2.1. Literature review of AI and expert translators in IR

Analyses of AI policymaking must involve an understanding that AI is an ambiguous, constantly constructed concept. As a general-purpose technology (GPT) of the future, nothing about AI is given – in essence, in definition, or in practice. In the broadest sense, this paper defines AI as a combination of algorithms, large amounts of data and computational processing power that can adaptively self-improve in performance, cross-apply learnings to new tasks not explicitly programmed within the technology, and is capable of producing outputs or products that would ordinarily require human intelligence (Stryker and Kavlakoglu Reference Stryker and Kavlakoglu2026). Like other GPTs, AI may play a dominant role across security and economic functions and provide an advantage to the governments and/or militaries able to quickly make use of it, and it will also diffuse across a multitude of sectors (Scharre Reference Scharre2023). The legal realm has wrestled with defining AI, as scholars recognize “it isn’t any one thing” – AI can “refer to systems that play games, produce coherent text, predict protein structures, diagnose eye diseases, or control nuclear fusion reactions” (Schuett Reference Schuett2023). Experts have coalesced around a general understanding that AI will have world-changing effect as it matures as a technology, but at least 55 analogies were used by 2023 in major public discussions or media coverage of AI (Maas Reference Maas2023).Footnote ¹ While AI and other disruptive technologies have been securitized in the public sphere, core to AI’s concept are also countervailing or at times contradictory narratives about dual-use purposes that also benefit humanity or, for instance, industrial productivity and thus one nation’s economy. References to AI have become commonplace in public discourse over the past two decades, especially given newsworthy technological advancements, but pinning down what AI will do for us (or the ultimate ceiling of what current forms of AI will be able to do once fully scaled) remains riven with disagreement even within scientific fields. Moreover, there exist high technical barriers for understanding and shaping AI’s essence fairly removed from the traditional profile of political leaders, who are nonetheless required to make crucial policy decisions about AI.

This paper takes steps to analyze the epistemic space surrounding AI to better disaggregate the messages and frames getting amplified from top-down state doctrine. This work diverges from established scholarship in IR that works within the rationalist tradition of IR have focused upon how AI might fuel security dilemma dynamics, due to issues of reciprocal threat perception, secrecy and uncertainty (e.g. Geist Reference Geist2016; Horowitz and Lin-Greenberg Reference Horowitz and Lin-Greenberg2022; Jensen et al. Reference Jensen, Whyte and Cuomo2020; Sharikov Reference Sharikov2018). While some forays into modeling state behavior in AI-related decision-making could see AI-based techno-security competition as a modified form of a traditional prisoner’s dilemma or similar “game” pertaining to “arming” a state (e.g. Armstrong et al. Reference Armstrong, Bostrom and Shulman2016), few of the analyses around AI have adequately linked domestic, sub-national discourses and instead choose to take a unitary view of state decision-making – despite the fact that truly frontier AI expertise for the first time lies outside of the purview of national laboratories or public sector knowledge banks. Thus, the modeling of state behavior in the AI space from a rationalist standpoint provides an incomplete picture: within these multilayer problem settings, states do not necessarily “share the same representations of the world,” and, to come to a consensus on how best to implement or regulate AI, “states must come to have similar understandings of what the global is made of and how it works” (Allan Reference Allan2017).

This paper instead looks to extend upon the existing literature on epistemic communities to continue a focus on the role of hybrid actors in politics. IR scholarship has already grappled at times with how expertise – usually drawn from other sectors, such as the hard sciences – can influence state behavior, going as far back as the literature on epistemic communities (Haas Reference Haas1992) as well as responsive scholarship critiquing the original epistemic communities’ framework due to the fundamental inseparability of knowledge production and power and further discussions of the “crossing over” of that expertise or knowledge from the academic sector (Litfin Reference Litfin1994). Allan highlighted the role of substate agencies as well as the “contingent influence” of scientific experts and other actors in the processes of “designation, translation, and problematization” (Allan Reference Allan2017). Scientists and experts can use their authority to reframe objects in along the lines of security or other power-related subjects, link disparate objects together and prescribe certain ways of dealing with objects all in public discourse; ultimately, too, this is more of an active process of translation and shaping of discourses in the public sector, rather than a phenomenon that is automatic or constant across all forms of expertise (Allan Reference Allan2017, 142).

While all these expansions, extensions and critiques of the initial “epistemic communities” concept have added much-needed dimensionality to this sector of society as it impacts IR, they need some updating to fully engage with the contemporary landscape of AI discourse and policymaking. Already, authors like Roberge et al. have looked at attempts to “translate” AI (Roberge et al. Reference Roberge, Senneville and Morin2020). As AI diffuses as a subject matter from niche fields of science into the policy realm, “translation,” akin to Litfin’s conceptualization of “knowledge brokers” comes into play across hype, predictions, metaphorical references and other discussions that are being had publicly trying to make sense of the technology. That translation does not occur through a “disembodied” fashion – instead, they ask: “Who speaks? Through which channels and with how much success?” (Roberge et al. Reference Roberge, Senneville and Morin2020, 2). Schiff’s work looked at policy formation in USG for AI from 2016 to 2020, positing that “policy entrepreneurs” and “problem brokers” are necessarily separable classes of actors that raise problems and solutions in turn and ultimately formulate a policy agenda (Schiff Reference Schiff2023). Schiff’s summary stops short of further theorization of the specific identity of the “problem brokers” and policy entrepreneurs in question but instead focuses on the substance of the narratives at play. Other analyses have examined the narratives or arguments themselves that result in specific policy prescriptions about AI. For instance, Buchanan and Imbrie taxonomized the way that AI has been politicized, generating three categories of interest: “Evangelists” advocate for AI to be implemented as fast as possible to solve key political and social issues, while “warriors” identify AI usage by hostile entities as an impending threat and make the case for heightened implementation of AI as a form of deterrence against future harms, and “Cassandras” see AI itself as a threat that is reorganizing society or as something that will cause harm from its immaturity, cautioning against indiscriminate applications of AI technology across critical services (Buchanan and Imbrie Reference Buchanan and Imbrie2022).Footnote ²

This article probes deeper into one of the cases where a very specific type of “knowledge broker” seemingly has made a major difference in policymaking outcomes – specifically, that of centering AI within U.S.–China geopolitical competition. This new theory adds empirical context that can help clarify why some knowledge brokers more effective than others. In AI, rather than merely valuing hands-on engineering experience or advanced degrees, elite policymakers specifically look to those who control the underlying technologies themselves as technical authorities. Power in the AI sector has been concentrated within just a few firms, which has meant that epistemic expertise and credence has also been thereby concentrated in the private sector and among its spokespeople. Successful knowledge brokerage, in this setting, leverages the existing platform and adherents of each of these figures while controlling knowledge’s production and interpretation simultaneously in a way unmatched by purely academic scientific experts and the public sector.

Extant theories around interest groups and corporate impact on policymaking assume that primarily financial relationships dictate most of policy outcomes through observable mechanisms that interfere with democracy. However, often, technical expertise can transcend the forces of moneyed power or corporate lobbying, as, in the AI era, frontier AI founders do often have private company backing and funds but have personalist status that arises from their perceived expertise as firsthand witnesses of AI advancements in the field. This article instead explores the deference to hybrid epistemic authority within sense-making about AI. The next section conceptualizes and operationalizes the concept of hybrid epistemic authority and examines the concept of “hybridity” in action more broadly outside of the AI realm, ahead of the empirical portion of this article.

2.2. Defining hybrid epistemic authority

Unlike a more traditional format of expert-hood or epistemic communities, who work within institutional domains (the traditional academia to government pipeline, for example), hybrid epistemic authority as terminology reflects the case in which a person occupies multiple strata as, for instance, technologists and private corporate founders or business executives, as well as, simultaneously, public intellectuals and even straightforwardly political actors through informal or formalized advisory roles or eventual government agency positions. Here, hybridity thus specifically refers to the simultaneous informal or formal occupation of traditionally distinct or not as commonly overlapping domains, such as technical or commercial roles, public intellectual-hood, and engagement or presence in the public sector.

This article does not argue that its analysis of the “hybrid” nature of these new “knowledge brokers” in the field of AI is definitionally unique to AI. Other fields have hybrid epistemic authority in action. Broader analysis of the “revolving door” phenomenon between public sector roles and private sector roles involves some conceptualization of this phenomenon (e.g. Tsingou and Seabrooke Reference Tsingou and Seabrooke2009), but engagement with the epistemic aspects of status – beyond moneyed connections and interests – has been more limited.Footnote ³ For example, Li has written about how individuals can wield the “revolving door” to build credibility through “credentials, bureaucratic competence, and claims to policy expertise” constructed via career moves: “Spinning through the revolving door – moving between sectors multiple times – allows individuals to renew capital that might be fading by collecting new credentials, refreshing bureaucratic know-how, and developing broader claims to policy expertise” (Li Reference Li2025, 10). This literature thus contributes to the broader field studying hybridity across sectors of governance and life, both as a descriptor of individuals and entities, as well as a descriptor of forms of power, when it comes to relations between corporations and their state-hosts. Gjesvik, for instance, noted that, within a study of the weaponized interdependence of key cyber infrastructure, the assumption that a state and private companies within are automatically aligned has increasingly deteriorated (Gjesvik Reference Gjesvik2023, 722). Similarly, Gertz and Evers also highlighted this trend pointing out that “close, cooperative relations between businesses and the government act as a force multiplier” for state power (2020). The authors argue that alignment between commercial interests and state security interests is not necessarily a given but is actively being constructed and contested by each side. Much of this analytical work on a “revolving door” or sectoral hybridity has taken place at the systems or organizational level, and less so at the individual, but individual hybridity within the realm of technology has often been cited by public policymakers as a necessary competence rather than a sign of potential capture within the role. For instance, President Obama celebrated Megan Smith in 2017 as the first U.S. Chief Technology Officer “with technical experience” as she crossed over from Google to join the White House (Fessler Reference Fessler2018). In other sectors, such as more traditional defense contracting, critics have often pointed out when new appointments of leaders rotating back into government directly from working at companies like Raytheon might pose a conflict of interest, although, even if there is more criticism in public fora, rarely has that pushback been enough to prevent the confirmation of defense leaders (Turse and Emmons Reference Turse and Emmons2020).

Most forms of exposure to the AI private sector and hybridity between the AI sector and areas of public policy are celebrated by political decision-makers in some countries rather than scrutinized as a potential conflict of interest. As a recent instance, scholarship covering the EU’s creation of the Global Tech Panel analyzed the invitation of corporate actors to join decision-making processes, finding that the AI world is subject to “an emerging hybrid regulatory security state based on ‘liquid’ forms of epistemic authority that empowers corporate actors but also denotes a complex mix of formal political and informal expert authority” (Bode and Huelss Reference Bode and Huelss2023, 1230). As another example, Sam Altman, in testifying to Congress about AI, embodied hybridity as OpenAI’s CEO and that position’s concomitant business interests and a public expert and epistemic authority on AI as an intellectual subject matter, and Congress members asked him directly for advice about what to do about AI (Fung Reference Fung2023). Private AI developers often take on this dual-mantle of being technical experts, political advisors and commercial actors with very little challenge from government.

What makes this manifestation of hybrid epistemic authority a particularly urgent field of study stems from how the unique aspects of the AI landscape have contributed to these actors’ rise. As a technology that is referenced often as one of the defining emerging disruptive technologies of the 21st century (e.g. Johnson Reference Johnson2019Footnote ⁴), AI has a specific, threefold nature, being highly technical, general-purpose and highly speculative. Combined with the fact that a highly select few well-capitalized private industry laboratories are firmly at the cutting-edge of delivering this new technology, these facts arguably coalesce to promote the focal role of hybrid epistemic expertise in shaping and framing narratives, and, ultimately, policymaking around AI. This article notes that these structural conditions are not unique to AI only; rather, that the status of the AI industry today and specific qualities of AI enable these forms of hybrid authority to be particularly salient. Though beyond the scope of this article specifically to assert, this article would hazard that these kinds of hybrid actors would have most relevance in other fields of science where there is high technical complexity, a concentration of inputs and innovation in the private sector only, any high-stakes relevance to national security of a dual-use or general-purpose nature, and a rapidity of innovation.

3.1. Eric Schmidt, tech and defense

Beginning from around 2016, former Google CEO Eric Schmidt’s hybridity across defense, elite tech and policy spaces enabled him to take on the role of a “translator” of how to view the rise of AI and what to do about it within national security realms. He was afforded strong leeway to shape how AI was ultimately framed within U.S. policy discourse, and, particularly, in how to regard China’s role and rise. Schmidt holds prominence as a former tech-affiliated leader with deep ties to the security sector, including through roles held on the Defense Innovation Advisory Board and other positions; even while he was still at Google, Schmidt reportedly already held influence as a donor and senior adviser during the Obama administration and Hillary Clinton’s tenure as Secretary of State. Schmidt was Google’s CEO from 2001 to 2011 and then served as Executive Chairman until 2015, and so he had credence from overseeing the company’s development as a highly dominant technology cloud/search engine platform and big data product powerhouse, and he maintained a strong tempo of public speaking and media appearances to build wider recognition. His becoming an intellectual, versatile “translator” between Silicon Valley and Washington was a personalist strategy specifically for Schmidt, distinct from Google as a company, and his personal wealth in the tens of billions amplified his access to policymakers and tech executives alike. Schmidt embodies liminality and hybridity, given that he nurtured his role as “translator” particularly after departing from Google leadership, though while still maintaining many venture-capital commercial ties to the AI and tech sectors in general.

As McInerney has pointed out – before 2016, there were fewer than 300 Google results for “AI arms race,” with very few articles discussing the topic as a possibility (McInerney Reference McInerney2024). One can also note the absence of what would later become ubiquitous references to such a “Racing” narrative in the early years of this period under study. One of the final acts of the Obama administration, on January 1, 2017, had been the President’s Council of Advisors on Science and TechnologyFootnote ⁶ (PCAST) report on “Ensuring Long-Term U.S. Leadership in Semiconductors.” This report identified semiconductors as a key supply chain component of many critical technologies, of which AI is one, but the report mainly tapped into existing conversations about how China’s industrial policies were often seen as unfair to non-Chinese sources or manufacturers rather than raising any novel concerns about AI and “Racing” (PCAST 2017). The early days of the Trump administration also did not see AI as a particularly strong feature. Throughout 2017, the USG was not yet pursuing a centralized, security- or competition-oriented framing around AI itself. Instead, under the Trump administration especially, there were limited White House-led initiatives to crack down on perceived imbalances in trade. AI-specific national security-relevant initiatives were mainly confined to the defense and intelligence sectors.Footnote ⁷ Similarly, before 2017, Schmidt in public often spoke in general terms about his own personal worries for the field of AI; even on February 16, 2017, he noted his primary concerns of internet “Balkanization” if “countries lock down their borders to prevent citizens’ personal information flowing,” as well as the fact that “states are developing … AI-powered cyber-weapons” (Thomson Reference Thomson2017). He did not publicly speak about China very much. In a previous 2013 book, he wrote that “China’s future will not be bright,” and has also spoken in the past about the benefits of Google’s engagement with China (BBC News 2011; Maslin Reference Maslin2013).

Things changed by late 2017, at which point Schmidt talked very forcefully and publicly about how the U.S. was mishandling “Racing,” such as at the November 13, 2017, keynote at the Center for a New American Security’s AI and Global Security Summit (CNAS 2017). Schmidt’s words throughout November 2017 – in multiple forms and outlets – were some of his first and most public from figures of his caliber that cast a critical light on China’s AI development as a direct threat to U.S. competitiveness. This article does not assume that Schmidt’s about-face came from nothing, as it surely also reflected the “mood music” of the era in how China was being regarded as a whole by U.S. policymakers, especially during the early days of the Trump administration – though this article would also argue that AI did not receive very much attention as an aspect of this intensified competitive posture until after Schmidt’s words gained uptake. This article also does not point out this trajectory to make the normative argument that Schmidt ought not update his beliefs or arguments when faced with new information. In fact, Schmidt referenced in his CNAS speech that he had spent time reviewing new documents from China detailing state planning around AI, which then drove him to speak up (CNAS 2017).

Temporally, however, it is notable that this November 2017 moment then led to Schmidt taking even more active steps to shape the narrative that AI, if capitalized upon and fully realized by China in its vision to be an AI powerhouse, would pose geopolitical risk. Schmidt’s about-face in 2017 also enhanced his hybrid credibility in public to interpret technical AI developments or news as critical national security-relevant takeaways, using his unique positioning as a former Google executive to pivot toward more formalized defense advisory roles.

Just months later, the defense sector had already picked up Schmidt’s message, which he had also been agitating for privately at the Defense Innovation Advisory Board meetings in the prior year. The 2018 National Defense Strategy (NDS) was published in unclassified form on January 19, 2018, and drew attention to “technological developments” becoming more available for “state competitors and non-state actors” alike, “a fact that risks eroding the conventional overmatch to which our nation has grown accustomed” (U.S. Department of Defense 2018a). The NDS formalized China as a strategic competitor in an era of “rapid technological change” (Ibid.). This framing grew outward from the defense silo and continued to amplify Schmidt’s own role in steering this policy trend. On May 15, 2018, the Defense Secretary Jim Mattis wrote to Trump urging him to enact a national AI strategy by arguing that the U.S. was not “keeping pace with the ambitious plans of China and other countries” (Metz Reference Metz2018). Mattis’ memo reportedly called for a government-led “commission” to see how AI momentum in the private sector could be better leveraged for national security and to respond to China.

Subsequently, on August 13, 2018, the National Security Commission on AIFootnote ⁸ (NSCAI) was established, chaired by Eric Schmidt. Schmidt was a notable choice of chair, compared to an academically-credentialed authority or a lawmaker/policymaker or military figure. The duties of the Commission overall were “to consider the methods and means necessary to advance the development of AI, machine learning, and associated technologies to comprehensively address the national security and defense needs of the U.S.” (NSCAI n.d.). The NSCAI over the next few years submitted submit interim and final reports to Congress and other branches of the Federal Government. This appointment for Schmidt further crystallized his influence by providing a formalized para-structure adjacent to many fields of national security policy for his sanctioned views to continue to be cited in tech and security policy for the U.S.

On September 7, 2018, the Defense Advanced Research Projects Agency (DARPA)Footnote ⁹ announced a $2 billion investment over five years into AI advancement, referring to the need to fund a “Third Wave” campaign for adaptable autonomous machines (Harwell Reference Harwell2018). An expert at the Center for a New American Security called this announcement “The first indication that the U.S. is addressing advanced AI technology with the scale and funding and seriousness that the issue demands … China [is] willing to devote billions to this issue, and this is the first time the U.S. has done the same” (Ibid.). This initiative emerged out of the defense establishment that in this era so often continued to parallelize the work advocated for by Schmidt both directly and indirectly, underscoring the way Schmidt had amplified the importance of AI-specific strategies driven by the military–defense sector in order to keep up with and outcompete China.

On November 8, 2018, the Department of Defense (DoD) published its AI Strategy, charging the DoD’s Joint AI CenterFootnote ¹⁰ (JAIC) to carry out encouragement of AI development and use within the DoD. AI is equivocated to power and vastly expanded capacity, and the strategy document argued that “China and Russia … are making significant investments in AI for military purposes, including in applications that raise questions regarding international norms and human rights” (U.S. Department of Defense 2018b). Around this time, it was reported that the former U.S. congressman from Texas (and the top Republican on the Armed Services Committee from 2015 to 2019) William McClellan “Mac” Thornberry traveled around the U.S. speaking to industry leaders and other community stakeholders about the dangers posed by Chinese surveillance technology. He later said of these visits at a keynote that “There’s a real competition about what the future is going to look like between government control and not,” juxtaposing the Chinese approach to surveillance with the U.S.’ one of freedom (Kaye Reference Kaye2022d). And during 2019, the “Racing” framing was amplified further. On February 11, 2019, the Trump White House issued Executive Order (EO) 13859, “Maintaining American Leadership in Artificial Intelligence,” which reflected the DoD’s strategy, establishing the “American AI Initiative” to promote U.S. leadership in developing technical standards around AI (Hine and Floridi Reference Hine and Floridi2024).

The White House increasingly unified its approaches against Chinese industrial policies that created perceived unfair advantages for specific Chinese entities with “Racing” logics. New top-down policies of enforcement and prohibition crystallized efforts for a “decoupling” from Chinese technologies and infrastructures; with a greater understanding of AI as a competition driver also came a wider understanding about the actual interlinked nature of technology infrastructure and the broader innovation system. On May 15, 2019, President Trump declared a “national emergency” over “threats against American technology” via EO, allowing the Commerce Secretary Wilbur Ross to block any transactions involving critical technologies or infrastructure that “poses an unacceptable risk to the national security of the U.S” (The White House 2019). The order added the major Chinese technology supplier Huawei onto the “Entity List”Footnote ¹¹ of companies that are regarded as threats to U.S. national security. The U.S. State Department emphasized this rhetoric, describing Huawei as an “arm of the Chinese Communist Party’s surveillance state” (Macias Reference Macias2020). The claims around Huawei and other Chinese providers invoke the threat of China’s policies for civil–military fusion – not just as a reason for Chinese firms’ advantage that the U.S. ought to respond to through uplifting its own firms, but also as a reason to be wary of even civilian-oriented technologies from Chinese companies (Higgins Reference Higgins2019). On May 21, 2019, U.S. policymakers engaged in efforts to warn away would-be private sector investors from engaging with Chinese markets further (Byers Reference Byers2019).

Additionally, during the NSCAI’s first conference on November 5, 2019, Secretary of Defense Mark Esper called for an increase in partnerships across the tech industry and academia with government to prevent the Chinese military from gaining a “decisive advantage” (Han Reference Han2019). “Racing” framing was gathering steam following Schmidt’s amplification among military leaders. Also in November 2019, the U.S.–China Economic and Security Review CommissionFootnote ¹² released its 2019 report to Congress, which stated, “U.S. economic competitiveness and national security are under threat from the Chinese government’s broad-based pursuit of leadership in AI … China’s ability to capitalize on new technology has been enhanced by what it learned or stole from foreign firms” (U.S.–China Economic and Security Review Commission 2019). On August 20, 2020, the Department of Commerce added more “Huawei affiliates” and fine-tuned the export controls levied on Huawei under the Export Administration RegulationsFootnote ¹³; while the impact of these measures is still being calculated overall, Huawei did lose access to advanced chips, and the sanctions reportedly had a “debilitating effect” on the company (Federal Register 2020). On November 12, 2020, the Trump administration declared a national emergency through the International Emergency Economic Powers Act and EO 13959 which prevented securities investments going to certain companies in China, and which was later expanded by the Biden Administration through EO 14032 in June 2021 (The White House 2021a).

Efforts in preventing China from shoring up a semiconductor advantage have also been used to justify bringing in the Taiwan Semiconductor Manufacturing Company (TSMC)Footnote ¹⁴ to invest $12 billion in Arizona for building a new manufacturing plant within the U.S. for semiconductors on May 22, 2020 (Aldane Reference Aldane2023). Later, on May 31, 2020, it was reported that the American semiconductor industry would start lobbying for additional $37 billion from federal sources to support factory building and semiconductor R&D in order to “keep the U.S. ahead of China and other countries” (Davis et al. Reference Davis, Fitch and O’Keeffe2020).

Internal documents – for instance, the Congressional Research ServiceFootnote ¹⁵ report on AI and National Security published on November 10, 2020 – also took up this language as well, referencing the fact that China was a leading competitor rivaling the U.S. in AI, though the report also called for caution when applying AI to military contexts (Sayler Reference Sayler2020). The report named China as a leading competitor rivaling the U.S. in the AI market, though calls for caution due to the risks that AI brings to military contexts.

During this period, mentions of China’s surveillance technology – especially among right-wing members of the House of Representatives or the Senate – indicated an additional dimension of “Racing,” which involved a “values competition” underlying this race. For example, the U.S. Undersecretary of State Keith Krach’s Techno-Democracies-10 plan was announced on May 1, 2020, and involved 10–12 countries that would focus on “the development, protection, dissemination, development and use of emerging technologies” and “serve as the stewards of technology norms” (Krach Reference Krach2020). Membership included states as well as private sector and academic members across the 10 critical innovation sectors of the Economic Security Strategy, which includes semiconductors, AI, and other advanced applications. On February 14, 2020, during the Munich Security Conference, U.S. House Speaker Nancy Pelosi warned Europe that allowing Huawei to implement 5G infrastructure within their borders or allowing their systems to become dependent on Huawei infrastructure in any way, would result in the triumph of “autocracy over democracy” (Warrell and Peel Reference Warrell and Peel2020). USG CTO Kratsios critiqued draft plans for EU AI regulation, calling it out as getting in the way of the U.S.’ efforts to counter “China’s digital authoritarianism” (Council Reference Council2020).

Schmidt began authoring more public-facing articles and pieces from 2020 onward further spreading this viewpoint of the U.S. risking falling behind China in the AI race, including a popular op-ed with Harvard academic Graham Allison, who achieved more recent renown through popularizing the “Thucydides Trap” framework that drew upon historical case studies to argue that U.S.–China conflict was inevitable (Schmidt and Allison Reference Schmidt and Allison2020). He later followed this up with a book coauthored with Henry Kissinger and Eric Huttenlocher in 2021, titled The Age of AI, and subsequent op-eds for the Wall Street Journal. These engagements and publications that further bridged the academic and policy worlds greatly aided in mainstreaming “Racing” in the lexicons of both communities.

Furthermore, on August 1, 2020, Secretary of State Mike Pompeo announced the “Clean Network,” which gathered an “alliance of democracies and companies” based on democratic values. More than 60 nations and 200 telecom companies jointly signed up to the Clean Network – although several researchers noted that the proposal was met with “indifference” among major European nations and added to concerns about a fragmenting internet (Boadle Reference Boadle2020; Wadhams Reference Wadhams2020). The Wall Street Journal, on the other hand, hailed the Clean Network Alliance as potentially the “most enduring foreign-policy legacy” of the Trump administration (Braw Reference Braw2021).

One of the biggest accelerants on this discourse was a landmark document published on March 1, 2021: the NSCAI’s final report, overseen by Eric Schmidt as previously stated (NSCAI 2021). Although technically, the document was an unofficial, nonbinding report that contained both fact-finding and recommendations to government, a number of public statements and additional written sources confirmed that the NSCAI final report stimulated discussion and change, especially within the DoD (Simonite Reference Simonite2021). Schmidt, alongside his associates recruited to staff the Commission, embedded his specific vision of AI – as a security-oriented asset of critical national security significance – into the building blocks of executive branch policymaking. The report advocated for wide boosts of defense funding for AI R&D up to $32 billion a year by 2026, the growth of computer science talent in the public sector, and the prioritization of AI in the DoD. All of these recommendations were justified through “Racing” language:

The NSCAI final report magnified the impact of Schmidt and his like-minded associates at scale – constructing AI as fundamentally inseparable from great power competition.

The Biden administration’s seminal texts and leadership statements softened some of the harder language used in the Trump administration about competing with China but continued to frame American leadership as being important bulwarks against inimical foreign competitors as well as a beacon of democracy in the modern age. During his first speech to Congress on April 28, 2021, he stressed the need for the U.S. to prevail in the technological race: “China and other countries are closing in fast” (The White House 2021a). Also in April, Schmidt continued to do press around the topic of the NSCAI report’s conclusions. Along with discussing the statements and recommendations of the report, Schmidt proposed an additional resonant frame that would also catch on at the upper echelons of government – especially the military – that AI “should be done with American values” rather than Chinese ones, arguing rather ahistorically that, for instance, U.S. values reject “discrimination” and “bias” (CBS News 2021). Here, Schmidt used a moral argument in addition to the competition-based one for why the U.S. should come first in AI competition, thus combining layers of rhetoric to further their influence as a sum total of defining AI policymaking.

The Biden administration passed some key EOs that attempted to rein in China’s supply chain leads and support domestic supply chain onshoring efforts. On June 3, 2021, President Biden signed an EO that prevents U.S. investors from investing in 59 Chinese companies, such as Huawei, China’s largest chipmaker Semiconductor Manufacturing International Corporation, aimed at making sure that investments from Americans are not funneled toward Chinese military efforts (The White House 2021b).

Since the NSCAI concluded its work and final report earlier in the year, the organization transformed on September 22, 2021, into the privately funded Special Competitive Studies Project (SCSP) organization, which states in its mission that it will “make recommendations to strengthen America’s long-term competitiveness as AI and other emerging technologies are reshaping our national security, economy, and society” and that it wants “to ensure that America is positioned … to win the techno-economic competition between now and 2030, the critical window for shaping the future” (SCSP 2025). The SCSP is a privately funded non-state organization and still includes Schmidt as chairperson of the SCSP and the former executive director of the NSCAI (who has a background at the NSA) as the president and CEO. Though privately funded, the SCSP states that it intends to be a callback to the 1956 Cold War Special Studies Project and has positioned itself as a vehicle to attempt maintaining the same reach and impact of the NSCAI (AI Now Institute 2023b).

Momentum continued into 2022. On February 1, 2022, the White House put out a list of critical and emerging technologies as a reference for additional national security measures, such as export controls or investment reviews (Fast Track Action Subcommittee on Critical and Emerging Technologies 2022). In some ways, the landmark CHIPS and Science Act signed into law by President Biden on August 9, 2022, could be seen as one of the major correlative accomplishments or side-effects of “Racing” language. The act was an amalgam of the CHIPS Act, the Research and Development, Competition, and Innovation Act, the Supreme Court Security Funding Act of 2022, and the rewritten version of a previous bill called the U.S. Innovation and Competition Act of 2021 (117th Congress 2022). The act is overwhelmingly geared toward competing with China over the semiconductor supply chain and additional science funding. Companies are also subjected to a ten-year ban prohibiting them from working on chips more advanced than a certain standard in China and Russia if they receive CHIPS Act subsidies; the Act incorporates a “foreign direct product rule,” and thus the Act extends to govern companies operating anywhere that work with U.S. technology and equipment (Ibid.). Over the next months, Biden charged the Committee on Foreign Investment in the United States (CFIUS) with examining investments into critical technologies including AI, and October 2022 saw further restrictions on selling semiconductors to China; October 7, 2022, included a directive from Commerce to “undercut Chinese AI military advancements” using export controls (The White House 2022a).

On September 12, 2022, the SCSP published its first report, centered on U.S.–China AI competition. The SCSP reportedly hosted board meetings and panel meetings with more than 225 experts – ranging from government stakeholders to tech industry personnel, academia and others – and held additional “engagements” with over 400 others (SCSP 2022). SCSP named “competition” as its organizing principle and identified technology as the “central element of the rivalry between the U.S. and China” (SCSP 2022, 3). SCSP’s nongovernmental status was offset by the high profile of the individuals it could convene under its roof and its overall influence within government; for instance, the former Secretary of State Henry Kissinger wrote the opening statement for this report. On September 16, 2022, National Security Advisor Jake Sullivan spoke at the SCSP Global Emerging Technologies Summit, emphasizing the need to shore up existing technical advantages and ensure that resources and intellectual property (IP) do not get to adversaries (The White House 2022).

Values-laden language initially used by Schmidt to compel the U.S. to implement AI quickly for the competition with China along economic, technostrategic and moral lines continued being used. This language called for more attention and funding according to “Racing,” and the narratives continued to trickle across to a range of agencies, committees and bodies in the USG. On September 29, 2022, Miriam Vogel, cochair of the White House National AI Advisory Committee, called at an event for democratic values to be “embedded” within AI that gets developed, which mirrors a call made in the NSCAI final report (Politico 2022).

More than ever, decision-making was grounded in being able to compete adequately with China. Representative Ro Khanna was quoted on October 1, 2022, as saying of the CHIPS Act that it wouldn’t be enough to combat China properly: “China, as Eric Schmidt says, does one of these Chip Acts every year” (Kaye Reference Kaye2022a). Similarly, FBI Director Christopher Wray gave a speech on January 20, 2023, arguing that China’s AI program would indubitably be weaponized, and so that the U.S. had to be prepared (Collingridge and Elliott Reference Collingridge and Elliott2023). The House Committee on China announced on February 12, 2023, that it hoped to spotlight human rights abuses, highlight Chinese “economic statecraft,” devise strategies for decoupling the U.S. ecosystem from China, and investing in AI, robotics and other technologies in order to compete better (Wong Reference Wong2023). On March 9, 2023, the U.S. Chamber of Commerce issued a report on AI that explicitly named China’s development of AI as a way of “establishing dominance” and called for the U.S. to adeptly handle “competition from China’s broad-based adoption of AI” (U.S. Chamber of Commerce 2023).

And leading into the next election cycle, more partisan leaders or figures began to lean into the language as well. For instance, Democratic Senator Chuck Schumer introduced the Global Technology Leadership Act on June 8, 2023, after vowing to address AI within his role. The Act would create an office (the “Office of Global Competition Analysis”) that would report on how competitive the U.S. is with reference to other near-peer technological competitors like China and would involve intelligence community, DoD, and other agencies coming together to provide input (Brown-Kaiser Reference Brown-Kaiser2023). And on June 15, 2023, Republican Senator Ted Cruz criticized his fellow congress members for not being “tech savvy” and cast aspersions on the USG’s ability to regulate AI well, all while referencing the threat from a tech-savvy China:

Following on from Schmidt’s strong level of involvement with the Defense Innovation Board and similar advisory bodies at the Department of Defense, the defense sector has since remained a clear voice calling for more attention to be paid to China’s specific capabilities and other metrics for determining the path of its tech sector. On July 17, 2023, the head of the CIA stated that AI – “the most profound transformation of espionage tradecraft since the Cold War” – would mean that the CIA may not “be able to keep pace with intelligence rivals like China, or keep ahead of them” (Aldane Reference Aldane2023). Meanwhile, the discussion also continued to gain even more resonant framing, taking on Schmidt’s proffered language from as far back as 2021 of a values-based “clash of civilizations,” such as with Air Force General Lt. Gen. Moore Jr. arguing within the context of AI competition that the Pentagon’s use of AI would always be more moral than alternatives: “Regardless of what your beliefs are, our society is a Judeo-Christian society, and we have a moral compass. Not everybody does … there are societies that have a very different foundation than ours” (Decker Reference Decker2023). This language underscores how the epistemic framing of “Racing” as a civilizational struggle, requiring whole-of-society mobilization around AI, became further entrenched within national security discourses from USG sources.

On August 2, 2023, three major former White House leaders – the ex-Joint Chief of Staff, the ex-CIA Deputy Director, and the ex-White House Counterterrorism and Homeland Security Advisor – all jointly wrote a piece encouraging more public–private collaboration to keep up with China’s AI progress and keep China “from winning the tech race and disproportionately influencing the future” (Dunford et al. Reference Dunford, Townsend and Michael2023). And Former Deputy Assistant Secretary of Defense for Strategy and Under Secretary of Defense for Policy Michele Flournoy penned a piece toward the end of 2023 for Foreign Affairs advocating for the USG to “accelerate – not slow – its adoption of responsible AI” – otherwise, “Washington could lose the military superiority that underwrites the interests of the U.S., the security of its allies and partners, and the rules-based international order” (Flournoy Reference Flournoy2023).

While Schmidt stands out in this section due to his centrality to USG policymaking on AI as part of the U.S.–China relationship, this article also highlights other tech industry spokespeople who took on this role, too.

3.2. Tech industry spokespeople

Even throughout the earlier years of this “AI race” narrative, industry actors in several instances have used “AI race” logics to criticize antitrust momentum in the U.S., advocate for better and faster government procurement or detract from regulatory efforts targeting Big Tech (AI Now Institute 2023b).

As one example, following the scandal around Cambridge Analytica, which showed how 87 million Facebook users’ data had been harvested for firms aiming to directly influence voting behavior, Facebook CEO Mark Zuckerberg was called for public hearings on April 10, 2018, during which he stated in response to pointed questioning by officials that American companies still ought to be able to harvest data more freely from consumers – “Or else we’re going to fall behind Chinese competitors and others around the world who have different regimes for different, new features like that” (Lomas Reference Lomas2018). Similarly, on May 23, 2019, both Facebook COO Sheryl Sandberg and Eric Schmidt called for a relent on criticizing big tech companies, stating that “Breaking up Big Tech will only help China” (Tiku Reference Tiku2019). Sandberg directly linked the fears around the power of the tech sector in the U.S. to having a negative impact on America’s competitiveness abroad: “There’s also a concern in the U.S. with the size and power of Chinese companies, and the realization that these companies are not going to be broken up” (Ibid.). Schmidt also stated that “regulatory bias” was having a negative impact on American competitiveness which is best realized through its tech companies, arguing that “Chinese companies are growing faster, they have higher valuations, and they have more users than their non-Chinese counterparts” (Ibid.).

“Racing” messaging was already getting amplified by the defense sector, as well, and the broader conditions of geopolitical competition with China likely made it so that the resonance of “Racing” could not be avoided by strategic actors like tech industry leaders. Increasing references to “Racing” were made over 2019, 2020 and 2021. On September 13, 2021, the main tech industry advocacy organization Computer and Communications Industry Association (CCIA) harshly criticized antitrust, arguing that the U.S. “is at a critical inflection point in its innovation race with China” with high stakes, and antitrust would be “in serious tension” with the “overall U.S. national innovation strategy to combat China … and may inadvertently undermine U.S. national security” (CCIA and King & Spalding LLP 2021). Just two days later, 12 former security officials (who were then found to have been engaged in direct or indirect contracts with private tech firms) issued an open letter to Congressional leadership arguing that antitrust actions and bills against large U.S. tech firms would jeopardize innovation and empower China over the U.S. (Basu and McGill Reference Basu and McGill2021). And on January 18, 2022, Google’s president of global affairs posted a blog piece arguing that any further antitrust legislation for the tech industry would “[give] a free pass to foreign companies … and risks ceding America’s technology leadership and threatening national security” (Walker Reference Walker2022).

On April 20, 2023, the subcommittee of the Senate Armed Services Committee heard testimony from experts about how the U.S. couldn’t afford to hold back any further AI research and development while China inexorably moved ahead with its own programs. The language was value-laden and direct. The President and CEO of the RAND Corporation stated that “Democracies [should] lead the norms and standards around AI” (Kasperowicz Reference Kasperowicz2023). Meanwhile, the Chief Technology Officer of Palantir stated that a pause would provide China advantages in the AI field: “A democratic AI is crucial” (Ibid.). According to reports from the committee meeting, the AI experts called on the government to “quickly start incorporating AI into U.S. defense systems” (Ibid.). As these figures continued to also leverage their hybrid positionality, they joined the growing chorus of epistemic influencers seeking to amplify messaging around AI to serve their own purposes.

As the originator of AI technologies, modern tech companies hold outsized sway in this space, as policymakers often tend to defer to their expertise about technology governance, though critics of these “Racing” framings have pointed out that this narrative encourages the U.S. policymakers to think that anything good for big tech companies is good for the U.S. as well as the obverse, even as U.S. tech companies maintain some market interactions in China for commercial benefit (AI Now Institute 2023b). In defense of this framing, the Executive Director of NSCAI, Ylli Bajraktari, said, “How do we stay ahead and compete against China if we’re not able to utilize our private sector’s expertise and knowledge and advantages in this space?” (Ibid.) As mentioned, the concentration of AI development in purely private hands has resulted in these developers being seen as primarily authoritative interpreters and translators of what frontier AI ought to mean for the policy world, to far-reaching effect.

3.3. Analysis

Several elements are implicated in why “Racing” loomed so large over the 2015–2023 period. Within advanced technologies like AI, discourse over China’s revisionism has been coupled with a sense of surprise. As Suttmeier and Simon wrote,

USG actions in response to Chinese trade practices were domestically considered “long overdue” by some partisan groups; the Chinese government’s alleged practices of economic policies include but are not limited to IP violations, currency suppression, espionage and the preferential treatment of Chinese corporations (Bradford Reference Bradford2023). As a result of the policies pursued between 2017 and 2023 along the lines of “Racing,” well over 250 Chinese companies have been added to the “Entity List” preventing them from selling in the U.S. market (Ibid., 189). Infrastructural decoupling is taking root, to some extent: as of the 2020s, Qualcomm can no longer provide 5G chips to Huawei and Google’s Android system cannot be installed on Huawei phones (Ibid.: 190).

The “Racing” narrative became the mobilizing impetus behind a cascading series of policy decision-making, culminating in this paper’s examined time period with the Biden administration’s CHIPS and Science Act and further export controls on chips. Again, this article is not aiming to provide a uni-causal argument: it is not solely the fault of epistemic influencers that the “Racing” narrative’s framing was particularly trenchant and impactful in the public sphere. Some of this can be attributed not just to the direct work of epistemic influencers but also to the resonance of the “Racing” narrative’s framing and references due to the context of a heightening U.S.–China confrontation, given China’s rise and the traditional understandings of disruptions from power transition, including, for instance, the narrative of a “Thucydides Trap.” “Racing” as a narrative within AI policymaking and beyond has a self-perpetuating quality, as actors who for their own reasons desire certain policy outcomes might note how effective it is at motivating bipartisan support and thus lean into it, further perpetuating the narrative itself.

However, this paper argues that hybrid actors did play a particularly key role in how AI came to be understood within the broader context of U.S.–China relations as a central component of the countries’ economically competitive outlook. While Schmidt was not the only advocate active in the space – and, of course, this pivot toward intense “Racing” language over the 2017–2023 period cannot in the least be attributed solely to his work or influence – it is worth noting the extent to which his existing influence and power in relation to the USG as well as his standing as a scientific and policy authority in the space of AI worked overwhelming in his favor such that whatever framing or narrative he chose to begin advocating for would surely go far. Following on from the way that the NSCAI report resulted in “19 of its recommendations to Congress [being] included in the defense budget approved in December 2020” (Simonite Reference Simonite2021), “Racing” has now been implicated in vast expansions of official state support for private sector AI developers in the name of competing with China, as well as further encouraging the view that what’s good for big U.S. tech industry heavyweights is good for all of the U.S. and national security (AI Now Institute 2023b). Across U.S. political parties, “Racing” involves advocating for further development and rollout of AI and pushing back on any checks to that momentum such as binding ethical or regulatory frameworks and antitrust lawsuits: “[‘Racing’] has perhaps been the single most productive argument behind the proliferation of policy instruments that increase government support and funding for the development of AI” (Ibid.). As Bryson and Malikova summarize: “Our concern is that the AI Cold War is often mentioned in the context of discussing digital market regulation. […] These sectors have to date seen very little regulation… [and] market consolidation has been swift in each” (Bryson and Malikova Reference Bryson and Malikova2021).

Another, lesser-covered dimension of this “Racing” narrative is its racialization, which was recently unpacked in an article by McInerney, who argues that the rhetoric of an arms race has gone well beyond the bounds of “technological dominance” but has grown to be regarded at the civilizational level and “specifically draws on previous racialized configurations of anti-Asian sentiment, such as techno-Orientalism and the Yellow Peril” (McInerney Reference McInerney2024). She points out that arms race discourse has not stopped after casting the Chinese government as a threat but has even extended this to Chinese individuals, or citizens of other countries, building on a long U.S. history of xenophobia and anti-Chinese/anti-Asian sentiment, and evidenced as alive and well in the modern day by the Trump-era China Initiative which sought to root out spies within American universities and companies (Ibid.). The framing of the AI arms race as a clash between values systems has been particularly effective and resonant in U.S. politics, McInerney argues, because of historical resonance with Orientalist and racist modes of viewing the “other” in the East and a mistaken belief that the values systems across China and the U.S. are so distinct as to be incompatible with one another.

In summary, “Racing” grew to reference a wider, values-based competition with China, framing the exceptional role of AI as one of the focal points, if not the focal point, of ongoing competitive dynamics between states (Hine and Floridi Reference Hine and Floridi2024, 259). Broader concerns about the effects of “Racing” can involve those around arms racing. Analysts are worried that “Racing” narratives will contribute to the negative impact of arms race-esque thinking within the U.S. and China, leading to a “race to the bottom” on AI safety measures (e.g. Scharre Reference Scharre2023, 5). In short, some actors have perceived a trade-off between the speed of AI development and robust safety procedures and checks, and “Racing” pressures the implementers of AI to critical state functions to lean harder on the former over the latter, especially when presented with a narrative that a foreign challenger is going faster. Furthermore, “Racing” language – especially in the name of preserving shared democratic values against an authoritarian threat – may spread across to other allies of either country, which may also cause further de-prioritization on AI testing and evaluation.

While the next section will contend with several alternative explanations for the rise of the “Racing” narrative and its primacy in U.S. AI strategy that focus on other variables besides these hybrid experts, some authors have already pointed out that the involvement of hybrid experts – Schmidt as a key example, or otherwise – have resulted in an intensification of the fraught profit-driven ties between private technology companies developing AI and the “National Security State” (Rikap Reference Rikap2024). Schmidt’s own venture funds have certainly been involved in active shaping of the new procurement relationships further supercharged by the U.S. legislation he has positioned his expertise to inform, such as the CHIPS Act of 2022. As this paper has discussed, the relationship between the public sector and private AI firms is far from straightforwardly top-down and state-driven, given the unprecedented power within these relationships afforded to these few firms who control the inputs of AI and inculcate deeper dependence within the public sector due to the capacity-reducing nature of black-box contracting and procurement relationships and brain drain. Looking at the narrative power of these new “hybrid experts” sheds light on the extent to which these actors can access epistemic capital and indirectly or directly steer decision-making in public bodies in ways that have international ramifications and are often self-perpetuating.

3.4. Alternative explanations

The empirical sections above have described how Eric Schmidt and other tech industry leaders made a key difference in elevating the “Racing” narrative within policy, as opposed to competing explanations that might point toward other aspects of politics for why this narrative initially took on primacy.

While instead one could view the “Racing” narrative’s expansion as the result of standard interest groups from the tech sector getting involved with politics, it is not clear that commercial gains uniformly and clearly result from drumming up fear around AI – as if the technology is highlighted as too securitized, policymakers might opt for highly curtailing its deployment and usage in critical functions, as well as restricting the trade environment and cutting off access for multinational companies developing AI to other jurisdictions for selling their products. Furthermore, the conscious recasting of AI as a critical national security and military focus within U.S.–China strategic competition took place in a way that didn’t necessarily serve as a continuation of existing bureaucratic politics in the U.S.; for instance, documents that came out at the tail end of the Obama administration in late 2016 were specifically oriented toward exploring the way in which AI and automation might change the domestic workforce or capacity of the U.S., rather than seeing AI as a primary agent of geopolitical change and competition as Schmidt argued for beginning in late 2017. If anything, certain hawkish viewpoints might have seen AI as a small part of a vast constellation of possible angles of economic competition, especially as, for instance, a headline from the Wall Street Journal that read “China Gears Up in Artificial Intelligence Race,” but limited to discussing the post-AlphaGo moment relative levels of technical sophistication, startup funding and level of workforce automation rather than outright militarization (e.g. Yuan Reference Yuan2016). Trump’s Deputy Defense Secretary Robert Work spoke on AI as part of the broader Third Offset strategy that required specific help at the end of 2016, but his remarks were part of a broader set of conversations and perspectives that were oriented much more toward building up conventional forces in response to an “ascendant China” and a “re-strengthened Russia” (see further below); what’s more, the October 2016 White House “National Artificial Intelligence Research and Development Strategic Plan” makes no mention of China as the nation toward which the U.S. aims to exhibit international leadership in AI (Livingston Reference Livingston2016). While around late 2017, more and more U.S. and international press coverage reacted to the 2017 New Generation AI Development Plan from China, specific framing of an “AI race” between the two countries did not trickle into actual bureaucratic documents or decision-making within government beyond generalized defense statements until Schmidt sounded the alarm with his public remarks in November 2017. The at the very least associative relationship between Schmidt’s public declarations, the uptake of these statements in the media, and subsequent fervor and focus among DoD actors and other agencies across the USG on U.S.–China AI competition, and Schmidt’s eventual role with NSCAI, can clearly thus be identified.

Importantly, the involvement of hybrid actors that pivot between private incentives and exposure to frontier AI development and a public profile does not always seem to accompany pure company-based commercial profit motive as we might see in traditional lobbying, especially because, as mentioned, “Racing” is paradoxically double-edged for some of the epistemic influencers most active in the “Racing” space and does not always result in specific commercial gains for the firms developing technology particularly relevant to this “Racing” framing. “Decoupling” and accompanying policy or legislative actions have thus been oriented toward maintaining the U.S. competitive edge, shoring up any espionage-based or infrastructure-based critical weaknesses (especially if the U.S.–China relationship were to downgrade further), and inflicting pain on China’s own technological standing. However, these moves have arguably sparked added tension with China and spurred on additional attempts in China to become technologically self-sufficient and further divorced from a U.S.-allied supply chain for critical technology inputs, which in turn raises the profile of China’s threat in the eyes of the U.S. – almost like a continuously-reinforced “flywheel” for further mutually competitive behaviors. Other sources examining China’s own policy doctrine have called attention to the fact that Chinese leaders have classified threat it faces of potential “chokepoints” toward its domestic technology development, a resonant frame given China’s long historical narratives of “humiliation” at the hands of foreign powers on the basis of trade relations (e.g. Tan et al. Reference Tan, Dallas, Farrell and Newman2025Footnote ¹⁶). Analysts and policymakers are also worried about any “Racing” measures sparking further escalation: “The prospect of the ongoing tech war morphing into something even more dangerous, including a US–China military conflict, cannot be excluded” (Bradford Reference Bradford2023, 217). Especially given other contentious issues in the U.S.–China relationship, such as the status of Taiwan, there is always a chance of U.S.’ actions backfiring in a way that renders the U.S. less secure (e.g. Al Jazeera 2023). Furthermore, technological protectionism’s edge can be felt most acutely in some ways by the commercial tech sector itself, or by key private stakeholders in the U.S. economy at large (Bradford Reference Bradford2023, 190). Deutsche Bank estimated that the “tech war” would incur costs of more than $3.5 trillion between 2020 and 2025, which – due to mutual boycotts, supply chain reconfiguration, and “tech walls” that force companies to deal with other companies’ systems and networks – would cause at least $1.5 trillion of that cost to fall on tech companies themselves in some way (Nuttall Reference Nuttall2020).

As more illustrative examples, further restrictions have hurt both the acquisition of talented workers to the U.S. ecosystem and potential business prospects, both of which could be argued runs counterintuitive to the goals of those invoking “Racing.” On June 23, 2020, across the board, tech industry representatives condemned new Trump-era immigration restrictions – which were implemented in the name of personnel and IP security, among other justifications – including a freeze on temporary work visas and a refusal to allow students on online-only courses to remain in the U.S. The tech industry actors stated that the moves were myopic and would reduce the competitiveness of the U.S. versus other powers (Savitz Reference Savitz2020). Immigrations and Customs Enforcement (ICE) later rescinded the rule about foreign students on online-only courses after Harvard and the Massachusetts Institute of Technology (MIT) sued and 200 colleges and universities joined the challenge; tech firms Google, Facebook and Microsoft also joined the lawsuit (Caldera and Kurilla Reference Caldera and Kurilla2020). The nature of the technology ecosystem is still heavily interlinked, involving the transnational flow of information, especially due to the nature of software engineering and cyberspace, as well as the most popular dreams of where young people want to study and work. A March 16, 2022 report from Stanford’s Institute for Human-Centered AI stated that U.S.–China bilateral partnerships were the most common type of partnership in the AI space compared to all other possibilities (Zhang et al. Reference Zhang, Clark and Perrault2022). Given the considerable difficulties in acquiring skilled technical talent – especially in frontier AI research and other niche areas of expertise – these “Racing”-inflected measures would have hurt the domestic tech sector instead of helping it. Other lawmakers have criticized “Racing” language for being needlessly polarized and harmful to both people and broader U.S. interests; Representative Andy Kim, who served on Obama White House’s national security team, stated about “Racing” on February 12, 2023: “There’s a fine line between deterrence and provocation … [Cold War comparisons are] also just false” (Wong Reference Wong2023). Scholars have also warned against blank comparisons to the Cold War era: “Policy-makers should examine new claims of a ‘race’ in critical technologies dispassionately and rationally and beware of suboptimal arming in response to claims of adversary capabilities … History … shows that such fears can be overblown and costly” (Belfield and Ruhl Reference Belfield and Ruhl2022).

More tech sector activity has also come under scrutiny due to the sanctions, export controls and other “trade war” type tactics employed by the U.S. to stay ahead. The U.S. Semiconductor Industry Association stated in 2020 that they were against new export controls: “Commercial products [sold] to China drive semiconductor research and innovation here in the U.S., which is critical to America’s economic strength and national security” (Ravi Reference Ravi2020). Tech companies were also cited in September 2022 as worried that CFIUS enforcements may skew too broad, given that AI has been a fundamentally cross-border technology involving inputs across countries (Kaye Reference Kaye2022d). Also in September 2022, U.S. senators called for the U.S. intelligence community to examine a deal between Apple and a Chinese chipmaking firm on the basis of national security (Sevastopulo Reference Sevastopulo2022). Companies have also cited frustration with onshoring efforts: though TSMC and Taiwan as a whole have benefitted from TSMC’s significance to electronic supply chains in the U.S., reports around the decision to bring some of TSMC manufacturing to Arizona stated that TSMC had done so “begrudgingly” (Aldane Reference Aldane2023). Generally, tech companies seek economic opportunities across new markets, and export controls often chip away at the bottom line.

Thus, use of the “Racing” narrative has some risks toward bottom-line profitability for tech firms. And, of course, tech industry players have as a result not all pushed the same variety of narrative. Private industry figures have occasionally sounded similar to one another on some of these issues, but they have also provided some countervailing points of view from time to time. This discordance also lends credence to the fact that this is not merely “interest group politics” or traditional lobbying as an overarching phenomenon – as otherwise the tech industry might see more uniformity across its spokespeople or influential founders’ logics and advocacy. Some tech companies have previously encouraged some restraint in the decoupling process (Bradford Reference Bradford2023, 191). For instance, in 2021, Microsoft even launched further presence in China, dedicated to mutual causes of AI and 5G research, in partnership with Chinese research universities (Kaye Reference Kaye2022c). Instead, this article points toward a more personalist and individualist strain of politics, in which hybrid actors at the intersection of private interests and public platforms – particularly provided their platform due to the inherently privatized and power-concentrated nature of AI – uniquely are able to shape sense-making and narrativization around AI in public policymaking.

Nonetheless, this article also does not argue that U.S.–China competition is bad for business overall; just, instead, that the relationship between Big Tech and the U.S. security apparatus is a complicated one, and that some hybrid influencers have been faster to capitalize on the “Racing” narrative for personal, monetary and political benefit compared to others coming from the same field. Rikap has also mentioned the “strategic yet asymmetric alliance” felt and the growing military-tech industrial complex that itself will prove a ripe subject for future research efforts, growing from and extending far beyond just the “Racing” narrative (Rikap Reference Rikap2024).

Along those lines, another potential explanation for this phenomenon might stem from Schmidt and similar figures’ ability to leverage purely personal capital with leading figures in power, such as President Trump or President Biden. But Schmidt’s influence clearly extended beyond the mere personal connections facilitated by his prior roles and wealth; these markers would not result in the deference toward Schmidt as an authoritative translator of technical news into geopolitics, and the extent to which Schmidt and similar current and former tech leadership figures continue to be referenced as defining what AI competition meant, and what responses would be appropriate. Additionally, the hybrid experts mentioned throughout this paper are distinct as a class of actor from more traditional forms of scientific expertise or academic expertise that might beget authority in a field. One can compare and contrast the high penetration of the “Racing” narrative as particularly amplified by the hybrid epistemic experts compared to other narratives on what to do about AI. These potential alternatives for a comparative study might range from highlighting AI as a source of existential threat to humanity that should thus be paused to AI as the source of further marginalization of the disempowered due to embedded, locked-in forms of algorithmic bias – these are just two illustrative examples of alternative dominant imaginaries to “Racing.” While each of these narratives have security-relevant components, and each of these narratives received some level of “success” via penetration into policymakers’ consciousnesses due to each having generally well-regarded spokespeople from other expert roles and throughout the academic sector, they ultimately did not receive as high degree of institutionalization across U.S. policymaking, and one could argue that the specific role of hybrid experts over the time period in question is one of the determining reasons for that.

Finally, there is the possibility that as U.S.–China competition intensified overall, the securitization of technology was inevitable, as it would spur a military–industrial complex resulting in locked-in, self-perpetuating tech-oriented strategies. One can examine several new accounts of the Obama–Trump–Biden administration’s transformation of U.S.–China strategy to see how technology factors into existing narratives. Shambaugh has described, for instance, how almost the exact same timeline examined in this book coincides with the rise of a “Counter-China Coalition” in U.S. politics, as the first Trump administration overtly took on the mantle of “strategic competition” with China and effected a major shift for U.S. foreign policy (Shambaugh Reference Shambaugh and Shambaugh2025, 129). Similarly, McCourt has outlined the overall arc of the U.S.’ foreign policy transformation and how consensus shifted about the strategically optimal way to deal with China in the same period of time (McCourt Reference McCourt2024). While Shambaugh’s account points to the role of Trump’s brain trust, such as the role played by individuals like Matt Pottinger, most of whom instead of being “hybrid” have held prior roles within the U.S. national security or foreign policy establishment, what is less immediately obvious from the general shift from engagement to competition between 2015 and 2018 is how AI turned from a footnote of the story of military modernization and surveillance to a centerpiece of U.S.–China competition and a singular eponymous an widely recognized “race” in its own right by the end of the Biden administration. McCourt’s work points to the evolution of the “Third Offset” strategy within the DoD as context for the increasing emphasis on viewing China’s military modernization – especially in technological terms – as a threat, referencing a RAND report on the “Third Offset” that largely credits former Deputy Defense Secretary Robert Work and other similar figures for channeling increasing DoD energy toward the problem of technological competition as a whole (Gentile et al. Reference Gentile, Shurkin, Evans, Grisé, Hvizda and Jensen2021). As these accounts argue, there was well-founded interest in technological competition with China and military modernization in several siloed sources throughout the DoD and government more broadly, but these sources do not convincingly trace how specifically the civilian AI sector became such a focal point of U.S.–China competition, even as far as serving as an indicator of the health or success of either nation in the competition itself. Schmidt remains key to that story, as sources above show that he began his engagement with Washington on emerging technologies as early as 2012, temporally predating the overtly competitive shift in U.S.–China relations (Conger and Metz Reference Conger and Metz2020).

Ultimately, the hybrid actors highlighted within this article themselves remain an important component to how narratives like “Racing” grew in profile among elite policymakers in the U.S. Even given some broader understanding of the conditions of competition between the U.S. and China, just as an example, it is not a given that the USG would have engaged in the exact process of onshoring, de-coupling, and the empowerment of the private sector in the AI field had specific epistemic influencers not steered the USG in that direction.