The premise that animated Washington’s approach to artificial intelligence for the better part of three years — the notion that the American technology industry and the American defense establishment shared a unified vision for the development, deployment, and governance of AI systems — was always more aspiration than reality, more talking point than strategic alignment. That premise is now collapsing under the weight of its own contradictions, and the consequences of its collapse will shape not only the trajectory of AI development in the United States but the global balance of technological power for decades to come.
The fracture lines are multiple and deepening. The most visible runs through the question of military application. The Department of Defense, through its Chief Digital and Artificial Intelligence Office, has spent the past eighteen months building what it describes as a sovereign AI infrastructure — classified compute clusters, secure model training pipelines, and deployment frameworks designed to integrate frontier AI capabilities into the full spectrum of military operations, from logistics optimization to target identification to autonomous weapons systems that operate, in the Pentagon’s carefully chosen terminology, with “human oversight” rather than “human control.” The distinction between oversight and control is not semantic; it is the difference between a human who must approve every action and a human who monitors a system that acts on its own unless overridden. The defense establishment wants the latter. Several of the companies building frontier models are unwilling to provide it.
Anthhropic has published its responsible scaling policy and drawn explicit boundaries around military applications of its models. OpenAI, which modified its usage policies in early 2024 to permit certain defense and intelligence applications, has nevertheless declined specific contracts whose terms it deemed incompatible with its safety commitments — decisions that have generated considerable frustration among Pentagon procurement officials who regard the company’s selectivity as both commercially naive and strategically dangerous. Google, whose workforce revolt over Project Maven in 2018 remains the defining episode of the industry’s internal debate over military AI, has quietly rebuilt its defense business through contracts that it describes as defensive and analytical rather than offensive and kinetic, a distinction that its own engineers have questioned in internal forums.
The export control debate has opened a second front. The Commerce Department’s semiconductor export restrictions, designed to deny China access to the advanced chips necessary for training frontier AI models, have been embraced by the defense establishment as essential to maintaining American technological superiority and resisted by the technology industry as both commercially damaging and strategically counterproductive. Nvidia, whose data center GPU revenue depends substantially on the Chinese market, has lobbied with the intensity and resources of a company whose core business model is under regulatory threat — which it is. The company’s argument — that export controls will accelerate China’s development of domestic chip alternatives without meaningfully delaying its AI progress — is not without merit, but it is also not without self-interest, and the difficulty of disentangling the two has made the debate nearly impossible to resolve on its technical merits.
The regulatory question — who governs frontier AI development, and by what authority, and with what enforcement mechanisms — has produced a third axis of divergence that cuts across the first two. The defense establishment, which wants unrestricted access to the most capable models for national security purposes, opposes regulatory frameworks that would constrain model development or deployment. The technology companies, which want regulatory certainty and protection from liability, support frameworks that are predictable and navigable but oppose those that would grant government agencies direct oversight of model training processes or mandatory pre-deployment safety evaluations. The result is a policy landscape in which every major stakeholder opposes some essential element of every proposed regulatory framework, and the legislative process has produced a series of draft bills that are introduced with fanfare and abandoned with quiet acknowledgment that the interest-group dynamics are, at present, unresolvable.
The international dimension of the fracture is perhaps its most consequential aspect. The United States has presented itself to its allies as the leader of a democratic technology bloc that will develop AI in accordance with shared values and deploy it in defense of the rules-based international order. This narrative requires, at minimum, the appearance of domestic consensus — a technology industry and a national security establishment that are pulling in the same direction, even if they disagree on speed and method. The visible divergence between Silicon Valley and Washington undermines that narrative in ways that America’s strategic competitors have been quick to exploit. Chinese state media has covered the American AI governance debate with evident satisfaction, characterizing it as evidence that democratic systems are structurally incapable of marshaling the centralized coordination that advanced technology development requires.
The lobbying expenditures tell their own story. The five largest AI companies — Alphabet, Microsoft, Amazon, Meta, and Anthropic — collectively spent approximately two hundred and seventy million dollars on federal lobbying in 2025, a figure that represents a threefold increase over 2023 and that does not include the substantially larger sums directed toward state-level lobbying, think-tank funding, and the cultivation of academic voices sympathetic to industry positions. The defense industry’s lobbying apparatus, which dwarfs the technology sector’s in both expenditure and institutional experience, has been deployed with equal vigor in the opposite direction — pressing for procurement authorities, classification frameworks, and legal protections that would enable the rapid integration of commercial AI into military systems without the consent frameworks that the technology companies insist upon.
The resolution of these tensions will not come through consensus, because the underlying interests are genuinely incompatible. A company that builds its brand on responsible AI development cannot simultaneously supply autonomous weapons systems without constraint. A defense establishment that regards AI superiority as essential to national survival cannot accept a commercial sector’s veto over military applications. A regulatory framework that satisfies the industry’s desire for predictability will not satisfy the government’s desire for flexibility, and vice versa. What will emerge, instead, is a set of pragmatic accommodations — some public, some classified, some deliberately ambiguous — that allow each faction to claim alignment with its stated principles while making the compromises necessary to function within a system that no single actor controls.
The myth of the united front served a purpose while it lasted. It projected confidence to allies, deterrence to adversaries, and coherence to a public that was and remains largely uncertain about what artificial intelligence will mean for their lives and livelihoods. Its passing does not make the United States weaker, necessarily, but it does make the country’s AI trajectory less predictable, less coordinated, and more susceptible to the friction and delay that characterize democratic governance at its most contentious. Whether that friction is a strength — the democratic system’s immune response to the concentration of transformative power in too few hands — or a weakness — the paralysis that allows more decisive competitors to seize the advantage — is the question that the next decade of American technology policy will answer.