The key is to draw a line between AI architecture and AI applications.
The key is to draw a line between AI architecture and AI applications.
A recent article in Politico magazine highlights growing tension in the Republican coalition over artificial intelligence. A White House executive order issued in late 2025 would lay the groundwork for significant federal preemption of state and local regulations of AI, but other Republicans are pushing for local control of AI policy.
While some (especially on the “tech right”) prefer a hands-off approach to AI regulation, many populists insist on the need to regulate AI, especially on the grounds of protecting families and jobs. This policy divide risks creating a political fissure in the GOP, and navigating this divide will be critical in the years ahead.
The White House and its allies should view the tech-skepticism of state and local Republican leaders less as a political inconvenience and more as a blinking red light. State-based efforts to confront illegal immigration were a harbinger of Donald Trump’s populist campaign for the presidency a decade ago, and AI could become a similar flashpoint in the late 2020s. Mass illegal immigration stoked such popular concern in part because it fed a sense of chaos; if conventional politicians were unable to maintain border controls, then it was time to turn to populist outsiders.
AI could replay that dynamic. If Washington seems helpless in the face of massively disruptive technology (and, worse, if Washington actively tries to stop states and cities from addressing that disruption), that policy vacuum could provoke another populist cascade. A GOP that proves deaf to the unrest of the “radical middle” on AI could risk being outmaneuvered by Democrats, who would claim the populist mantle by pledging to “take on” AI. Vermont Senator Bernie Sanders is already mounting a test run of this left-populist gambit.
These political pressures should not obscure the fact there is a major national interest in being at the forefront of AI, which has profound stakes for technological innovation, national defense, and even reindustrialization. Because the United States cannot afford to cede the AI technological frontier to the Chinese Communist Party or to other strategic competitors, American policymakers need to secure public buy-in for a regulatory infrastructure that encourages AI innovation.
One possible détente on AI policy would be to distinguish between architecture and application. Federal preemption is perhaps on its strongest grounds in terms of the underlying architecture of AI (training regimes, model structure, and so forth). A patchwork of local regulations regarding the inner architecture of AI models could imperil the ability of that sector to grow and adapt. If every state mandated its own “safety” or “diversity, equity, and inclusion” audit every time an AI company tweaked a model or algorithm, the fecundity of regulatory burdens could slow innovation to a crawl.
There is a case for the federal government to lay out certain rules of the road for AI architecture at the national level and establish safeguards to guarantee human control of AI. A draft executive order that circulated in December pointed in this direction. Emphasizing “human flourishing” as part of AI policy, it laid out a federal standard as a “floor” in order to ensure some basic safeguards and regulatory consistency across the country, but it also recognized space for states to innovate above that policy baseline.
The incentives for a uniform national policy are not nearly so strong when it comes to the application of various AI technologies. Different state-level regulations of applications need not imperil the innovative potential of AI development. For instance, requiring that medical providers inform a patient whether they are talking to a bot or a human being in no way threatens the American competitive edge on AI. Likewise, restrictions on the use of AI in public school classrooms or bans on the use of AI to create sexually explicit deepfakes would not meaningfully impede innovation for AI infrastructures. The future of human civilization will not turn on who fabricates the best deepfake porn.
Multiple efforts to move AI regulation through Congress have so far faltered, though California Representative Jay Obernolte is currently working on a new preemption bill. These legislative struggles can in part be attributed to the real political headwinds facing any sweeping AI agenda at the federal level. A more targeted bill — ensuring a national framework for how to regulate models while also allowing for a more decentralized approach to applications — might be better able to get legislative traction.
The physical infrastructure of AI — especially data centers — has been another area of contention. Here, too, some compromise is possible between the federal government and state and local actors. A robust federal pro-energy agenda could address understandable fears that a proliferation of electricity-binging data centers will drive up power costs for local communities. Particularly in areas where water supplies are already under pressure, there are also considerable concerns about the environmental impact of large data centers, which use water as a critical cooling resource. Disregard for concerns from people in the affected communities would likely feed political backlash.
For the United States to remain at the forefront of digital innovation, it will need a sustainable public consensus for tech policy. Avoiding regulatory sclerosis will mean fusing pro-growth sentiments with attention to the flourishing of the person, the family, and local communities. This will be especially important for a realigned Republican Party. Party leaders must think through a tech policy that both advances innovation and reinforces the underlying fabric of American life.