabetti, Reference Lyandres and Rabetti2023). Nevertheless, the data indicate that private and public token sales continue to provide capital for crypto startups worldwide. This underscores the continued importance of token-based fundraising for specific segments of the FinTech community, and it aligns with a need to rethink optimal approaches to regulating token sales.

Finally, building upon the preceding analyses of funding dynamics, Figure 4 offers a geographic perspective on FinTech startups by illustrating the global distribution of unique FinTech startups in each subcategory with traditional FinTech in green, AI in blue, crypto in red, and AI × crypto in purple. Each panel employs a color gradient, where darker shades indicate a higher concentration of startups, with thresholds ranging from 1 to 30,000 ventures. This visual representation accentuates the extent to which certain countries have become FinTech hubs, revealing both macro-level dominance and subcategory-specific strengths.

Two countries, the U.S. and China, display dense footprints across nearly all FinTech subfields. India, the United Kingdom, and Canada emerge as significant second-tier centers, outpacing most other regions in terms of FinTech startups. Russia, Singapore, and Brazil are also highlighted by darker gradients in specific subcategories, indicating entrepreneurial strength, albeit focused a somewhat narrower set of specific FinTech applications like crypto. Interestingly, the transition from conventional FinTech to the most recent AI × Crypto wave reveals subtle yet important shifts in geographic clustering. While numerous European countries exhibit some depth in AI or Crypto individually, the U.S. stands out for its incredibly high concentration of AI × Crypto startups, surpassing China and all other countries in this combined category while being on par in individual categories like AI. This suggests that the U.S. may occupy a relatively advanced position in integrating AI and blockchain innovations, reflective of a confluence of capital availability, technological expertise, and entrepreneurial dynamics that collectively shape its FinTech landscape.

3.1 The Supply and Demand Dynamics Driving FinTech Growth

The 2008 financial crisis is the primary institutional catalyst for the current era of FinTech innovation. The crisis revealed substantial shortcomings in regulation, highlighting how financial oversight struggled to keep pace with financial innovation, which was mostly centered around new products (e.g., new forms of securitization). This gap incentivized change through both necessity and opportunity. It prompted the development of new regulatory frameworks, entrepreneurial entry by FinTech startups, and questioning the financial system’s core risk management capability.

Specifically, in the 2008 financial crisis aftermath, a paradigm shift occurred in financial innovation, marking what, in retrospect, is the genesis of the modern FinTech era. The financial crisis shifted supply and demand. Namely, it increased the supply of talent working on financial inefficiencies and fundamentally changed the features consumers and investors demanded from their financial services providers. By exposing structural vulnerabilities within the traditional financial system, the crisis catalyzed both sides to realize that fundamental reforms were needed. Notably, given that certain financial innovations had contributed to systemic instability, there was a recognition that subsequent technological advancements would require a different approach.

On the supply side, this post-crisis environment reawakened those associated with the cyberpunk movement to work on solving the peer-to-peer money challenge. While many of the original pieces were there, it was the seminal whitepaper by Nakamoto (Reference Nakamoto2008), which introduced Bitcoin as a decentralized alternative to conventional financial infrastructure, and drew technical and engineering talent to financial system inefficiencies. The entrepreneurial talent in the high-tech sector started to reorient its expertise from cryptography and electrical engineering toward financial frictions and shortcomings of the system, therefore, it represents a defining characteristic of modern FinTech.

The financial crisis, however, did more than just shift talent. The 2008–2009 financial crisis, triggered by innovations in mortgage securitization, provides a stark illustration of how complexity and opacity can enable systemic risks to accumulate undetected. Securitization, which bundled residential mortgages into tradable securities, was initially designed to spread risk and lower borrowing costs. However, as Griffin (Reference Griffin2021) documents, this innovation became a vehicle for systemic risk due to pervasive misreporting, fraud, and conflicts of interest among key players. The complexity of these instruments obscured underlying risks while creating incentives for originators to extend credit to borrowers unable to sustain their loans.

This crisis experience shaped the subsequent FinTech revolution through a demand channel. First, the erosion of trust in traditional financial institutions created the demand for more trustworthy institutions. As such, business models emphasizing transparency and decentralization of power became more popular. Bitcoin’s introduction in 2008 explicitly positioned itself as a response to the crisis, offering a trustless, peer-to-peer financial system that would eliminate the need for potentially unreliable intermediaries. Second, increased regulatory oversight forced some banks to reduce traditional lending post-crisis. This created market opportunities for financial innovations that could improve access to financial services for these marginalized groups in a manner akin to the ATMs and credit cards of early innovation eras. For example, new platforms emerged to fill gaps in credit markets, with crowdfunding and peer-to-peer lending offering alternatives to traditional intermediation. Third, the 2008 financial crisis highlighted the importance of better risk assessment and monitoring tools. This demand for better intelligence, coupled with the massive cost reductions attributable to digitization, made it easy for executives to justify investments in the technologies that incorporate big data and advanced AI.

What are the massive cost reductions brought about by digitization? As Goldfarb and Tucker (Reference Goldfarb and Tucker2019) delineate, five cost changes came about because of digital technologies’ capacity to encode information as bits. Digitization reduces search, replication, transportation, tracking, and verification costs. This, combined with Moore’s law and Kryder’s law, which respectively posit the exponential growth of processing power and storage capacity, meant that computation and storage costs also declined. In such a state, it is natural to see an increase in the supply of applications using computationally intensive cryptographic proofs or big-data-driven AI algorithms.

3.2 Conventional FinTech Applications

Conventional FinTech applications, including digital payments, credit expansion via BNPL, peer-to-peer lending, and crowdfunding, are part of the first wave of FinTech innovations. Typically, traditional FinTech founders and innovators create a competitive advantage by introducing economic benefits such as reduced transaction costs and increased accessibility. However, these features that attract customers often pose novel regulatory challenges. For example, regulators face new concerns about consumer privacy and exploiting vulnerable populations. Despite conventional FinTech applications being incremental iterations of financial products and services that regulators are familiar with, the dilemmas they pose to regulators are similar to those brought about by more radical innovations like AI and DeFi. Therefore, a higher-level view of the economics of these conventional FinTech applications helps provide a valuable analogue to the economic issues that regulators face with the more radical innovations.

First, consider digital payments, which enable instant, remote transactions between two parties. These digital payments raise new privacy risks that do not exist with cash or traditional credit cards. Payment system operators can now collect detailed data on individual consumer purchases, including item-level data on what was bought, where, and when. This data can be combined with personal information like phone numbers, email addresses, and home addresses to build detailed profiles of consumers’ lives. Consumers are concerned about these infringements on their privacy, and over time, they have come to oppose their data being collected and shared (Goldfarb and Tucker, Reference Goldfarb and Tucker2012). Thus, a rationale for regulatory intervention in this setting is typically not about alleviating a specific constraint or distortion but about following the broad guiding principle of ensuring privacy. Thus, from a high-level perspective, it is worth evaluating how different ensuring privacy in payments is from settings like Meta’s more revolutionary attempt at a stablecoin or AI innovators’ decisions to include private data for training their models. Perhaps, surprisingly, the underlying economics are not that different.

At a high level, many U.S. laws and regulations encourage intervention if deception occurs. Now, consider this analogy. Just like large retail corporations cannot advertise a sale on TVs for $100 but only have one in stock, then lead the consumer who tried to buy the $100 TV to buy another similar yet much more expensive TV, in the FinTech context, startups should not be allowed to advertise convenience only to lead users to products that feature convenience bundled with data extraction, which is much more expensive to consumers. It is a classic bait-and-switch that exploits the underlying psychology of consumers. The lower advertised price “baits” the consumers, then the salesperson or AI-recommendation app “switches” the consumer to a higher-priced item. Therefore, users need to realize that whether it is a conventional FinTech application or a new Web3 application, some developers are attempting to deceive them, so they should exercise caution.

Insights from behavioral economics help to explain why consumers are vulnerable (Kahneman, Reference Kahneman2013). Consumers are more likely to trust new products and services when they are easier to understand. New technology often provides intuitive, user-friendly interfaces that make it easy for consumers to adopt without deep thinking. Yet thinking through the risks of using such technology requires one to allocate attention to effortful mental activities and make complex calculations (e.g., my probability of being a victim of a cybercrime from sharing my data). However, the Federal Trade Commission (FTC) has long classified certain deceptive or unfair advertising practices and prohibits violating those advertising rules. Since 1968, the FTC has not permitted bait-and-switch advertising. Typically, if the FTC takes action against a company for false advertising, they are likely to issue a cease-and-desist order, and in some cases, to go a step further and require that the company engage in corrective advertising, in which the company explicitly states that the former advertising claims were untrue. Thus, one potential regulatory solution could be to go through the FTC and encourage FinTech startups to stop doing this and engage in corrective advertising that would serve a FinTech literacy role. FinTech literacy as a regulatory solution is discussed in greater detail in Section 5.

Next, consider a similar scenario with a mobile payment app that enables purchases with a single tap, and a proposed regulatory solution in California. The payments app collects data on every transaction, including what was purchased, where, and when. The app uses this data to provide personalized recommendations and offers, enhancing the user’s shopping experience. From an economic perspective, the convenience provided by the app is valuable. It saves time, simplifies budgeting, and offers tailored discounts, making the shopping process more efficient and providing users with higher utility. Psychologically, the user is driven by the ease of use and immediate benefits. The app’s interface is designed to be intuitive, requiring minimal effort to operate. The instant gratification of quick payments and personalized deals appeals to the user’s present bias. The user might also trust the app due to its association with a reputable brand, reinforcing their willingness to share personal data.

What did California regulators do when faced with a trade-off between enabling the growth of efficient digital payment systems and protecting consumer privacy? California implemented an outright ban, as outlined in the Song-Beverly Credit Card Act. The parallel from that Act would be to prohibit payment systems and merchants from requiring or requesting personal information as part of payment transactions, unless the consumer affirmatively opts in. This would preserve the privacy status quo from the cash/credit card era and maintain a level playing field across payment types. However, it could slow the adoption of digital payments because specific business models would not be viable without the ability to monetize consumer data. In this sense, startup entry would be lower with such bans, which could reduce overall consumer welfare. Thus, if one wants to preserve entrepreneurial entry, a solution based on FinTech literacy may be more appealing.

Another area of innovation in payments brought about by FinTech companies is BNPL, which is a form of short-term consumer credit that allows users to split purchases into installment payments, often with minimal or no fees if paid on time. BNPL has seen explosive growth in recent years, with transaction volumes rising from $33 billion in 2019 to an estimated $120 billion in 2021. This growth has been driven by increasing adoption by both consumers and merchants. From a consumer perspective, BNPL can help alleviate liquidity constraints and smooth consumption. By allowing consumers to break up payments over time with limited upfront commitment, BNPL can make purchases more affordable, especially for those with limited savings or credit access. Consistent with this, Balyuk and Williams (Reference Balyuk and Williams2023) find that liquidity-constrained consumers are more likely to use BNPL and that BNPL access facilitates expenditure smoothing around income shocks.

However, BNPL also carries risks of encouraging overspending and leading to debt accumulation, especially among younger and lower-income users. Maggio et al. (Reference Maggio, Katz and Williams2023) show that BNPL access increases spending levels for all consumers, even those not liquidity-constrained. They argue this is hard to explain with standard intertemporal substitution motives. It is more consistent with the “liquidity flypaper effect” where the additional retail liquidity from BNPL sticks where it hits, fueling short-term spending.

From the merchant’s perspective, offering BNPL as a payment option can significantly boost sales. Berg et al. (Reference Berg, Burg, Keil and Puri2023) conduct a randomized experiment with a large e-commerce retailer and find that the mere availability of a BNPL option at checkout increases sales by approximately 20 percent. The spending response to BNPL availability is much larger than to the availability of other payment methods like PayPal. Interestingly, Berg et al. (Reference Berg, Burg, Keil and Puri2023) find that the merchant’s profits from increased sales facilitated by BNPL outweigh initial costs in terms of consumer defaults, helping to rationalize the rapid merchant adoption of BNPL. However, to the extent that the market becomes saturated and BNPL becomes so ubiquitous that it leads to defaults, the long-run gains may not be so advantageous.

Thus, like we saw with payments and regulators’ difficulty in achieving a broad mission of ensuring privacy when behavioral economics is considered, behavioral responses to BNPL suggest regulatory challenges. In the long run, evidence indicates the potential for welfare-reducing overconsumption, increased defaults, and worse credit in equilibrium. For consumers facing liquidity constraints, the additional purchasing power and flexibility of BNPL can smooth consumption and improve welfare. However, the same features may induce some consumers to overspend in ways that ultimately reduce welfare. For merchants, the sales boost from BNPL can outweigh the costs, unless defaults and other risks are poorly managed. From a regulatory perspective, regulators must grapple with the counterfactual. BNPL is far from first best because of behavioral tendencies, but it may be superior to a counterfactual of payday lending. Moreover, regulatory interventions that may require segmenting populations based on their behavior are complex, often leading to framing solutions like opt-out versus opt-in, rather than one-size-fits-all information disclosures.

Third, consider P2P lending platforms, which offer the potential to expand access to credit by enabling direct matching of borrowers and lenders via online marketplaces. P2P lending can serve borrowers who struggle to get loans from traditional banks. They may also offer better rates to borrowers and higher returns to lenders by reducing overhead costs and enabling more granular risk-based pricing. However, P2P lending raises concerns around credit risk, investor protection, and fair lending compliance. Current P2P platforms make minimal disclosures to lenders about how loans are underwritten. Lenders may not understand the risks they are taking on. There are also concerns P2P lenders could bypass consumer protection laws and enable predatory lending or unfair discrimination using non-traditional data.

Regulators must balance the goal of expanding access to credit against the need to protect potentially vulnerable lenders and borrowers. Overly restrictive regulations could eliminate the cost advantages of P2P lending. However, under-regulation risks enabling excessively risky or abusive lending practices. One approach could be to subject P2P platforms to disclosure and oversight requirements similar to traditional consumer lending, including standardized reporting, usury limits, ability-to-repay rules, and fair lending audits. This would create a more level playing field. However, it may increase the costs of running the platform and reduce the supply of credit that such platforms provide.

Another option would be to limit investment to sophisticated or accredited investors, reducing risks to vulnerable lenders. However, this would cut off P2P lending as an investment option for most households. Therefore, like with BNPL, targeted protection is needed for vulnerable populations, but targeting is often tricky. One potential solution discussed in 5 is RegTech, which would involve incorporating regulatory solutions directly into the product or service so that regulators can provide some form of protection or have knowledge to prevent systemic risk from excessively risky lending. Another might be the adoption of industry codes of conduct, which limit the exploitation of vulnerable populations, or if a startup wants to do business with such populations, requiring registration similar to what payday lenders must do.

Next, consider that some regulatory challenges conventional FinTech poses are nearly identical to well-established regulatory challenges in financial services. Consider crowdfunding, which reduces the cost of entrepreneurs accessing capital. Crowdfunding platforms, like Kickstarter, allow individuals to raise funds from many people via online campaigns. Compared to traditional early-stage financial resources for entrepreneurs, crowdfunding offers some benefits. It lowers search costs by providing a centralized marketplace where creators can pitch directly to a large pool of potential funders. It mitigates geographic constraints, allowing creators to access funding from anywhere globally. Further, receiving crowdfunding provides a new set of signals, like the number of backers, funding raised, and social media engagement, that can help founders, funders, and future investors assess quality.

However, crowdfunding also creates regulatory challenges because it is especially prone to distortions from information asymmetry and, to a lesser extent, behavioral economics (e.g., herding behavior). Investors on the crowdfunding platform often have limited information about creators’ true ability or intent to deliver, making it hard to screen for low-quality or even fraudulent campaigns. Agrawal et al. (Reference Agrawal, Catalini and Goldfarb2014) determine that this can lead to a market for lemons problem (Akerlof, Reference Akerlof1970). In particular, adverse selection drives high-quality projects toward traditional financing with stronger signaling mechanisms, while moral hazard incentivizes crowdfunding projects to exaggerate their prospects or misappropriate funds post-financing. Moreover, given that investors usually do not obtain formal control rights, this reduces incentives to monitor creators post-funding, which is one of the major benefits of traditional early-stage financing sources like VC (Hellmann and Puri, Reference Hellmann and Puri2002; Eldar and Grennan, Reference Eldar and Grennan2023).

So what can be done? Strausz (Reference Strausz2017) demonstrates that properly designed platforms can mitigate these distortions through conditional pledging mechanisms that enable consumers to implement deferred payments, thereby controlling entrepreneurial moral hazard. Other regulatory responses to these market failures have evolved along a common financial regulatory path, with policymakers requiring graduated disclosures and enforcing investment limits. The empirical evidence suggests this regulatory framework creates a complementary role for crowdfunding alongside traditional financing sources rather than a substitute relationship. For example, Chang (Reference Chang2020) shows that a fixed funding mechanism, where money is refunded if funding goals aren’t met, makes crowdfunding attractive to resource-constrained startups and established firms. Enabling these solutions affirms what practitioners have witnessed in the decades since crowdfunding was first introduced. Namely, better contracts and proper regulation in crowdfunding markets have enhanced efficiency by reducing demand uncertainty, allowing resources to be allocated to projects that might otherwise remain unfunded due to information frictions in traditional financing services.

To see the analogy between conventional FinTech and more radical innovations like crypto and AI, consider the parallels between crowdfunding and initial coin offerings (ICOs). ICOs allow entrepreneurs to raise funds through token sales, which may grant token holders various rights such as access to a future product, platform, or a share of profits (Lyandres and Rabetti, Reference Lyandres and Rabetti2023). Howell et al. (Reference Howell, Niessner and Yermack2020) show that ICOs are an effective mechanism for financing growth, with companies that raise more funding in their ICO seeing higher post-ICO employment and valuations. Lyandres et al. (Reference Lyandres, Palazzo and Rabetti2022) further find that an ICO’s success, again measured by the amount raised, positively correlates with survival and future financing. At the same time, ICOs also carry significant risks for investors (Lee et al., Reference Lee, Li and Shin2021). Many ICOs have proven to be scams or failed to deliver their promised products and services. The lack of standardized disclosures and oversight in ICO markets can make it difficult for investors to distinguish high-quality from low-quality projects. As discussed later, regulators have brought significant litigation and enforcement cases over whether tokens are securities that should be subject to the securities law.

In many ways, ICOs can be seen as crowdfunding with added risks. Why? ICOs leverage blockchain technology and tokenization to streamline issuing and trading of a novel type of “hybrid” security with some equity and debt-like features while introducing new and poorly understood risks, such as perverse incentives caused by early liquidity for founders and developer teams. Like other innovations in the conventional FinTech space, there could be benefits from ICOs, but determining how to protect tokenholders is hard. A good first start is to draw upon crowdfunding solutions like graduated regulatory frameworks, contractual solutions, enhanced disclosure requirements, and investment limits, all of which appear to help preserve the benefits of innovation without negative consequences. What will be interesting for regulators and lawyers is to consider the additional implications and risks that emerge from the combination of liquidity, price discovery, and programmable contracts. For example, many token issuers and users are learning that tokens with high fully diluted valuations and low initial circulating supplies can fall into zones that prohibit real progress on the development of the project and ecosystem.

Thus, while these conventional FinTech innovations may, at first glance, appear to operate through different mechanisms than AI or crypto, they are very similar. Conventional FinTech innovations increase access to financial services but also introduce nontrivial risks. Some economic distortions introduced by conventional FinTech are common challenges associated with informational asymmetry, for which well-established solutions exist. Other challenges, like privacy and exploitation of vulnerable populations, will require alternative solutions. However, there are many familiar solutions once the economic changes brought about by the innovation are appropriately framed. For example, regulations can be embedded into the point-of-sale through RegTech or more conventional pathways, such as using the FTC and making counter-advertising or FinTech literacy part of the penalty paid to the FTC. In each case, regulatory solutions are possible and made more evident by understanding the evolution and institutional context that preceded the innovation in the first place.

3.3 AI

Advances in AI are helping to decrease the cost of prediction and exploration (Agrawal et al., Reference Agrawal, Gans and Goldfarb2018). Reinforcement learning (RL), a sub-area of AI, allows agents to learn behaviors in an unknown environment by finding a balance between exploring uncharted territory and exploiting current knowledge. Modern LLMs use RL via RL from Human Feedback (RLHF), which has become a standard technique for aligning LLMs with human preferences and intentions. This approach is valuable for tasks with complex or ill-defined goals, such as determining what makes a response helpful or aligned with ethical values. Agentic AI combines RL with LLMs to create systems that can respond to unforeseen situations while adhering to basic rules and constraints. In this sense, RL teaches optimized writer and critic agents to refine the same output through repeated interaction. This back-and-forth, which can involve two or many agents, enables chain-of-thought reasoning. Thus, AI’s capability moves from simply predicting the next word to accurately predicting the type of judgment needed to interpret complex or ill-defined instructions.

Together, this set of cost reductions may encourage incumbents to substitute other inputs, such as labor for AI, as part of the production process. Similarly, reductions in prediction and exploration costs increase the value of complements, such as owning data, and of humans’ nonpredictive, nonexploratory skills, such as persuasion and social-emotional skills (Cao et al., Reference Cao, Jiang, Wang and Yang2022; Grennan and Michaely, Reference Grennan and Michaely2022). This theme of complementary skills implies that AI is a skills-biased tool. While AI can augment productivity, it favors experts with more expansive job skills. For example, a study of managers found that AI benefits the likeable yet incompetent manager, does little to improve the likeable and competent manager’s performance, and significantly reduces the perceived performance of the competent jerk (John et al., Reference John, Monnot, Mueller, Saleh and Schwarz-Schilling2024).

Decreases in the cost of producing predictions or allowing exploration also bring about more subtle yet unintended consequences for labor markets. Consider how much easier it is to improve resumes with AI tools and to send multiple applications. Evidence from the gig labor site, Upwork, shows that post-ChatGPT, there was an increase in applications per job. In the deluge of applicants, the high-quality freelancers who saw themselves getting less recognition adjusted their application strategy by applying to fewer jobs and tailoring themselves to specific skill niches to differentiate themselves relative to their prior behavior (Yiu et al., Reference Yiu, Seamans, Raj and Liu2024). In contrast, an expansion in the data provided on tech workers’ productivity (i.e., Github’s choice to display private domain commits, not just public ones) and thereby, the ability of employers to use AI to screen potential applicants productivity in greater detail, enabled positive assortive matching in the labor market (Gupta et al., Reference Gupta, Nishesh and Simintzi2024).

An economic consideration related to AI is privacy and the many data sources used by AI to train its prediction models, including alternative ones. As researchers search for ways to improve AI’s predictive accuracy, they collect data that humans generally would not want to reveal. Moreover, data collection and ensuring data integrity can be costly. For instance, researchers are creating annotated data sets with “visible thoughts” where humans describe what they thought would happen, what they believe is about to happen, what they are paying attention to, where the current sources of tension are in a situation, and so on. The goal of collecting data on annotated thoughts is to add an intermediary chain-of-thought prediction layer that helps to create more accurate AI-generated content. Yet, in making this data, the human advantages brought about by obfuscation are diminished. Suppose data with transparent thoughts were the key to more accurate predictions. In that case, it is possible that many researchers would attempt to collect such data, and inevitably, the data could be used to exploit that person’s emotional and psychological vulnerabilities. For example, antitrust scholars are concerned about using such data for price discrimination (Asker et al., Reference Asker, Fershtman and Pakes2022). Treating these models as if they understand meaning is challenging from a regulatory perspective because they are only sophisticated pattern-matching systems. At the same time, assigning liability to developers is equally fraught with challenges, suggesting that potentially new models of regulation, perhaps akin to those applied to animals, make more sense (for a detailed discussion, see Section 4.3).

Finally, it’s worth recognizing that the consequences of lower prediction costs can appear subtle but amplified through repetition. For example, AI is weak at predicting rare probability events such as tail events. In addition, AI is much more accurate at predicting events in the near future than in the distant future (Dessaint et al., Reference Dessaint, Foucault and Fresard2024). This tension across horizons may have unexpected consequences for innovation and capital allocation, key drivers of economic growth and consumer welfare. Consider consumer-facing industries, which may fall into the trap of using predictive AI to suggest iterative product improvements rather than exploratory AI to find new products. As Henry Ford famously said before the introduction of the first car, if you were to ask consumers what they want, they would tell you a faster horse. Predictive AI models are much more likely to offer incremental near-term solutions, although AI agents could encourage more creative solutions through the reduced cost of exploration.

3.4 Blockchain, Digital Assets, and Smart Contracts

Advances in blockchain are helping to decrease the cost of state verification (Yermack, Reference Yermack2017; Catalini and Gans, Reference Catalini and Gans2020). Digital assets like cryptocurrencies and tokens are issued and transferred using blockchain technology. Anything can be stored on blockchain technology. For example, a smart contract is computer code that automatically performs predetermined actions and is stored on a blockchain. Blockchains like Ethereum, Solana, and Cardano are specifically designed to support smart contracts as a core functionality. With smart contracts, it is possible to automatically exchange a certain amount of currency for another between two counterparties when a prespecified threshold is met. Suppose the smart contract code verifies the presence of the required threshold and the currency from each counterparty. In that case, the smart contract will execute the transaction, thus eliminating the need for third parties to facilitate it.

The cost reductions brought about by blockchain and smart contracts encourage those who need to engage in a transaction to adopt blockchains as a substitute for more costly forms of intermediation typically necessary to verify that certain conditions are met. This suggests that industries with high degrees of centralization, such as finance, may benefit the most from this cost reduction. Nevertheless, there may be barriers to adoption. For example, high-fee institutional arrangements often remain entrenched even in the presence of more-efficient alternatives (Judge, Reference Judge2015). Another barrier to achieving these benefits may be the hold-up problem, which occurs when multiple participants bring together valuable assets to do business, but one or more of the participants must make relationship-specific investments. Such relationship-specific investments put those who contribute early into an unfavorable negotiating position later on, which allows later participants to take advantage of the early participant (Holden and Malani, Reference Holden and Malani2021). These hold-up problems are a common economic theme that emerges across financial innovations as discussed in Section 3.8.

Despite the benefits of decreases in the cost of state verification, there may be unintended consequences. Some blockchains are permissionless, meaning the validation of state is part of the computer code, so no intermediaries are necessary. Permissionless blockchains commonly use proof-of-work (PoW) or proof-of-stake (PoS) to ensure security. Both approaches have advantages and disadvantages (Budish, Reference Budish2018; Cong and He, Reference Cong and He2019; Saleh, Reference Saleh2020; Abadi and Brunnermeier, Reference Abadi and Brunnermeier2022). PoW is bad for the environment as it requires lots of energy; PoS is susceptible to large stakeholder collaboration (e.g., monopoly-like actions) or distortions brought about through intermediation chains (Azar et al., Reference Azar, Casillas and Farboodi2024) given that only a handful of validators participate in mining the blocks. Alternative consensus mechanisms, such as proof-of-authority (PoA), proof-of-access, or proof-of-personhood (PoP), have yet to achieve broad adoption.

To understand the full extent of the negative consequences potentially imposed by blockchain applications, consider the noteworthy example of Meta’s attempt at launching a stablecoin (i.e., Libra/Diem). Much like the Sarbanes-Oxley Act in 2002 was a reaction to Enron, or Dodd-Frank Act in 2010 was a reaction to the financial crisis, arguably, recent regulation and proposed legislation appears to have been a direct reaction to the idea of Libra/Diem (e.g., while this study focuses on the U.S., the European Union’s Markets in Crypto-Assets Act (MiCA) regulation surrounding asset-referenced tokens appears to be a direct response). Before the sale and eventual abandonment of Libra/Diem, the regulatory concern was that Meta, a company with a repository of social data covering approximately three billion people worldwide, could match that social data with transaction-level spending data. This could lead to systematic violation of privacy, exploitation of behavioral tendencies, and ultimately, a form of power that could compete with sovereigns.

Moreover, if algorithmic or fiat-based stablecoins achieved widespread adoption, this could introduce new risks to the financial system (Gorton and Zhang, Reference Gorton and Zhang2021; Catalini et al., Reference Catalini, de Gortari and Shah2022). Some of the main concerns justifying financial regulator scrutiny are: (i) the potential for run externalities, where a rapid sell-off of one stablecoin could trigger systemic problems (e.g., similar to what occurred with Terra/Luna), and (ii) the gradual reduction in traditional bank deposits that might occur without consideration for the externality this would impose given the role these deposits play in the payments system (Bertsch, Reference Bertsch2023). As U.S. regulators prepare stablecoin legislation, additional political and economic concerns around potential negative consequences of stablecoins are emerging, ranging from the loss of monetary policy autonomy and thereby, the inability to stabilize local business cycles, to reduced efficacy of economic sanctions, to potentially dampening feedback effects such as existing institutions catering to stablecoin issuers preferences which inadvertently reduce real economic activity.

Much of the public discourse about cryptocurrency regulation has focused on the dangers of cryptocurrencies. These risks exist, and there is a need for targeted regulation to address potential harms that flow from this. However, it is equally important not to overemphasize these risks and lose sight of the substantial benefits, economic or otherwise, that may be gained through legitimate applications of the technology. In fact, despite the potential for negative externalities, other blockchain applications beyond stablecoins are more complementary to financial services. Consider Ripple’s RippleNet which transforms cross-border payments by replacing Society for Worldwide Financial Telecommunications (SWIFT)’s error-prone manual system with an instant, traceable network that reduces transaction costs and eliminates delays. Further, Ripple’s on-demand liquidity feature leveraged its token (XRP) to eliminate the need for prefunded nostro/vostro accounts, freeing up capital while providing smaller financial institutions a competitive advantage in global transactions.

Another benefit of decreasing the cost of state verification via blockchain is that the cost savings can be passed to the customer. This is important for customers who face extraordinarily high fees, as the reduction in cost could be what brings them into the financial system. As with the Ripple example, tremendous costs are associated with settling international transactions. The average transaction fee for cross-border payments is around 10% and has an average clearance period of two to three days. In contrast, various blockchains enable users to transact quickly and at nearly zero cost. By dissolving the financial barriers between developed and developing nations and offering a low-cost network of frictionless international payments, blockchains have the potential to bring financial services to billions of people who would otherwise be left behind. While solutions have been created outside of the financial system, like Transferwise, blockchain-based solutions can provide cost savings even for transactions that require provenance.

Another specific benefit of blockchain is its potential to catalyze innovation because it helps solve the problem of hidden intangible value. From an accounting perspective, only patents, noncompetes, customer lists, and other intangible assets that can be tied to a specific cash flow stream are typically recorded as intangible assets on the balance sheet. Otherwise, it is only through an acquisition that the intrinsic value of an organically developed intangible asset (e.g., influential brand) converts to a market value and is assigned a value on the balance sheet. Interestingly, smart contracts can be adapted to potentially transform any transaction that requires intermediated validation, such as certifications of authenticity or future cash flows. This certification of authenticity is at the core of incentivizing content creation and/or unlocking the idle collateral value of these organically created intangible assets awaiting sale.

Finally, machine-to-machine payments enabled by blockchain-based technologies are beneficial because these micropayment systems can mitigate market distortions. Basic Attention Token (BAT) exemplifies this benefit in digital advertising, where it creates a direct value exchange between content creators, advertisers, and users based on actual attention metrics. By allowing browsers to be automatically compensated by publishers for content consumption, yet only requiring advertisers to pay for verified attention, BAT eliminates the ability of tech platforms to extract a disproportionately high share of the rents simply from enabling a match between consumers and products. This direct machine-to-machine payment system reduces fraud, improves targeting efficiency, increases equity for content creators, and restores proper price signals to a market previously distorted by opaque algorithms and centralized gatekeepers. The result is a more transparent ecosystem where resources flow according to genuine value exchange rather than platform leverage.

3.5 DeFi and DAOs

DeFi builds upon the advances brought about by blockchain. The analogy to blockchain’s reduction of the cost of state verification and AI’s reduction of the cost of prediction and exploration is that DeFi and DAOs reduce the need for labor-driven intermediation (e.g., underwriters or managers), concentration of power (e.g., systemically important financial institutions), and enforcement costs. DeFi protocols often run on smart contracts. DAO governance runs on smart contracts. Since smart contracts create binding, automated agreements that execute precisely as programmed without requiring trusted intermediaries, the DeFi contract’s commitments reduce counterparty risk and enforcement costs. From an aspirational point of view, one could argue that DeFi and DAOs minimize the cost of building and/or coordinating complex financial and managerial services.

Yet, as we will see, DeFi and DAOs also introduce new coordination costs because a distributed structure is costly to maintain. Further, while the promise of DeFi is often presented as fully automated and trustless, human involvement remains necessary in many aspects, creating complications that challenge the ideal of pure commitments through code and reduced costs of coordinating complex activities. In particular, complications that continue to arise even in this new, supposedly more efficient setting are governance challenges in DAOs, coding errors in smart contracts, potential for manipulation when accessing external data from oracles, and the need to interact with traditional legal and regulatory systems should emergencies arise (e.g., hacking). These human elements in DeFi create a tension between the theoretical ideal of pure code-based commitments and the practical reality where human judgment, coordination, and trust are still required at various levels. This is why some critics argue that many DeFi projects are not genuinely decentralized but represent different trust and authority arrangements.

Despite these caveats, DeFi does allow parties to create precisely tailored and highly complex economic arrangements that execute via smart contracts almost instantaneously and ideally without the need for an intermediary or other trusted party. Moreover, all DeFi transactions are recorded on the ledger of a public blockchain network. Blockchains are useful not only for settling transactions but also for coordinating activity. Thus, an increasingly popular way to manifest coordination is through DAOs. Given the cost reductions and efficiencies associated with blockchains, developers can create transparent organizations, referred to as DAOs, that are less hierarchical in nature. These DAOs use smart contracts to prevent people from stealing pooled assets or engaging in other forms of agency costs, thereby creating trust within the organization. DAOs often give people the ability to use smart contracts in voting-based systems to allow users to provide judgment or governance in a more decentralized way. In this sense, one may solve the classic agency cost problems first described in Berle and Means (Reference Berle and Means1932) by creating headless organizations not run by a board or professional managers.

The products and services associated with DeFi often resemble offerings from traditional, centralized financial markets. For example, users can obtain a loan upon posting collateral, much like conventional collateralized loans. Yet DeFi takes on many additional forms, including asset trading and derivative transactions. While the earliest applications were built on blockchain-based tokens or stablecoins, recent applications incorporate real-world assets (RWAs), such as real estate, IP rights, and private credit. Similarly, the original DAO was formed in 2016 and operated much like a VC fund, by pooling assets and voting in a decentralized manner on potential investments to make profits. Ultimately, the original DAO was shut down due to a human coding error, but many new DAOs are forming, including those that have limited liability (LAOs) and larger user bases, such that they are operating similarly to a community of users with a bank account.

The key to offering DeFi products and services cheaply is the fact that they are open source, interoperable “stacks.” The stacking or layering of a base blockchain with protocols, assets, and ultimately, user-facing aggregation layers, re-creates complex financial services by modularizing the input components. Moreover, since there are so many different possibilities for combining modularized components, some DeFi service providers allow users to pick and assemble a fully customized set of services. In fact, the variety of finely tailored products that can be created is extraordinary and likely to grow. Consider that there are efforts to combine predictive AI, such as those determining what financial services a user needs, with a fully customizable set of DeFi services rather than a few prepackaged offerings, such as those currently offered by robo-advisers (D’Acunto et al., Reference D’Acunto, Prabhala and Rossi2019). AI × crypto applications, such as these, are discussed in more detail in Section 3.7.

DeFi products and services can be cheaper and more customizable, so they serve as an attractive substitute to traditional financial intermediation. In particular, boilerplate language that bankers and lawyers routinely place into transactions can be converted to code, substantially reducing the need for expensive human capital. Yet, DeFi and traditional intermediation are not perfect substitutes. In particular, relative to traditional financial services, DeFi applications incorporate new risks (Schar, Reference Schar2021). In part, the risk with DeFi arises because regulatory and compliance practices differ on-chain from what happens in the traditional system. For example, there is no deposit insurance on DeFi, prices are volatile, and the whole system is subject to financial crime and cyber risk.

The novel risks associated with DeFi may resolve themselves through continuous improvement of the DeFi product, or create a segmented market if incumbents do not embrace the underlying technology in the same way they have for AI and private blockchain applications. As an example of an improvement against cyber risk, consider decentralized exchanges (DEXs), which enable direct peer-to-peer cryptocurrency trading without intermediaries, primarily using automated market makers (AMMs) instead of order books. Trading on a DEX reduces some risk because users maintain custody of their funds instead of placing them directly in a custodian’s control. Yet a constant product formula (x × y = k) is commonly the mathematical foundation for many DEXs like Uniswap, where x and y represent the quantities of two tokens in a liquidity pool, and k is a constant that must be maintained after every trade. This mechanism automatically adjusts prices based on supply and demand, creating a self-balancing market that operates entirely through smart contracts. While using DEXs solves for custody risk, they introduce entirely new risks like price slippage for big trades and front-running on users’ pending transactions. Thus, just as with AI and other blockchain applications, we are seeing that advances in DAOs and DeFi applications eliminate certain economic inefficiencies while simultaneously creating new economic distortions. These distortions may resemble familiar problems (like front-running) or represent novel inefficiencies.

3.6 NFTs, SBTs, RWAs

NFTs represent ownership of a digital or nondigital asset and can be created on most public and private blockchains, although they are most popular on Ethereum. They differ from traditional fungible tokens because they cannot be exchanged for an identical asset. Thus, NFTs represent an innovation in digital property rights because they solve a key challenge brought about by Web2, namely, ownership and content creation verification. While the previous section outlined how blockchain generally reduces verification costs, NFTs specifically solve the double-spending problem in digital content markets. The double-spending problem in digital content markets refers to the inherent challenge that digital content, unlike physical goods, can be perfectly copied and distributed infinitely without diminishing the original. This creates a fundamental economic obstacle. If a digital asset can be replicated without cost or quality loss, establishing scarcity becomes impossible through technical means alone, undermining traditional ownership models and creators’ ability to monetize their work. Put another way, prior to NFTs, digital content could be perfectly replicated at zero marginal cost, enabling widespread distribution but undermining creators’ ability to capture value. This market failure led to the rise of platform intermediaries, who aggregated and monetized content, while creators often captured minimal economic rents.

NFTs resolve this tension through a novel combination of technological and economic mechanisms. By enabling verifiable scarcity in digital goods, NFTs create artificial rivalry in otherwise nonrival digital goods. Thus, while NFTs do not prevent the copying of the underlying digital content itself, they establish a cryptographically secure record of ownership for a specific token that represents the official or authentic version of that content. The immutable nature of the blockchain means that the NFT’s blockchain-based record can serve as a solution to digital asset ownership. These benefits help to transform the economics of digital content markets in several fundamental ways. First, NFTs enable direct monetization of digital creation without requiring traditional intermediaries. Second, NFTs create secondary markets for digital goods, enabling price discovery and speculation. Third, NFTs enable new forms of bundling between digital ownership rights and ancillary benefits or governance privileges. Fourth, NFTs facilitate the digital–physical divide by providing a pathway forward for tokenizing unique RWAs like real estate.

NFTs have moved beyond the digital art applications popularized by Beeple, providing a useful mechanism for converting previously illiquid intangible assets into tradable financial instruments. However, research reveals significant market inefficiencies with NFTs that may extend to other assets. For example, Huang and Goetzmann (Reference Huang and Goetzmann2023) document investor over-extrapolation bias and selection effects when sales increase and prices rise for NFTs. White et al. (Reference White, Wilkoff and Yildiz2022) establish correlations between NFT news and subsequent returns, and Borri et al. (Reference Borri, Liu and Tsyvinski2022) demonstrate how rarity impacts NFT valuations. These findings parallel historical periods of financial innovation, where the novelty interacted with users’ behavioral tendencies in ways that were not welfare-enhancing.

Despite these potential inefficiencies, NFTs have particular relevance for IP markets, where traditional copyright and patent systems face increasing strain from digital technologies. NFTs enable automated usage rights enforcement through smart contracts, potentially reducing monitoring and enforcement costs in IP markets. Moreover, the ability to fractionalize IP ownership through NFTs creates new possibilities for risk sharing and capital formation in creative industries. In the case of NFTs, fractionalization is often used as it allows smaller investors to pool resources to purchase fractional interests of an NFT. For example, there are platforms exclusively focused on fractional investing and enabling the purchase of truly illiquid assets like the full skeletons of dinosaurs.

However, these benefits come with some economic tradeoffs. Sockin and Xiong (Reference Sockin and Xiong2022) highlight how artificial scarcity in digital goods can create deadweight loss through reduced access. This raises questions about the optimal balance between creator incentives and social welfare in digital content markets. Additionally, the speculative nature of NFT markets, especially for art and other collectibles, can lead to bubble dynamics and coordination failures, particularly when value depends heavily on social consensus rather than fundamental utility. In particular, Oh et al. (Reference Oh, Rosen and Zhang2024) propose a framework for understanding NFTs as digital Veblen goods: consumers demand them partly because other consumers do. This makes demand for NFT collections fragile, and issuers respond by underpricing their NFTs in primary markets, creating profit opportunities for scalpers.

At the same time, there seem to be some promising applications for NFTs, especially connected to RWAs. For example, consider the benefits in the real estate context. A piece of real estate represented by an NFT allows for the transfer of ownership, which can be executed automatically via a smart contract when the NFT is sold. It is well established in finance that securitization and equitization have contributed to more efficient public markets by enabling fractional ownership, investor accessibility, and improvements in liquidity, and it appears in some instances that these benefits can be extended to a more extensive set of activities. However, these services have primarily been focused on more homogeneous assets associated with public market exchange. NFTs are helping to extend the concept of fractionalization to traditionally illiquid private markets. Thus, RWA tokenization represents a significant innovation in financial market infrastructure, enabling the representation of traditional financial assets like Treasury bonds, real estate, and private credit as digital assets on blockchain networks.

As of December 2024, the total value of tokenized RWA reached $15.3 billion in Assets Under Management (AUM), with private credit ($9.1 billion), treasuries ($4.1 billion), commodities ($1.1 billion), alternative investments, such as collectibles ($0.5 billion), institutional funds, such as opening up private equity giant KKR’s Strategic Healthcare Fund through tokenization ($0.4 billion), stablecoins backed by RWA-collateral other than traditional fiat ($0.2 billion), and real estate ($0.2 billion) according to data from RWA.xyz. The RWA sector has exhibited remarkable growth, expanding from $2 billion in early 2023 to its current scale. The market’s development has been driven by major institutional participants, including BlackRock’s BUIDL Treasury fund and Franklin Templeton’s tokenized asset initiative, suggesting a structural shift in how traditional financial institutions approach digital asset markets. This rapid institutionalization of RWA tokenization indicates the desire for tradability, accessibility, and liquidity in these traditionally illiquid, nonhomogeneous private assets.

Another innovation in tokenization from an economic perspective is SBTs, which extends the concept of a unique ownership of a digital asset to nontransferable assets such as digital credentials. As explained by Ohlhaver et al. (Reference Ohlhaver, Weyl and Buterin2022), SBTs address fundamental economic problems in identity and reputation markets. Traditional credential systems suffer from several market failures: high verification costs, limited portability, and vulnerability to falsification. SBTs potentially resolve these issues through cryptographic verification and permanent wallet association. The economic implications of SBTs are particularly relevant for labor markets, professional certification, and social cohesion. By enabling credible, portable proof of skills, experience, achievements, or social status, SBTs could reduce information asymmetries in hiring, contracting, and trust-building. Weyl et al. (Reference Weyl, Thorburn, de Keulenaar, Mcchangama, Siddarth and Tang2025) demonstrate how SBTs might reduce screening costs and improve match quality in social media for purposes of trust building. These ideas can also be applied to labor markets as the nontransferability of SBTs creates novel interactions with traditional models of human capital development and signaling. For example, it could lead to positive assortative matching as high-moral or high-productivity workers match with those firms with similar traits.

Next, IP law could evolve in response to easily traceable, immutable ownership records. Traditional copyright law developed around assumptions of rival, excludable physical goods and centralized enforcement mechanisms. NFTs and SBTs challenge these assumptions by enabling new forms of digital property rights that are simultaneously public (visible on blockchain) and exclusive (controlled through private keys). This creates room for the law to be modified and new ways for thinking about the interaction between on-chain and off-chain rights enforcement. The implications for copyright law are particularly complex. While NFTs can represent ownership rights, the relationship between token ownership and traditional copyright privileges remains ambiguous. Bodó et al. (Reference Bodó, Gervais and Quintais2018) analyze how smart contracts might automate copyright licensing and royalty distribution, potentially reducing transaction costs in creative industries. However, this automation also raises questions about fair use, first sale doctrine, and other traditional concepts in IP law (Kappos et al., Reference Kappos, Bennett and Mariani2023). Some advocates argue this would allow one to easily track ownership and therefore, we should include a percentage of the resale back to the creator, which some platforms are experimenting with.

Looking forward, it is clear that NFTs, like other FinTech innovations, solve for one market failure while introducing new ones. Previously, there was a market failure in which creators often captured minimal economic rents because they could not establish scarcity. However, by artificially limiting access to digital goods, NFTs create deadweight loss through reduced access, where some consumers who value the good above its marginal cost but below market prices are excluded. Further, behavioral biases in NFT markets, such as investor over-extrapolation and the digital Veblen goods phenomenon, parallel conventional FinTech concerns about exploiting vulnerable populations. Finally, the ambiguous relationship between on-chain NFT ownership and traditional copyright privileges introduces new forms of coordination failure. These new economic distortions suggest that targeted regulatory approaches, such as distinguishing between consumptive and investment NFTs, clarifying IP rights, and preventing market manipulation, would enhance efficiency without stifling innovation in this emerging asset class.

3.7 AI × Crypto

While capital started to be meaningfully raised for AI × crypto projects in 2020, primarily in the data and compute space as highlighted by Cong et al. (Reference Cong, Tang, Wang and Zhang2022), the generative AI boom of 2022 catalyzed crypto developers to embrace the intersection of the two technologies and integrate even more enthusiastically. So far, most projects can be grouped into one of four AI × crypto subcategories, including compute, data, models, and applications. Compute is the earliest and largest subcategory in this space, with multiple projects creating sizable GPU marketplaces. For example, founded in 2017, Render Network is a high-performance distributed GPU rendering network that leverages software to facilitate a compute marketplace between GPU providers (e.g., idle home computers when you are at work) and GPU requestors (e.g., major media companies like Disney and HBO, who have already signed deals with Render). The economics underlying Render are the same peer-to-peer economics driving the “sharing economy” in other areas like real estate (i.e., Airbnb) and vehicles (i.e., Uber).

The idea behind compute AI × Crypto business models is to create a two-sided marketplace for something that otherwise is not realizing its full value, such as idle compute. GPU owners monetize idle compute capacity that would otherwise go unused, creating new revenue from existing assets with minimal additional costs (i.e., mainly energy costs). This benefits creators, who are the primary target audience of Render Network, given their GPU-intensive rendering needs. Specifically, the creative users gain access to distributed GPU power without the capital expenditures of purchasing high-end hardware or committing to fixed cloud computing contracts.

Suppliers also benefit from the two-sided platform model as it can compete with centralized compute companies through lower costs, dynamic pricing, and scaling flexibility. Specifically, by tapping into underutilized GPUs, the network leverages existing hardware rather than requiring new infrastructure builds. At the same time, pricing can be dynamic and adjust based on supply and demand, potentially offering lower rates than traditional cloud rendering services, especially during periods of high supply. Relatedly, the market can naturally expand or contract based on market conditions. These features are important. The incremental compute supplier, who is being incentivized to add capacity at peak times, is much more agile than a centralized compute business, which needs to predict demand ahead of time and make large, up-front investments to build capacity.

Interestingly, like other leaders in the blockchain space (e.g., Ethereum Foundation, Solana Foundation, Stellar Foundation, Optimism Foundation, etc.), Render Network has structured itself as a Foundation but has also established a DAO to serve as a decentralized governance mechanism that enables all community members to participate by submitting and voting on Render Network proposals. This focus on allowing the community to directly influence the future development and operation of this decentralized GPU computing network stands in stark contrast to some computing networks’ seemingly black-box, authoritarian nature. These embedded governance features are more than just “vibes” as they serve a dual marketing purpose for the protocol.

The synergies between crypto and AI in the compute area are part of a larger swath of the blockchain space known as Decentralized Physical Infrastructure Networks (DePIN). The idea behind DePIN is to incentivize individuals and organizations to contribute to and manage physical infrastructure like energy grids, transportation networks, or computing power in a decentralized manner, essentially allowing anyone to participate and earn rewards for contributing resources to the network through tokens. Many business models are targeting generative AI’s need for compute with their services. For instance, Aethir is a compute DePIN that connects GPU providers with enterprise clients needing high-powered chips for professional AI tasks. Helium, Akash, Storj, and Flux all offer similar business models focusing more or less on some specific niche compute areas. A broader application of DePIN is to companies like Roam, which aims to address the structural challenges facing telecoms by combining decentralized wireless infrastructure with tokenization. Through a global WiFi OpenRoaming network, Roam offers automatic, secure connectivity across millions of access points, replacing manual logins with blockchain-based authentication.

Another focus in the AI × crypto space is enabling access to high-quality data sources. As companies race to create the best frontier AI model, the total universe of human-generated data is estimated to be depleted between 2026 and 2030. While humanity will not run out of data, it is the kind of clean, well-labeled information vital for training advanced AI models that may become scarce within the next few years. In economic terms, the demand for high-quality data is rising faster than its readily available supply, creating a need for better data marketplaces and novel privacy-compatible exchange mechanisms.

Thus, similar to Render’s solution in the compute space for using idle compute, efficiency gains in the data space are coming from crypto-enabled data exchanges that make use of currently siloed data. Traditional data markets often suffer from high transaction costs and lack of transparency. Blockchain-based platforms, however, can reduce these frictions. This is made possible by three key advantages: micropayment infrastructure, enhanced property rights, and reduced intermediation costs. First, crypto enables machine-to-machine transfers of value at near-zero cost. By reducing the transaction costs associated with compensating small-scale data contributions, crypto helps expand the potential data supplier market. Second, by codifying data ownership on a blockchain, data contributors receive property rights typically ignored in Web2 and, thus, can be assured of fair compensation and continued control over how their data is used. This arrangement encourages the supply of higher-quality data because contributors trust the system to reward them appropriately. Finally, several AI × crypto business models use smart contracts to reduce expenses associated with centralized control and storage of data. By eliminating the need for data infrastructure, businesses can be profitable sooner and, thereby, pass some value directly back to the data owners while still delivering data at a lower price than centralized models charge AI developers.

Data quality is also an issue that AI × crypto can help with. In response to looming data scarcity, novel business models aim to take advantage of the industry shift from “more data is better” to prioritizing high-quality data. This shift may involve synthetic, multimodal, and detailed domain-specific data or simply higher-quality, human-labeled data. Generating synthetic datasets can supplement real-world data, though there is concern about model collapse if models predominantly learn from their own generated outputs. Economically, synthetic data can reduce the dependence on rare or expensive real-world data but must be balanced against potential diminishing returns in model quality. Multi-modal data refers to information spanning multiple senses or modalities. Commonly, this means combining text with images, audio, or video. Multi-modal learning enables generative AI systems to gain a richer, more nuanced understanding by aligning information across multiple vectors.

Finally, one of the more promising areas combining AI × crypto in the data domain is using cryptographic tools to enable access to specialized data in medicine, law, engineering, and other fields. Typically, this data is too sensitive for public release but extremely valuable for the context it provides. High-quality, domain-specific data often resides behind privacy or regulatory barriers, particularly in healthcare. Cryptographic primitives such as Zero-Knowledge proofs (ZKps), Fully Homomorphic Encryption (FHE), and secure multiparty computation enable AI models to learn from private data without exposing the raw inputs. For example, some protocols help individuals tokenize their biometric data and then lease this information through smart contracts to companies. From an economic perspective, these technologies reduce risk and liability for data holders, lowering sharing costs and increasing overall supply.

These decentralized data solutions are made possible by blockchain-enabled machine-to-machine micropayment systems that when combined with smart contracts streamline compensation for data owners, making it viable to supply and purchase high-quality data contributions of any size. Because blockchain transparency enables participants to price data quality more accurately, resources flow more efficiently to the high-quality data that yields the greatest marginal improvement in frontier AI models. Finally, by reducing privacy risks, more data contributors, ranging from large institutions to individual end-users, are willing to participate in tokenized data markets.

As an example, consider Grass, a powerful data-generation engine. Imagine you want to build a vast, real-time database of virtually everything online, from product prices to job listings to social media updates. You could run your job through a single service provider and use big server farms, or you could tap into millions of ordinary people’s devices worldwide. The latter is essentially what Grass does. When a firm or researcher needs to collect online data, they submit their requests to Grass’s network. Grass then routes these tasks to participating user devices (Grass nodes), which retrieve the data from target websites as though the users themselves were browsing. The participating user devices (idle computers at home) receive micropayments in cryptocurrency for their service. The multimodal data retrieved from the nodes is then cleaned and structured by Grass. Effectively, Grass serves as a centralized coordinator, relying on validators and routers to ensure data quality and provenance. By harnessing millions of contributors worldwide, Grass can efficiently gather large volumes of publicly available web information on a daily basis. This decentralized model also reduces blind spots in the data gathering process as it gets around the barriers that traditional web crawlers encounter (time-outs, request limits), and therefore, Grass can provide higher-quality data.

Turning from data and compute, the third subcategory of AI × crypto is AI models. AI models are trained systems that can perform predictions based on input data, whether predicting the next word in a sentence, as in the case of LLMs, or predicting the best trade for an investor’s portfolio. Notably, the parameters that define frontier AI models are typically chosen by humans, introducing a way for bias to leak into the system. For instance, data labeling, feature selection, imbalance in class representation, postprocessing on model outputs, and hyperparameter tuning all influence a model’s performance. Importantly, even narrow finetuning such as on a small amount of insecure code or even inaccurate numbers can produce broadly misaligned LLMs on almost everything else they output such as offering misinformation (Betley et al., Reference Betley, Tan and Warncke2025).

Like centralized entities seeking to achieve AGI, DGI is a primary goal of the business models in this space. DGI is the notion that an AI system, rather than running as one monolithic model in a single data center, could operate across many different machines or nodes (potentially owned by different people or organizations). The hope is that by distributing data, computation, and decision-making power, the model has greater resilience, less influence from any single controlling entity, and the potential to scale in ways traditional centralized systems cannot.

In practice, current large-centralized AGI efforts from OpenAI to Google’s DeepMind and others have the lead. Nevertheless, the promise of DGI remains an attractive investment to some as it is one of the only ways to combat centralized control of powerful frontier AI models. For example, Sentient Technologies, which is often associated with Peter Thiel, is experimenting with a decentralized approach to AI models. It focuses on e-commerce AI models and spreading computational tasks across a large, distributed infrastructure. This approach of focusing on a specific problem is more common at the intersection of AI × crypto.

For example, Numerai is a hedge fund that crowdsources AI models from data scientists worldwide. It does so by supplying high-quality, anonymized financial data for free. In return, data scientists compete to develop predictive AI models, staking Numerai’s token on their models. If their models perform well, they earn rewards in proportion to how much Numerai uses their model. As a second example, Bittensor is a decentralized protocol that rewards developers for creating and sharing AI models. Like Numerai, Bittensor uses token-based incentives to encourage high-quality model contributions. However, the two differ in both purpose and structure. Bittensor aims to build a global marketplace for AI services, in which model creators earn tokens as payment for others using (or benefiting from) their models. Thus, Bittensor is one of the few projects attempting to combine blockchain incentives with an open, general AI framework. Developers retain more control, and an open network can potentially aggregate thousands of specialized models to create a competitive model with AGI. Thus, Bittensor’s approach counters centralized AGI models by incentivizing a global community to share and refine models openly.

The final category of AI × crypto is applications, which include AI agents and authentication services. This area is growing fast because, whether or not DGI is achieved, blockchain developers can integrate frontier centralized AI models into applications. An AI model, for instance, could be implemented into a DeFi-style application to rebalance a portfolio optimally. As some proponents argue, adding these AI models to blockchains makes the most sense since blockchains provide a deterministic, transparent, and tamper-resistant platform where smart contracts already function much like simple autonomous agents. Extending this idea, some believe that more advanced AI agents, which are capable of making decisions, owning digital assets, and interacting with external data via oracles, could naturally live on-chain. Because blockchains enforce contract rules programmatically and are difficult to shut down or censor, agentic AI running on them could, in theory, operate with minimal human intervention.

In practice, AI agents can be constructed to self-reflect, use tools, plan, and collaborate with other agents, allowing them to mimic traditional workflow loops more closely. From an economic perspective, agentic AI offers specialization, which is helpful in the workplace. Specifically, AI agents can be programmed to concentrate on a specific function, leading to greater efficiency, high-quality outputs, and continuous improvement through repeated practice. By excelling at narrow tasks, agentic AI systems can be aggregated to tackle complex problems in a highly coordinated and efficient manner. In this sense, agentic AI brings the composability that blockchain is famous for offering customers in a setting previously dominated by broad, one-size-fits-all automation. They hold meaningful promise because specialized AI agents can be created and modified more rapidly, adapt to new tasks more flexibly, and continuously refine their skills.

When these agentic AI systems operate on a blockchain, they benefit from a transparent and tamper-resistant infrastructure that enables seamless collaboration. The blockchain provides a reliable ledger of contributions, ensuring each specialized agent’s work is recorded and verified without centralized oversight. It also automates incentives: agents can be rewarded in proportion to their contributions, encouraging participants to develop ever more specialized and effective models. This self-reinforcing loop, where each agent gains feedback, improves its function, and is compensated for its utility, leads to a more dynamic and innovative environment. It is also the case that all of these agents are “unbanked” and the traditional financial system is unlikely to want to onboard these agents, leaving crypto as their primary avenue to both make and receive payments.Footnote ³

Thus, it’s worthwhile to think through a concrete example. Imagine a decentralized entity with a chatbot interface run by various AI agents. The chatbot could build a following by posting appealing content to social media and generating income in multiple ways from its audience. It could manage and invest those earnings through cryptocurrencies. It is quite feasible that such an entity could become a fully autonomous billion-dollar entity in time. In later sections, we will consider the gaps regulators face in serving these DAEs, but for now, it is worth recognizing that there are already examples of this, ranging from Truth Terminal, which gained popularity on X, to robots like Gaka-chu, which makes a living as an entrepreneurial painter and manages its’ earnings through cryptocurrencies (Ferrer et al., Reference Ferrer, Berman, Kapitonov, Manaenko, Chernyaev and Tarasov2023). In fact, many blockchain analysts are predicting a dramatic increase in the role of AI agents in the ecosystem. VanEck predicts one million AI agents on the blockchain by the end of 2025, and others argue that Web3 agents will be blockchain’s leading creators and consumers.

Finally, AI has made creating and spreading misinformation like deceptive deep fakes cheaper. With lower barriers for malicious actors who want to circulate false content, this, in turn, has increased demand for verifying the accuracy and origins of what users encounter online. In this environment, crypto-based protocols offer a unique way to authenticate data. They do this by providing an on-chain proof of the source, effectively allowing anyone to confirm that a piece of information actually came from a specific user or AI system.

From an economic standpoint, authentication services rely on supply and demand in a new market for verifiable data. On the demand side, users, businesses, and organizations want a trusted method to ensure the information they receive is legitimate. This demand will likely grow as more deepfakes and misleading content appear, making reliable verification more important. On the supply side, crypto projects are developing decentralized protocols that allow anyone to register or prove the authenticity of data. By tokenizing these services, they create financial incentives through token rewards for participants to maintain the system’s integrity.

Several examples illustrate both the promise and the challenges of these authentication services. Worldcoin, from OpenAI cofounder Sam Altman, focuses on proving that a user is a real person through an approach called “Proof-of-Personhood.” As a second example, consider Numbers Protocol, which aims to label AI-generated content and register digital media files on-chain. The main economic driver is whether enough people need reliable data verification and whether these platforms can meet that need in a scalable way. Suppose users find the solution superior to traditional verification methods, like centralized IDs or manual verification. In that case, they may be willing to pay for, stake, or support the tokens that power these authentications.

3.8 Common Economic Forces

The breadth of recent financial innovations is unusual in its expansiveness, encompassing advancements in conventional FinTech, AI, blockchain, cryptocurrencies, smart contracts, and even integrating across dimensions to create novel financial capabilities within the AI × crypto space. Before describing the regulatory responses to this wave of innovation, I examine the economic forces that underlie these innovations. In doing so, it becomes clear that driving forces often push in different directions, creating new risks and tensions. For example, have the improvements in transparency brought about by blockchain meaningfully mitigated information asymmetry, or have they introduced new distortions and complexities? Similarly, have the novel governance structures associated with decentralized organizations reduced agency costs, or do they persist and manifest in other ways?

By analyzing the extent to which recent financial innovations have brought about changes in fundamental economic phenomena, including agency costs, information asymmetry, moral hazard, hold-up problems, and coordination challenges, this subsection helps identify the new and old channels through which economic distortions may arise. These five common economic forces provide a foundational reference for subsequent sections examining regulatory and policy responses. Thus, it is necessary to have an eye toward potential solutions for mitigating such distortions and enforcing accountability for bad actors in the FinTech space.

3.9 What Areas of Financial Services Are Technology Resistant?

It is worth noting that while emerging financial technologies can replicate and expand the current offerings of financial services firms like banks, insurance companies, and asset management firms, there are still areas of financial services in which emerging technologies are mostly absent. Consider investment banking, which includes underwriting for debt and equity issuances, and advising companies involved in mergers and acquisitions. These deals are still driven by human capital, client experience, reputation, and excellence in execution. The industry is comprised of many small, boutique investment banks that specialize in providing bespoke, artisan-like services to their clients. These complex tasks still appear to be technology resistant, although some aspects of the overall service are being offered more efficiently (e.g, investment banking analysts can save time using generative AI to help write first drafts of pitchbooks). Finally, while some AI and alternative data providers are helping in the due diligence process for M&A deals, there is not a decentralized application for sourcing M&A deals or getting clients through negotiations and regulatory review. Similarly, while token sales or DeFi loans may represent an alternative way for startups to raise capital to fund their business, the underwriting quality and size of the underwriters’ networks still matter for these offerings. Thus, it is more common for applications to focus on the tools such as providing an AI chatbot that offer investors and traders answers to bond-related covenant questions rather than underwriting and placing the bond specifically. Nevertheless, some in the crypto space are making use of rights granted to investment circles and similar smaller entities to explore the potential of fully autonomous underwriting processes.

4.1 AI

In the case of AI, a prominent failure meriting regulatory intervention is that algorithms can be biased (Bartlett et al., Reference Bartlett, Morse, Stanton and Wallce2022). Such bias causes harm, especially when the systems are deployed too hastily (e.g., audits of face recognition systems repeatedly show racial and gender bias; audits of credit screens show that they are biased). Other market failures relate to manipulation that targets vulnerable populations (e.g., realistic deep-fakes that vulnerable populations do not realize are AI-embedded). There are also, of course, externalities, such as the climate impact of energy-guzzling deep learning models, and the potential displacement of high-skilled workers.

4.2 Blockchain Technologies and Digital Assets

The release of the Bitcoin whitepaper in 2008 by Satoshi Nakamoto is characterized as radical innovation without precise applications. Its decentralized, peer-to-peer network provided a stage for inventors to revive peer-to-peer ideas from the early Internet days. It made new use cases possible through the elimination of breaches of trust. From a regulatory perspective, the U.S. approach toward regulating blockchain and DeFi applications appears to be targeted and specific. For example, regulators clarify regulatory uncertainty through speeches, which allow well-intentioned developers to thrive without imposing potentially high compliance costs. At the same time, this targeted approach will enable regulators to focus on bad actors who may be using the system in ill-intentioned ways (e.g., for money laundering or tax evasion). In what follows, I highlight key decisions that have shaped innovation incentives. Importantly, regulatory goals are numerous, including investor protection, market efficiency and integrity, capital formation, financial inclusion, the prevention of illicit activity, safety and soundness, and financial stability.

4.3 AI × Crypto

Legal scholars and courts in the U.S. have previously grappled with the nature of emerging entities, and the courts will soon likely confront the challenge of defining whether an agentic AI is entitled to legal recognition. The courts have not formally taken on a case along these dimensions, but past scholarship suggests a few potential paths forward. One approach would involve arguments that treating corporations as legal persons offers a blueprint for extending certain rights and obligations to AI systems. Another potential line of reasoning would draw on the legal precedent that assigns animals to a distinct status, which lies outside the conventional binaries of either personhood or property. Such an analogy would offer a potential parallel framework for thinking about how we might classify advanced AI systems. Finally, a third viewpoint suggests that AI functions much like an advanced piece of machinery, implying that any harmful outcomes it produces fall under the responsibility of the developers, owners, or users who manufacture and operate it. This reasoning aligns with well-established tort law, where legal accountability attaches to those designing or manufacturing the product rather than the underlying technology itself. In this subsection, I examine the arguments for and against these three viewpoints and provide additional context for their basis.

5.1 Traditional Regulatory Tools

Traditionally, financial markets are governed by two primary regulatory mechanisms: price-based interventions, like interest rate caps, that correct market failures by altering economic incentives, and entry/exit controls, like licensing regimes or merger reviews, that determine which entities can participate in a regulated activity. Beyond these two approaches, regulators also employ tools like disclosure requirements, targeted activity restrictions, macroprudential buffers, and time-based mechanisms like trading blackout periods – each addressing specific economic distortions that competition alone cannot resolve.

Similar regulatory principles are being adapted and applied in the FinTech space. For example, stablecoin reserve requirements use the same tool (i.e., pricing) to restore distorted incentives. Regulatory challenges with stablecoins, in some ways, parallel those encountered in fractional reserve banking. By promising 1:1 redemption while often investing in less liquid instruments, stablecoin issuers create maturity mismatches that expose them to classic bank-run dynamics whenever users lose confidence in the adequacy of reserves. Compounding this vulnerability is the relative absence of traditional safety netsFootnote ⁷ Price-based regulation through mandatory reserve requirements better aligns incentives. Namely, by increasing the cost of stablecoin issuance through the obligation to hold high-quality, liquid assets, issuers internalize the systemic risks they generate.

In addition, discussions about crypto exchange licensing regimes and listing requirements mimic traditional exit/entry controls utilized by regulators. For example, if regulators believe that cryptocurrency exchanges suffer from information asymmetries and principal-agent problems with users unable to verify exchange solvency or security controls. Then, one would expect to see catastrophic collapses (e.g., Mt. Gox, FTX). The classic regulatory solution is entry-based regulation, given that the economic costs extend beyond direct losses to include diminished market participation due to trust deficits and misallocation of capital toward excessive self-custody solutions. Specifically, licensing requirements would establish minimum standards for exchange operations. These would likely include proof of reserves and segregation of client assets, robust cybersecurity standards, regular audits, verifiable corporate governance mandates, market surveillance capabilities, and financial reserves. Society derives economic benefits from addressing the quality issue (e.g., by screening out low-quality exchanges), as licensing can create a separating equilibrium where consumers can more confidently participate in the market. This regulatory approach mirrors traditional securities exchange regulation.

Of course, not all exchanges for digital assets are centralized. This suggests that determining an appropriate way to license DEX to enable proper competition is vital. DEXs, however, face a different set of problems, such as front-running. Thus, while DEX resolves numerous frictions in financial services by reducing costs and bringing accessibility to more investors at a transparent/constant exchange rate, they also pose new risks to traders and liquidity providers, such as sandwich attacks and slippage if liquidity dries up in a particular pool. Thus, in an ideal world, regulation could address these novel exploitations of users and provide a policing or enforcement mechanism that reduces the bad actor’s ability to take such actions (Harvey et al., Reference Harvey, Hasbrouck and Saleh2024).

Entry/exit decisions also come into play when considering which tokens are listed on an exchange. However, the economic incentives in this context make it much more apparent that the burden should be on the regulator rather than the exchange itself. For example, suppose one wants to continue to enable innovation and experimentation. In that case, the onerous should not be unduly on the entrepreneur to pay high compliance fees to be able to list a token for her project. Instead, it seems that the onus should be placed on the regulator. As long as a token reasonably meets the criteria for listing (e.g., something more similar to a commodity than a security), it would then be the responsibility of the regulator to clarify which digital assets do not meet listing standards. Importantly, the regulator would also give centralized exchanges and DEXs a grace period to delist noncompliant tokens.

Placing the onus on the regulator incentivizes exchanges and token issuers to work within regulatory frameworks rather than seeking less regulated jurisdictions. It also prevents regulators from going after well-intentioned projects due to the law’s ambiguity. Instead, it encourages regulators to use their limited resources to focus on projects where bad actors are trying to exploit consumers. Of course, many types of tokens exist, so it seems prudent to ask if any tokens should automatically be exempt based on their risk profile. On this front, it does seem evident that NFTs, at least creative ones, should be exempted, as regulators should not be tasked with judging artistic expression. However, for financialized NFTs that represent RWAs, traditional risk-based financial regulation seems more useful than a commodity-like listing. Finally, in this vein, memecoins, which have been exempted from security status, seem like they could be pushed even further into a public lottery-like framework and not be in the same category as other digital assets or NFTs. The main idea is that with any digital asset, you want to prevent fraud, not speculation. Even in the memecoin form, speculation is reasonable, but that speculative feature can be channeled into a better product like a public-lottery-ticket memecoin that serves the greater good (Grennan et al., Reference Grennan, Liu and Tan2025).

Exchanges are traditionally viewed as a regulatory bottleneck point where information and actors come together and, thus, are ideal for regulatory intervention. As the blockchain ecosystem continues to embrace a multi-chain architecture, a new “bottleneck” point is becoming apparent: bridges. Substantial systemic risk emerges from such cross-chain dependencies. When a bridge protocol fails, as was the case with Wormhole, Ronin, and numerous others, the contagion quickly spreads across previously isolated blockchains, creating correlated failures. Individual projects lack incentives to fully account for these systemic risks, creating a classic externality problem where interconnection decisions do not reflect their true social costs.

One regulatory mechanism available is cross-chain risk buffers. Such a mechanism would require bridge protocols and cross-chain integration applications to maintain additional reserves proportional to their systemic importance. Unlike simple capital requirements, these buffers would precisely scale with measures of interconnectedness: the volume of cross-chain transfers, the number of connected chains, and the concentration of economic value flowing through particular bridge points. The economic logic parallels systemic risk surcharges for globally important financial institutions, but is adapted to the unique topology of blockchain networks. By imposing costs proportional to the systemic risk created, this mechanism would encourage more robust bridge architectures, appropriate diversification across connection points, and improved security for critical cross-chain infrastructure.

Finally, another prime candidate for regulatory mechanisms linked to pricing would be to introduce transaction fee caps for blockchain networks. Evidence increasingly shows that blockchain miners act in ways that are not fully competitive (Lehar and Parlour, Reference Lehar and Parlour2024). During high-demand periods, this creates significant deadweight loss through congestion pricing. When network activity spikes, transaction fees can increase by orders of magnitude, as seen during NFT minting frenzies or large market swings. The underlying economic problem stems from the fixed short-term transaction capacity and highly inelastic demand during peak periods. This creates textbook conditions for market power exploitation, where users with time-sensitive transactions must pay extraordinary premiums to access the network. One potential regulatory solution would be to implement fee caps, which are a form of price regulation. By setting maximum transaction fees, regulators could prevent extreme price gouging during congestion periods while incentivizing novel solutions like roll-ups or other forms of capacity expansion.

5.2 Adapted Definitions and Legal Frameworks

Regulators can maintain clarity and effectiveness while accommodating FinTech innovations by adapting legal concepts to encompass new market realities rather than creating entirely new regulatory structures. This subsection explores several adaptation strategies, from expanding the definition of exchange to include DeFi protocols, to reimagining custody frameworks for multi-sig wallets, to implementing safe harbors with sunset provisions that allow for time-limited approvals with built-in review mechanisms.

First, as an example of adaptation, consider the SEC’s proposed change to the definition of an “exchange.” After a series of back-and-forth for comments, in January 2022, the SEC acknowledged that advances in technology and innovation have changed the methods by which securities markets bring together buyers and sellers of securities. As the SEC explained, “[i]nstead of using exchange markets that offer only the use of firm orders and provide matching algorithms, market participants can connect to numerous Communication Protocol Systems, which offer the use of protocols and nonfirm trading interest to bring together buyers and sellers of securities”. (SEC, 2022). By including this broad definition of a Communication Protocol Systems within the formal definition of an “exchange,” the SEC is extending its regulatory framework to DeFi protocols. Under the new definition, protocols that bring together buyers and sellers of securities would have to register with the SEC as an exchange unless otherwise exempt.

Another definition that is ripe for adaptation or refinement is custody. Multi-sig wallets are a common decentralized security and governance feature that requires multiple private keys to authorize transactions, creating a distributed control model. Yet this challenges traditional regulatory frameworks. Does signing, accessing funds, and/or moving funds constitute custody? Is there a separate level of custody that could be exercised by each signatory over the wallet? Unlike conventional financial custody, where a single institution maintains complete control over client assets, multi-sig arrangements distribute this power across various stakeholders, potentially including the asset owner, service providers, and independent entities, creating graduated levels of control rather than binary custody relationships. Current regulatory approaches typically employ simplistic custodial/noncustodial distinctions that fail to capture the nuanced reality where a service provider holding one of three required keys exercises fundamentally different control than one holding two of three keys (effectively having custody). A more sophisticated regulatory framework would implement a control-based spectrum approach that considers factors such as threshold requirements (e.g., 3-of-5 signatures needed), key distribution, emergency recovery mechanisms, and operational practices to determine regulatory obligations. This would allow for proportionate oversight while preserving the security benefits and governance innovations that multi-sig technology enables. Finally, regulators could help by clarifying the legal standard expected of third-party signatories, who act as custodians, making it clear to those doing so that they must adhere to qualified custodian standards, such as transparency, provision of insurance, and conflict resolution mechanisms.

Similar to discussing adapted definitions, it would be useful for regulators to think of adapted legal frameworks. This seems to be especially relevant for regulatory issues surrounding agentic AI systems and AI × crypto applications. As discussed in Subsection 4.3, while AI technologies appear in court cases, U.S. judges have only addressed them using conventional doctrines (e.g., judges treat AI as a tool and focus on the people or corporate entities behind it). No existing decisions suggest that AI, on its own, can be a defendant or hold rights. Instead, the prevailing assumption is that liability or entitlement ultimately rests with identifiable human or corporate stakeholders. Yet there are arguments to be made for AI to be considered in its own right, such as a hybrid manner where, under certain circumstances, it would be considered to have rights similar to animals.

Another way to adapt legal frameworks or to add flexibility is to introduce safe harbors and to include sunset provisions. These adaptive tools also allow for collaborative governance models, and so are discussed in more detail in Section 5.5.

5.3 Self-regulatory Solutions

Pure self-regulation is unlikely to address the economic distortions caused by FinTech innovations adequately. The core market failures, like information asymmetry, create incentive structures where voluntary collective action is unlikely to emerge organically or persist without external enforcement. Fundamentally, most FinTech startups possess substantially more knowledge than consumers. In traditional finance, this leads to disclosure requirements and consumer protection regulations.

For self-regulation to address this distortion effectively, industry participants would need to establish a standardized disclosure framework, create enforcement mechanisms with meaningful penalties, and overcome the competitive advantage of opacity. The economic challenge is that startups individually benefit from maintaining information advantages while collectively suffering from reduced market trust. This creates a classic prisoner’s dilemma where rational actors will not voluntarily provide optimal transparency without external enforcement.

Nevertheless, a few examples of self-regulation have been proposed by those involved with building digital assets and DeFi applications. For self-regulation to be a viable option, however, crypto industry leaders need to coordinate and potentially lobby key decision-makers. In terms of lobbying, the crypto industry was nascent at best, but did focus meaningfully in the 2024 election cycle. For example, when the now-dormant Infrastructure Bill was being debated in the fall of 2021, only a few of the big players in the crypto space intervened with meaningful money. Coinbase, whose business model would have been severely harmed if the bill passed, spent only $625,000 on lobbying in 2021Q3. In comparison, Meta spent $19.7 billion on lobbying in 2020. Another advocate to emerge during the Infrastructure Bill debates was the Blockchain Association, which is a member-driven policy group for crypto networks. The Blockchain Association has several proposals and the involvement of key players in the industry. Importantly, though, in the 2024 elections, the crypto industry joined together in the U.S. to bring their special interests forward. For example, the Political Action Committee, Fairshake, raised over $250 million in 2024. Ryan Selkis, a crypto executive, explained the logic: there are 50 million crypto holders in the U.S., and that’s a lot of voters (Yaffe-Bellany et al., Reference Yaffe-Bellany, Griffith and Schleifer2024).

To date, most self-regulatory proposals have flopped. In 2018, the Gemini crypto-exchange founders proposed a Virtual Commodity Association (VCA) as a self-regulatory organization for cryptocurrency markets. The proposal envisioned an organization like the National Futures Association (NFA) for futures markets, which coordinates with the CFTC rather than a fully congressionally approved self-regulatory organization like the FINRA, which acts to protect investors, safeguards market integrity, and resolves disputes. Like many self-regulatory organizations, a key aspect of the proposal was industry standards that promote transparency, efficiency, and price discovery. In general, industry self-regulation may have advantages over regulation by rule and enforcement or market-based regulatory mechanisms. For example, a common argument is that the industry is complex and those who know it best and, therefore, have an informational advantage are those participating in and building the underlying technology. Another potential advantage of self-regulation is that it could be global. Yet despite the potential advantages, historically speaking, as industries mature, self-regulatory approaches are associated with corruption and inefficiencies (Ogus, Reference Ogus1995).

As an alternative to a pure self-regulatory approach, a hybrid approach seems more plausible, such as a self-regulatory agency backed by a consortium of government agencies on a rotating basis. Such coordination and explicit rotation would mirror the approach with AI. Another hybrid option would be a version of audited self-regulation. Typically, in these approaches, the industry develops technical standards and best practices, but then the government provides oversight, and third parties verify compliance. Thus, there is a regulatory backstop for systematic failures. This approach has the advantage of leveraging industry knowledge while addressing incentive problems that undermine pure self-regulation. It’s particularly suited to areas where technical complexity makes direct government regulation challenging, but economic distortions make pure self-regulation insufficient.

Consistent with those conditions, there has been a push toward self-regulation with DAOs and verified third-party compliance. For example, a group of DAO technologists and founders created Metagov. This self-regulatory organization helps provide standards and easy tech plug-ins for its members. It also externally verifies membership, but currently, it does not verify compliance with its proposed standards. In this sense, a hybrid regulatory framework that utilizes this combination of industry expertise (e.g., best practices) and incentive-aligned solutions (e.g., easy onboarding) in a manner similar to Metagov, but with actual government or auditing oversight (e.g., verified compliance) would offer the most promising self-regulatory path forward.

5.4 RegTech Solutions

FinTech regulators face a paradox: the same innovations that users adopt because they enhance the efficiency of and access to financial services also simultaneously create new opportunities for criminal exploitation. Nearly all FinTech innovations outlined in this paper, besides conventional FinTech, inadvertently provide tools for cybercrimes ranging from money laundering to ransomware payments. As documented by Cong et al. (Reference Cong, Harvey, Rabetti and Wu2023c), DeFi platforms have become particularly vulnerable to exploitation, with their pseudonymous nature and cross-border functionality presenting unprecedented regulatory challenges. Yet, this technological revolution offers a solution. Namely, blockchain’s inherent transparency creates immutable audit trails that, when adequately leveraged through forensic techniques outlined by Cong et al. (Reference Cong, Grauer, Rabetti and Updegrave2023b), provide regulators with robust detection and attribution capabilities previously unavailable in traditional financial systems.

Most RegTech solutions leverage AI and big data analytics to automate compliance tasks, monitor risks, and detect potential violations in real time. One area where RegTech is being effectively used is in combating financial crimes, such as terrorist financing. AI-powered systems can analyze vast amounts of transactional data to identify suspicious activity and flag potential risks, enabling financial institutions to take proactive measures and report suspicious activities to regulators. RegTech is also being applied to improve the efficiency and accuracy of regulatory reporting with solutions that automate data collection, validation, and submission processes. Moreover, RegTech can help financial institutions navigate the complexities of compliance in an increasingly digital and globalized economic landscape, such as managing risks associated with cryptocurrencies and cross-border transactions.

A few key challenges are preventing RegTech adoption: concerns about data quality, security, and efficacy, infrastructure investments, and labor risk. For example, executives at financial institutions acknowledge difficulties in integrating new technologies within their existing systems and often hesitate to invest in upgrading. Even if the solutions are viable and offer an NPV-positive investment, behavioral and cultural explanations consistently show a lack of adoption (Graham et al., Reference Graham, Grennan, Harvey and Rajgopal2022; Mishra et al., Reference Mishra, Prabhala and Rajan2022; Gertler et al., Reference Gertler, Higgins, Malmendier and Ojeda2025). Another potential risk that may arise is labor risk, because there is a dearth of technical skills needed to supervise the use of RegTech solutions. Some may argue that generative AI can assist with coding and automation, yet this is a constant “arms race,” because any criminal organization can just as easily adopt generative AI productivity tools. Concerns about data quality, security, and efficacy often come down to the recognition that algorithms often underperform relative to humans in detecting tail risk events. Thus, while RegTech solutions are promising, regulators should take a balanced approach, one that acknowledges and showcases RegTech’s benefits but, at the same time, monitors the new tools used to enable complementarities with more holistic human assessment.

One potentially promising RegTech solution in the DeFi space is privacy-preserving regulatory oracles. It is well established that blockchain’s pseudo-anonymous nature enables illicit activity, tax evasion, and sanctions violations, creating negative externalities. However, full transparency eliminates the privacy benefits driving legitimate cryptocurrency demand. This represents a complex market failure where neither extreme produces a socially optimal outcome. Traditional approaches to this problem involve direct KYC/AML requirements that fundamentally compromise privacy. This creates substantial dead weight loss by eliminating beneficial privacy-preserving use cases and driving activity toward less regulated alternatives.

The RegTech solution is privacy-preserving regulatory oracles, which would enable selective disclosure, through ZK-proofs, allowing compliance verification without full transparency. These systems would utilize cryptographic techniques to confirm regulatory requirements (identity verification, source of funds, tax compliance) without revealing the underlying data. For example, a transaction could include a ZK-proof demonstrating the sender has completed required verification without exposing their identity. Similarly, proofs could verify that a transaction does not violate sanctions without revealing the specific parties involved.

The economic efficiency of this RegTech approach to compliance stems from ZK-proofs’ preservation of privacy as a feature while addressing the externality problems of completely anonymous transactions. Rather than forcing a binary choice between full transparency and complete anonymity, this mechanism enables a more nuanced middle ground that preserves beneficial privacy characteristics while mitigating social harms. This regulatory approach has no direct parallel in traditional finance, representing a truly crypto-native RegTech solution.

Another promising and novel RegTech solution is to use technology to provide insurance. Although smart contract code is publicly viewable, most users lack the technical expertise to audit it, creating various economic distortions related to information asymmetry. When vulnerabilities lead to exploits or hacks, users bear meaningful losses with little to no recourse. This represents a classic market failure: despite the theoretical transparency of open-source code, practical opacity creates conditions where users cannot distinguish between secure and vulnerable protocols. The resulting adverse selection problem leads to insufficient investment in security audits and formal verification since users cannot accurately price this quality differential. A RegTech solution, therefore, would be either incentivized or mandatory participation in smart contract insurance pools. This solution addresses the information asymmetry challenge without technology. Under this system, protocols deploying smart contracts would contribute a percentage of transaction fees or token value to a collective insurance fund that compensates users in the event of code-based failures.

The appeal of the RegTech insurance approach lies in its risk-sharing characteristics. By pooling risk across the ecosystem, the RegTech insurance protocol creates collective incentives for improved security while providing a safety net for inevitable failures. The system could implement risk-based premiums where contracts with formal verification or multiple independent audits qualify for reduced contribution rates, creating direct economic incentives for security investment. This regulatory mechanism resembles deposit insurance in traditional banking but operates through a decentralized risk-pooling arrangement rather than a centralized guarantor. It addresses the externality problem where individual projects underinvest in security because they do not fully internalize the reputational damage their failures inflict on the broader ecosystem.

5.5 Collaborative Governance and Complementary Regulation

It will take time for regulators to learn the points of tension in the financial ecosystem brought about by the new use cases of emerging financial technologies, which leaves room for incumbents and new entrants to influence policy. To that extent, developers, startups, and incumbents operating in the FinTech space may need to overcome their anarchist/anti-authoritarian roots and join a trade association or lobby politicians to achieve a regulatory solution that optimally balances all stakeholders’ interests.

A helpful approach could be engaging stakeholders and other nongovernmental actors in the regulatory process to harness their expertise, legitimacy, and market influence to promote more efficient and effective FinTech regulation. Given the sector’s complexity, regulators may need to work with researchers and academics who are experts and leverage their impartial views. Rigorous analysis can significantly contribute to our collective understanding of the implications of new technologies and to formulating policies that balance innovation with consumer protection and market integrity.

Similarly, firms could involve industry associations, academics, and consumer groups in developing regulatory standards and guidelines. Doing so can help to ensure that the suggested standards are technically feasible, commercially viable, and socially acceptable. Alternatively, it may be best to delegate certain regulatory functions to self-regulatory organizations or industry associations that can help leverage their market knowledge, especially their ability to respond quickly and flexibly to changing market conditions. This regulatory approach of involving external stakeholders early in a project’s lifecycle can also help build trust and legitimacy in the regulatory system and reduce the risk of regulatory capture or inertia later on.

It is worth noting, however, that involving stakeholders and other nongovernmental actors in the regulatory process also has some potential drawbacks and challenges. First, ensuring that all relevant stakeholders are adequately represented and have a meaningful voice in the regulatory process can be challenging, particularly for smaller or marginalized groups. There is a risk that well-resourced, special-interest groups may dominate the process and shape outcomes in their favor. For instance, Mattli and Woods (Reference Mattli and Woods2009) argue that the politics of regulation is characterized by a bias in favor of the most influential and wealthy actors, leading to regulatory capture in that well-organized industry groups can shape regulatory outcomes at the expense of broader public interests. One key to addressing this challenge is opening the regulatory process to new actors and ideas. This often requires regulators to be proactive in their efforts to level the playing field and enable meaningful participation by a broader range of actors.

Second, conflicts of interest may arise, such that the stakeholders’ competing interests can undermine the effectiveness or legitimacy of the regulatory process. For instance, industry groups may prioritize short-term profits over long-term stability, while consumer groups may prioritize access and affordability over innovation and competition. Finally, delegating regulatory authority to nongovernmental actors can create challenges for accountability and oversight. There is a risk that self-regulatory organizations or other intermediaries may lack the capacity, incentives, or legal authority to effectively monitor and enforce compliance with regulatory standards (Coglianese and Mendelson, Reference Coglianese and Mendelson2010). To mitigate these challenges, regulators should establish clear criteria and processes for stakeholder participation, ensure a balance of interests and perspectives, and maintain ultimate regulatory authority and oversight. Finally, capacity-building support or resources may need to be available to garner meaningful participation by all stakeholders.

An important prerequisite for meaningful stakeholder participation in FinTech governance is addressing the substantial knowledge asymmetries between industry participants and consumers. Traditional financial literacy education, while valuable, proves insufficient for navigating the complexities of DeFi protocols, smart contract risks, yield farming mechanisms, and cross-chain bridge vulnerabilities. This educational deficit creates market failures where consumers cannot adequately assess risks or make informed participation decisions, ultimately undermining the legitimacy of collaborative governance models that rely on informed stakeholder input. One promising regulatory mechanism involves leveraging existing FTC enforcement precedents to address this information gap systematically. When the FTC pursues enforcement actions against firms for deceptive advertising practices in the DeFi space, settlement agreements could include mandatory public education components funded by the violating firms. This approach, which has precedent in traditional consumer protection settlements, would create a self-funding mechanism for FinTech literacy programs while directly targeting the knowledge gaps that enabled the deceptive practices. Such educational remedies could take various forms, from funding university research centers focused on FinTech consumer protection to supporting the development of standardized risk disclosure frameworks that help users understand protocol-specific vulnerabilities. By making firms that profit from consumer confusion bear the costs of remedial education, this mechanism aligns private incentives with education goals necessary for effective collaborative governance.

One intuitive path toward collaborative governance includes permitting safe harbors, those protected spaces where developers can transition products from centralized to decentralized control without the crushing weight of compliance requirements during fragile growth stages. However, these safe harbors are only genuinely safe for other stakeholders when paired with sunset provisions. Time-limited approvals with renewal requirements allow experimentation with innovation while maintaining the power for a stakeholder-informed regulator to end experiments that harm rather than help, based on real-world outcomes rather than theoretical projections.

As an early example of the need for safe harbors for developers, consider the history of Bitcoin. In March 2013, when Bitcoin was still under $1000, Bitcoin suffered an accidental hard fork that lasted for several hours as two different versions of the network led miners to unknowingly start building two separate chains (Andresen, Reference Andresen2013). The centralized leadership team quickly detected the split, and the miners on the more advanced node took a financial loss to downgrade and restore the canonical chain. An attack was made on a merchant during the accidental hard fork. After the incident, Bitcoin leadership released an upgrade to fix the problem and sent a series of alerts to users of older versions asking them to upgrade.

As the Bitcoin example illustrates, some early-stage FinTech projects could appear unstable in their infancy, but with time, they may grow into stable, valuable products or services. Safe harbors allow developers to gradually transition their product or service from centralized to decentralized control without being subject to costly compliance requirements. Safe harbors enable key figures in a startup’s development, like founders, developers, and VCs, to monitor progress and respond quickly. However, an important consideration regulators need to make with any safe harbor or sunset provision is the definition or quantification of thresholds. Regulatory options typically include criteria like setting time limits or quantifying decentralization in terms of owners, miners, exchanges, and so on to grow out of or sunset your time in the safe harbor.

Many DeFi protocols are experimenting with transferring control to the user base through governance tokens and evaluating decentralization through voter engagement or other popular metrics for decentralization like Gini or Nakamoto coefficients (Srinivasan and Lee, Reference Srinivasan and Lee2017). Of course, when it comes to decentralization metrics, a natural question arises about how to weight accounts when users can have multiple accounts and some accounts are not active. But these are exactly the kind of questions that regulators, developers, and academics can collaborate on answering with real-world data after the fact (e.g., see work on re-centralization by Appel and Grennan (Reference Appel and Grennan2023) and Cong et al. (Reference Cong, Rabetti, Yan and Wang2025)).

The size argument is a key advantage of safe harbors. Regulatory wait-and-see approaches are often justified because innovation in its infancy is too small to have a meaningful impact on financial stability. Safe harbors allow regulators to focus on FinTech applications that have scaled up to the point where they could impact the system. Fortunately, by providing a safe harbor, the regulators incentivize the developers and inventors to stay in the U.S. and engage with the current system rather than relocating to other jurisdictions.

In an insightful model of regulatory uncertainty, Campello et al. (Reference Campello, Cong and Dietrich2024) analyze FinTech development through a strategic game between innovative firms and a resource-constrained regulator. The framework captures how innovation rates depend on regulatory budgets, skills, and the number of competing innovators, with a key finding that multiple equilibria arise from the complementarity between regulatory preparedness and innovation investments. While the model’s binary regulatory instruments (approve or ban) often capture the current crudeness of current FinTech regulation in the U.S., incorporating sunset provisions would address the multiplicity problem. Sunset provisions usually have time-limited approvals with renewal requirements. This could be a powerful equilibrium selection mechanism by creating a coordination device that anchors expectations and eliminates inefficient equilibria.

Taking the insights to the real world, it becomes clear that sunset provisions help formalize post-innovation learning, allowing regulators to allow innovators to enter the markets early and efficiently while preserving the option to terminate harmful innovations based on outcomes observed after the fact. While this has not been tried much with financial innovation, evidence from medical device regulation offers an intriguing parallel, as it shows post-market-entry surveillance meaningfully increases consumer surplus through reductions in consumer uncertainty (Grennan and Town, Reference Grennan and Town2020). Approving FinTech products early but reviewing additional data after some time would encourage experimentation with potentially valuable innovations that might otherwise face outright bans.

By combining collaborative governance models like safe harbors and sandboxes with flexible legal frameworks such as sunset provisions, regulators could provide more concrete guidance for FinTech startups. In doing so, they could also achieve an adaptive regulatory system that better matches the new open-source, composable, customizable, easy-entrepreneurial entry financial ecosystem. In this spirit, regulators could even sunset their involvement in the regulatory system and allow RegTech tools or hybrid approaches to self-regulation to take over after a certain threshold of stability or maturity is achieved.