It’s Time for AI Policy to Get Serious About the AI Adoption Gap

“If you build it, they will come” may have worked for a fictional farmer building a field of dreams in 1980s Iowa. But as a strategy for successful technology diffusion, it has failed more often than it has succeeded. Consider the early automobile industry. Decades of engineering brilliance gave Europe an initial lead as French and German carmakers replaced tillers with steering wheels and surface-fed carburetors with floats and spray nozzles. Yet their ingenuity only found a niche market, even as it also demonstrated what the modern car could become. With limited road infrastructure and companies like Daimler producing less than one vehicle per worker, high costs and low utility confined the European car’s appeal to the wealthy few. Soon enough, Europe ceded ground to the United States, where the moving assembly line, efficiency wages, and roadway construction famously generated both the supply-push and demand-pull needed to jumpstart a mass market.

When it comes to AI, Washington and Silicon Valley have embraced a daring “build-first” strategy, and the early results have been impressive. The Trump administration deserves enormous credit for policies focused on easing supply-side bottlenecks in chips, energy, infrastructure, and spectrum, which will stunt the United States’ ability to produce AI technology without solutions. Similarly, with the bulk of its deal volume flowing into developing and fueling the best possible general-purpose models, Silicon Valley and its massive investments in supply have given the United States an initial, and potentially vital, edge in AI innovation. They have also opened the world’s eyes to AI’s transformative potential, just as European ingenuity did for the car so many years ago.

But from a market perspective, this strategy will only succeed if the rest of the economy generates sufficient “pull” quickly enough to turn that supply into lasting impact. And from a policy perspective, the success of that strategy is essential to keeping U.S. industry and the defense industrial base competitive as global rivals ramp up their own AI capabilities.

With markets paying more attention to the demand problem, it’s time for Washington to take the same leap. U.S. AI policy needs bolder ways to boost the rate of adoption and diffusion of AI applications, even as it also continues to execute on expanding supply. Both are necessary to sustain U.S. leadership—not just in AI, but across the many economic and national security arenas where success will depend on its successful implementation.

In a time of fiscal constraint, new federal spending cannot be the only solution. Creative policy levers will also be critical—and early career workers are an obvious place to push.

Adoption Barriers Warrant a Bolder Response

Some might ask why policy matters—as AI matures, won’t markets generate as much demand as its capabilities warrant? While gaps in demand can be difficult to quantify, the United States has enough experience to know that market mechanisms will undershoot its full potential without policy support. Spillovers, asymmetries, and heterogeneity in use have long plagued the diffusion of simpler and less profound innovations. The risks of under-adoption—not to mention its costs—are much greater with technologies like AI where the strategic benefits, gaps in know-how, and complexity of integration are more pronounced.

Theory aside, data already paints a picture of a marketplace facing adoption challenges that will only grow more significant once the low-hanging fruit of AI integration is gone. At a macro level, there is substantial evidence that recent economic growth and stock market gains have been largely, if not almost entirely, driven by AI data center investment. While demand for that compute power no doubt exists, there is comparatively little evidence that use cases in vertical industries are emerging fast enough to sustain the spend. Years after the United States democratized access to foundational models, census surveys continue to show low levels of enterprise adoption and a persistent gap in adoption rates between small and large firms that began to narrow only as large firm rates started to fall. Further, at the firm level where AI integration will succeed or fail, the challenging mechanics of implementation are exposing the limits of the “as-a-service” model that worked so well for past digital applications. While platforms can make it easier to build enterprise AI capabilities, integrated solutions outside of software development are rarely turnkey and only partially scalable. In industrial enterprises, for example, meaningful AI adoption often requires heavily customized if not completely bespoke combinations of software, hardware, and organizational changes.

If this state of play persists, the United States will find itself in a precarious position, making it imperative for its economic and national security to enable widespread adoption by the private sector. That is especially so given the People’s Republic of China (PRC)’s ability and resolve to use its autocratic system to solve the demand side of the equation. More than half of China’s GDP growth may already be attributable to industrial digitalization, and China has already deployed more than 30,000 private networks across its top two operators alone as it lays the groundwork for future progress. More anecdotally, recent advancements in Chinese manufacturing have reportedly left U.S. executives reeling, and this past August, China doubled down with a 90 percent national AI adoption target by 2030. Against this backdrop, debates over which country’s frontier models are better may wind up missing the point. Because the benefits of transformative innovation often compound—and given the clear economic and national security implications of a false start—what may matter more is how quickly each economy puts its models to work in vertical applications.

AI Adoption Accelerators

The Trump administration’s AI Action Plan recognizes slow adoption as a bottleneck and proposes useful tools like sandboxes, collaboration on standards, regulatory relief, and full-stack export promotion. As policymakers identify ways to double down, there is another obvious lever to pull: deploying young workers that a short-sighted private sector has undervalued at this moment of rapid change.

A shifting labor landscape has led to rising rates of unemployment and underemployment among younger workers, regardless of whether they possess a college degree. The causes are complex, but uncertainty regarding AI appears to be part of the story. While percentage increases thus far have been modest, the delta represents tens of thousands of struggling U.S. workers in their prime—and a potential army of AI adoption facilitators ready to roll up their sleeves to meet the diffusion challenge. These are workers who are more likely to be digital natives and sophisticated in their ability to apply off-the-shelf technological capabilities. The United States should find ways to unleash them.

The Trump administration’s efforts to reform higher education present precisely such an opportunity. The administration has drawn criticism for advancing policy goals through concerted action against colleges and universities. Even so, there should be bipartisan consensus that the federal government need not blindly provide billions in funding to educational institutions without regard for the results. There is room for the federal government to ensure that students and the broader public receive a better return on investment without micromanaging the scientific process. In that context, the federal government should encourage colleges and universities to assume a greater role in ensuring that strategic technologies—first and foremost, AI—have home bases for deployment in local communities.

The current administration should incentivize colleges and universities to establish “AI Adoption Accelerators” designed to meet the urgency of national diffusion goals. These centers should forge strong partnerships with nearby businesses to understand their potential for AI integration and implementation challenges. Coursework covering the practical basics of architecting and deploying AI projects should be developed in response, tailored to local needs, barriers, and gaps in capability. Once trained up, AI adoption facilitators should be deployed to advance AI integration at willing firms through course projects, internships, or employment. Sponsoring institutions should explore industry partnerships to create incentives for local businesses to hire trained adoption facilitators, such as portable discounts on subscriptions and equipment, and cost-sharing for stipends that fund intern-to-hire pipelines.

AI Adoption Accelerators should focus shamelessly on the yeoman’s work of creating workable solutions for specific firms. In that regard, they should be designed to complement existing initiatives like AI Centers of Excellence, which focus more on regulatory experimentation and frontier research and education. Likewise, accelerators should reject costly and drawn-out delivery models better fit to proving out risky technologies than integrating ones already available in the marketplace. Finally, to maximize their impact on unemployed workers and local communities, they should be designed to benefit college and noncollege graduates alike.

Incentivizing Action

To spur action, the administration should reward institutions with increased research overhead reimbursements, provided that they sponsor or participate in accelerators meeting reasonable benchmarks.

In contrast to most programs that interface with higher education, AI Adoption Accelerators would focus on integrating already commercialized technologies and achieving near-term results. In these differences, however, lies opportunity. The Trump administration has aggressively sought to reduce indirect or overhead cost payments, which account for more than 40 percent of direct research funding according to one estimate. But it has also indicated a willingness to increase reimbursement rates for institutions that adopt certain policies. To speed up AI adoption, the administration should use this lever to incentivize AI Adoption Accelerators. While accelerators may not target any individual federally funded project, they would enhance the overall quality and relevance of university research by expanding U.S. capacity to absorb innovation and aligning research agendas with real-world demand. That makes accelerators not just useful for advancing national AI diffusion goals—but much more defensible as an overhead expense compared to most line items on the university budget.

The United States should take bolder action to close the AI adoption gap. Early career workers will be on the frontlines of the solution—and colleges should work with industry to deploy them more effectively.

Shiva Goel is an adjunct fellow (non-resident) with the Strategic Technologies Program at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Matt Pearl is the director of the Strategic Technologies Program at CSIS.