U.S. Foreign Policy: Power in the Age of AI

While Artificial Intelligence (AI) has been at the heart of American strategic concerns for over a decade, the Trump administration now considers itself in an "existential race with China" concerning AI (to reuse the words of Sriram Krishnan, Senior Advisor for AI at the White House), with an explicit desire to ensure their dominance over future systems. The White House is thus deploying a far more ambitious agenda than its predecessors, based on an unprecedented collusion between private Tech giants and the Trump Administration. Concurrently, the China-U.S. race for technological dominance is shaping the ongoing reconfiguration of the world order, accelerated by the reduction of the American military footprint outside the Western Hemisphere.

If not a grand strategy, the United States is forging a new paradigm of competition in this multipolar context, operating outside traditional international law.

In the United States, several strategic directions are currently emerging to maintain dominance in this area, which can be declined at three levels: fundamental research and the development of "frontier models"; the widespread deployment of AI across the U.S. economy, in anticipation of global competition with China (which is implementing an AI development plan supported by its national civil-military fusion strategy since 2017); and, more specifically, the adoption of AI within the American armed forces across all their functions.

AI Fully Integrated into the Conception of 21st-Century American Power

The Administration’s AI agenda was accelerated following the unexpected progress of the Chinese company DeepSeek, described by some as the next "Sputnik moment" of the United States. China has now made AI the pillar of their economic and industrial planning, with the aim of becoming the world leader in the domain by 2030. Simultaneously, the Chinese People's Liberation Army (PLA) has, now for over a decade, adopted a military doctrine of "intelligentization" of warfare integrating AI, big data, and quantum computing. The national strategy of "civil-military fusion" aims to leverage civil sector innovation for the benefit of defense. The credibility of the American posture towards China is thus now largely played out in the field of AI.

In this context of a zero-sum game with China and the advent of the "America First" agenda, the priorities of the United States in this matter are now clear: consolidate their sovereignty, secure energy resources and value chains, obtain first-mover status in strategic technological segments, and weaken the competitors’ capacity to rival them. Ultimately, the goal is to disseminate American "chips and models" in civil and military domains to allies and competitors of Washington alike. The emphasis on the diffusion of open-source models actually reflects this desire to impose American standards, including models conforming to the Trump Administration's ideological line (devoid of diversity, inclusion, climate, or anti-disinformation topics).

A recent paper from the Stimson Center asserts that the United States must aspire to become a new "nation-company hybrid" imposing its "physical and digital substrate" in order to reshape "hierarchy of great powers" However, it raises doubts about the sustainability of growth of computing infrastructures, in terms of energy and water supply.

From the standpoint of military posture, retrenchment strategists are also reliant on AI in changing the United States’ strategic calculations by reducing its footprint while maintaining substantial strategic depth as well as the ubiquity of its striking capabilities. Following the ideas around "hyperwar" theorized by John Allen, the establishment of a non-human, ultra-responsive force that would ideally deter adversaries while reducing points of vulnerability - for example, troops within striking distance in the Middle East.

In this perspective, where the imperative of maintaining a liberal international order is fading, the United States is moving towards a system where alliances and partnerships would be primarily dictated by infrastructure and value chain coalitions built around modern choke points, including semi-conductors, data centers, energy, and critical minerals. Conditioning aid to Ukraine on the signing of a "deal" on critical minerals, or the fixation on Greenland are symptoms of this reorientation of Washington's priorities towards new geoeconomic considerations. Equally revealing, the Development Finance Corporation (DFC), a tool created during Trump’s first term to mobilize private funds towards emerging countries, could be placed exclusively in support of a policy to secure value chains and promote American exports. Strikingly, the question of support for Taiwan also tends to be reduced to that of securing access to semi-conductors, as revealed by the proposal made by U.S. Secretary of Commerce, Lutnick, to relocate 50% of chip production to the United States in exchange for lifting tariffs on the island.

The "Hamiltonian" Shift: An Unprecedented Collusion Between Public and Private Sectors

To implement this strategy of geoeconomic dominance, the Trump administration is betting on an unprecedented collusion between the government and major private actors. In exchange for a radical deregulation program, unprecedented access to public contracts, and the promotion of their interests abroad,Tech companies have now sworn allegiance to the White House, promising to participate in its national reindustrialization project through massive investment projects.

"America’s AI Action Plan" notably reflects this strategy of laissez-faire and complementarity between public and private powers, summarized by the phrase "let them cook" coined by David Sacks, with whom Donald Trump has entrusted AI and cryptocurrencies strategy. Behind this new public-private pact lies a "Hamiltonian" strategic shift, in the way the American Administration weaponizes these tech companies in the service of its geopolitical project. The concept, developed by political scientist Walter Russell Mead in reference to the country's first Secretary of the Treasury, Alexander Hamilton, denotes, in this case, the blurring of the lines between the government and businesses’ interests.

Besides the protection offered by the White House against their foreign competitors and European regulations, large Tech companies can now hope to win some of the world's most lucrative public contracts, thanks to their increasing integration into government functions. The example of Palantir, which recently signed a $10 billion contract with the Department of War, promising to restructure all internal digital processes using AI, is symptomatic. In return, Tech companies seem to accept a certain degree of interventionism, as shown by the recent U.S. government’s acquisition of a 10% equity stake in Intel.

The rapprochement of tech and the state also reflects a profound cultural change within Silicon Valley, historically libertarian and as such opposed to any direct association with the government. To gauge the extent of this recent change, it is necessary to recall that Google's collaboration on Project Maven (a project dating back to 2017, using AI to process drone video archives) initiated by the Pentagon aroused significant enough opposition within the company in 2018 that Google had to halt its collaboration with the Department of War. The current turnaround marks the ideological victory of figures like Peter Thiel and Alex Karp, who have long been a minority within the Tech sphere. For years these individuals have been calling for the sector to adopt a "patriotic shift" to contribute to U.S. defense, and see the current evolution born out of the union between the research and military worlds as Silicon Valley simply going back to its roots. This new alliance finds its most complete symbolic realization in the creation, as controversial as it is unprecedented, of a reservist program composed of Tech executives named "Detachment 201".

Jensen Huang (CEO of Nvidia) and David Sacks (Special Advisor for AI and Crypto) believe that export restrictions would have a gravely deleterious effect on America's ability to dominate the market. While their influence has allowed a certain openness to suspending the White House’s export controls, Nvidia currently remains excluded from the Chinese market.

The Temptation of the "Sprint" towards AGI and the Problem of Adoption at Scale

This dominance of private actors is also reflected in the race towards Artificial General Intelligence (AGI) or Artificial Superintelligence (ASI). Many large tech companies are now strongly driven by what could be described as a "sprint" strategy, betting that first-mover status in the AGI race would offer the American economy a clear, decisive advantage over China. Following this logic, investment in the vertical development of superintelligences should be the main objective, a near-panacea that could, subsequently, address all other challenges. The Meta Superintelligence Labs entrusted by Mark Zuckerberg to Alexandr Wang, presented as a kind of "Manhattan Project" of AGI, are a clear embodiment of this gamble.

The advent of AGI and the resulting emergence of a new strategic stability are now conceived as a problem in itself. The most optimistic defenders of the "sprint" strategy estimate that once AGI is achieved, the exponential nature of the advances would offer sufficient deterrent power vis-à-vis China to end strategic competition. A segment of the tech industry even considers the contours of a new form of strategic stability based on AGI replacing or complementing the current model built on nuclear deterrence, owing to the new thresholds of mass destruction that such technology could create. According to this view, the AGI era must be accompanied by new geopolitical concepts inspired by Mutually Assured Destruction (MAD), for a security architecture based on the concept of "mutual assured AI malfunction" (MAIM): Given acts of espionage, sabotage, cyber, kinetic attacks, etc., MAIM must thus lead States to restrict their willingness to preventively or preemptively destroy the projects of their competitors and could be accompanied by an export control regime and agreements on information security and sovereignty.

Facing the temptation of the sprint, more and more voices are raising awareness on the cost of a strategy aimed at supplanting Beijing by any means necessary in the frontier models race, in terms of investments, infrastructure, and energy. Proponents of the so-called "marathon" approach emphasize the importance of the horizontal adoption of AI across society. However, it is precisely in terms of adoption at scale that the United States risks lagging behind its Chinese competitor, which invests more in application deployment in different economic sectors rather than in its frontier models. Unable to compete on the latter, Beijing has decided to bet everything on forced-march diffusion. Yet, a recent and much-discussed MIT report highlights that the enormous challenge in AI adoption remains the transition from pilot projects to successful integration into organizations, stressing that market scale is the main source of value creation. However, adoption is stagnating in American companies.

From the perspective of American strategy, the allocation of resources in a sprint or marathon strategy may, therefore, represent a real long-term dilemma.

The Military and AI: The Organizational Challenge

These strategic issues take on a particular dimension in the military domain, where the integration of AI into the conduct of war has become a growing priority for Administrations since Trump’s first term. While the Pentagon’s early strategies in the field of emerging technologies, including the one published in 2023, remained fairly conservative regarding AI, the war in Ukraine has undeniably spurred a new dynamic, acting as a first testing ground for AI warfare ("AI war lab").

Despite the real impact of AI on the conduct of war and the rise of a new military-technological complex, the American Department of War nevertheless faces the same difficulty of adoption at scale as the rest of the economy. Out of a defense budget of nearly a trillion US dollars, the share allocated to AI is 13.4 billion for fiscal year 2026. While there are over 800 projects involving AI at the Pentagon, the department remains at the very beginning of the transformation of its military model. For now, the main projects involving AI, many of which constitute a reinforcement of existing capabilities in the domain of planning (AI must ultimately allow for the creation of a battlefield management system enabling total operational vision through the fusion of data from all sensors in real-time - Combined Joint All Domain Command & Control, CJADC-2), targeting (battlefield management systems already exist at more limited levels, but in the domain of targeting, AI can allow armed forces to identify targets more quickly and precisely. Thus, Project Maven is one of the Pentagon’s biggest AI projects, aiming to process drone data in a logic of preemption of adversary targets) or cyber warfare rather than a real disruption of the nature of war. In terms of "force structure and employment," the development of autonomous drone swarms, embodied in the Replicator Project intended to produce large quantities of autonomous weapons or drone systems at lower cost, remains one of the most anticipated developments.

The new administration, particularly Pete Hegseth, has spurred an acceleration of the application of these solutions at scale, again betting on a fruitful synergy between the public and private sectors. The current permeability reaches such levels that one can truly speak of the advent of a new military-technological complex emanating from Silicon Valley, encouraged by private actors like Steve Feinberg, former CEO of Cerberus Capital who became Deputy Secretary of Defense. Conversely, US Vice President J.D. Vance worked at Mirthil Capital, a venture capital firm of Peter Thiel. Convinced that the United States can only "rebuild its strategic advantage" in the defense domain if the Department of War becomes a leader in AI adoption, Feinberg places priority on the rapid adoption of commercial AGI solutions: "Rebuilding our military advantage demands that the [DOD] become an [AI-first] enterprise; one that rapidly adopts cutting-edge commercial Al technologies, exploits data at scale to generate operational advantage, and leads the discovery of new ways to fight and win."

The Pentagon's openness to adopting commercial Large Language Models (LLM) models developed by large tech companies has been one of the first steps in this direction

The fact that these contracts amount to hundreds of millions, even billions of dollars, shows a willingness to create private solutions at scale.

A flagship project cited is the Thunderforgeinitiative, valued at $3 billion, associating agentic AI proposed by ScaleAI with Anduril systems to facilitate decision-making in operational theaters, planned for INDOPACOMM and EUCOMM.

The numerous bureaucratic reshuffles in recent months, however, underline the extent to which adapting an administrative behemoth to these imperatives of change remains a significant challenge and a problem of which the current government is well aware. Uncertainty still hangs over the Pentagon's true global strategy for adapting to the reality of AI. Indeed, the fear shared among observers is that institutions will not manage to offer a response equal to the changes required.

One of the main administrative, but also cultural, hurdles within the Pentagon remains the issue of acquisitions, which is key to transforming pilot projects into real capabilities, as shown by the preponderance of nuclear deterrence or naval platforms in the defense budget's OBBB [One Big Beautiful Bill Act, budget bill adopted in July 2025].

However, in the area of equipment, processes are extremely long, to the point that the latency period between proof of concept/innovation/research and development and procurement is often nicknamed the "valley of death". The slowness of the defense-acquisition system favors established companies over startups like Anduril, which base their model on taking risks upfront. Private-sector figures such as Shyam Sankar, the CTO of Palantir, have made acquisition reform a central focus, hoping to spark a true "revolution" replacing old industrial players like Lockheed, BAE, or RTX with new startups adapted to 21st-century warfare. Thus, here again, the question of resource allocation between traditional platforms and new capabilities developed by startups poses a real strategic dilemma that remains yet unresolved, despite a clear desire to upgrade scale.

Although there is not yet a global vision articulated by the new Administration, a form of ‘AI-led great power doctrine’ aligned with the "America First" agenda seems indeed to be emerging in Washington, creating a link between tech circles and the realist conception of foreign policy that dominates within the Administration. Nevertheless, the speed of developments in the field risks catching off guard any attempts at elaborating a truly long-term strategy. In this new Hamiltonian marriage between Silicon Valley and Trumpism, the interests of the private sector risk taking precedence over those of the government, as evidenced by the colossal investment of both energy and infrastructure resources in the race towards AGI. It is also noted that the race for AI adoption at scale remains a considerable challenge, whose economic and geostrategic winners remain unknown. However, lacking capacities comparable to those of the United States and China, in this race Europe must undoubtedly identify strategic sectors and play the diversification card in order to avoid finding itself in a state of irreversible vassalization.

Copyright image : Roy Rochlin / GETTY IMAGES NORTH AMERICA / Getty Images via AFP
Michael Kratsios (President's Science Advisor), Donald Trump, and David Sacks (President's New Technologies Advisor) during the presentation of the "Winning The AI Race" plan on July 23, 2025, in Washington.