Who Will Make Money on AI?

Introduction

The development of breakthrough dual-use technologies has historically involved cooperation and interaction between academia, the government, and the private sector. Researchers in academia often provide the initial seeds for new innovations, governments can marshal resources and direct research programs toward national security priorities, and as technologies mature and develop, the private sector can scale new capabilities, operating within regulatory frameworks that account for dual-use risks identified by the government. For example, the initial ideas for nuclear energy emerged from European academics. The U.S. government then launched the massive Manhattan Project to develop nuclear weapons capabilities in the context of the Second World War. A decade later, the Atomic Energy Act of 1954 paved the way for the private sector to deploy civilian nuclear power plants.¹

The development of artificial intelligence (AI), however, has followed a distinct trajectory. The private sector is playing a leading role in advancing the frontier of this new dual-use technology even in its early stages, as novel capabilities and associated risks are still being discovered. Because AI is developing within commercial firms rather than government programs, market forces are likely to play a significant role in shaping how these capabilities develop (as companies direct research resources to align with commercial incentives) and diffuse (as companies choose which markets to prioritize). There is an urgent need to better understand how market structures and commercial pressures will influence the evolution of AI technologies, and what these dynamics mean for AI risks and U.S. national security interests.

Today, there are several wide-ranging debates on future scenarios for AI commercialization and market development. The unique characteristics of AI as a product, including significant uncertainties on what product fits will ultimately be most important for a powerful general-purpose technology, make it difficult to draw clear historical lessons from other industries. Will companies developing frontier AI models identify sustainable business models? Will enterprise AI users find productive use cases to justify rapid uptake of AI tools? What layers of the AI stack will be most profitable, and what layers may become commoditized? Will today’s dominant incumbent companies across the AI stack maintain their positions, or will new upstarts encroach on their markets?

The starting point of this paper is that the answers to these and related questions will have important implications for U.S. national security interests. To date, however, experts focused on AI and national security have tended to pay more attention to technical AI developments, such as the trajectories and timelines to reach different AI capability benchmarks, than to companies’ commercialization strategies and market incentives. To the extent that commercial dynamics are considered, it has been primarily around how competition to release new models may give an incentive to cut corners on safety.² While important, this is only one channel through which AI commercialization may impact national security, leading to an important gap in understanding. This paper does not seek to provide definitive answers to the big outstanding questions on the evolution of the market for AI, but rather to better understand the contours of such debates to identify what is at stake for U.S. national security and assess how policymakers and private actors might better align commercial incentives with national security interests.

The AI Economy

AI has evolved from an esoteric research topic to an emergent general-purpose technology that could have transformative effects on the U.S. and global economies. Researchers are beginning to identify the impact of AI in the latest aggregate economic data on productivity and employment in the United States, although such analyses remain contested.³ Expectations of the future economic impact of powerful AI vary widely. Some leaders at frontier labs have suggested that AI will drive an economic revolution in the near term, with growth in gross domestic product accelerating to 10–20 percent a year.⁴ Economists studying AI tend to be more cautious, highlighting that even as AI capabilities improve exponentially, economic growth will still be limited by physical constraints in the real world—including the ability to scale chips and data centers—and that historical precedents suggest technology diffusion and adoption is a slow and staggered process.⁵ Some economists and AI experts warn powerful AI could spark mass unemployment, as AI becomes a cheaper and more capable substitute for nearly all knowledge workers.⁶ Other economists, however, are more sanguine, again often pointing to the history of earlier labor-saving innovations, which served as a complement rather than a substitute for labor and induced greater aggregate demand.⁷

Meanwhile, the American public is increasingly anxious about AI. One recent poll found that 57 percent of Americans believed the risks of AI outweigh its benefits, with only 34 percent holding the opposite opinion.¹² Seventy percent of Americans think advancements in AI will lead to a decrease in job opportunities, while only 7 percent expect an increase.¹³ Such perceptions are sparking a backlash against the prospect of transformative AI, epitomized in the fights state and local governments are launching against new data center construction.¹⁴ Whether these public pressures ultimately translate into policy shifts to restrict AI development and deployment, however, remains to be seen; Americans have also had predominantly negative views of social media companies for several years, but this has not resulted in substantial regulatory changes to date.¹⁵

Against this backdrop, companies across the AI stack are seeking to establish or cement profitable business strategies in rapidly evolving commercial AI markets. Individual companies’ commercial decisions are shaped by broader macroeconomic trends in the AI market, and likewise the aggregate economic impact of AI will be determined by the interaction of individual companies’ decisions. Certain segments of the AI tech stack, such as semiconductor companies, have more clarity on commercialization and have already reaped substantial revenues from the AI boom. However, fundamental questions remain about the path to commercialization and profitability for AI models and applications. A central question will be to what extent AI companies are incentivized to design their products for enterprise, consumer, government, or internal (e.g., indirectly monetizing AI capabilities by better targeting ads) uses. How companies balance between these market segments, and which grow faster than others, will drive commercial strategy decisions that will shape incentives for investing in AI safety, potentially leading to knock-on effects for U.S. national security interests.

Companies’ commercialization strategies will also impact market structure and concentration across the key layers of the AI stack: AI infrastructure, including chips and data centers; foundation models; and AI applications. Throughout the AI stack, companies are seeking to establish market positions that provide a moat to shield them from competition while also worrying about overconcentration among both their suppliers and their customers. These evolving market dynamics will influence what levers governments might use to shape AI development and deployment to ensure its consistency with U.S. national security interests.

Powerful AI capabilities may bring significant benefits for, but may also have serious repercussions on, U.S. national security interests. This paper considers three categories of national security interests: (1) how to enable beneficial AI uses while minimizing risks associated with the misuse of powerful AI, (2) how to ensure reliable and controllable AI system behavior, aligned with the intentions and values of their developers and users, and (3) how to maintain the United States’ strategic geopolitical advantage in the development and global diffusion of AI.

Realizing the benefits of powerful AI will require broad adoption and diffusion of AI capabilities. But expanding the use of AI may also increase security risks associated with the misuse of AI, namely the ability of malicious actors to deliberately employ powerful AI for harmful purposes. As AI systems become more capable, they can lower the barriers to conducting cyberattacks, developing biological or chemical weapons, generating disinformation at scale, or supporting adversary military operations. The April 2026 announcement by Anthropic that its Mythos model had discovered a dangerous number of zero-day cybersecurity exploits, and its decision to withhold a public release of the model, underscores that these risks are no longer hypothetical.¹⁶ AI developers build safeguards into their models to prevent misuse, but these guardrails are imperfect, susceptible to circumvention through jailbreaking, and require continual investment to maintain as model capabilities advance. They are also discretionary, left up to individual firm decisions. Moreover, the growing availability of open-weight models, where model parameters are publicly released, means that safety guardrails can be removed by anyone with sufficient technical capability.

The United States also has a national security interest in ensuring AI systems behave in ways intended by their developers or users and avoiding AI misalignment. At one end of the spectrum, misalignment includes relatively mundane technical failures, such as an AI system that misinterprets instructions, produces unreliable outputs, or acts unpredictably in novel situations. At the other end, it encompasses more severe scenarios in which highly autonomous AI systems pursue objectives that diverge from human intentions in ways that are difficult to detect or correct. At an extreme, misalignment of powerful AI systems may create large-scale risks to human safety. AI developers use approaches such as reinforcement learning from human feedback and related techniques to train models to better align with what users want, but “interpretability”—understanding why a model acts the way it does—remains a significant challenge, particularly as systems grow more capable and are deployed with greater autonomy.¹⁷

Finally, the United States also has a set of geopolitical national security interests associated with how AI development and diffusion may shift the balance of power between states. The United States has a strategic interest in maintaining a leading position in AI relative to competitors, particularly China, given AI’s potential military and geoeconomic implications. AI systems are already being deployed in U.S. military operations, and a dominant AI capability could substantially alter the balance of power between the United States and its adversaries. Beyond direct military applications, if U.S. competitors gain a large share of the global AI market, this could create systemic dependencies and provide them with a powerful tool for economic leverage over other countries.

Importantly, these three interests are not independent and may at times be in tension. For example, the geopolitical interest in maintaining more advanced AI capabilities than strategic competitors may lead to dangerous AI race dynamics, where concerns about misuse and misalignment are de-prioritized to develop powerful AI capabilities as quickly as possible. The reverse is also true: Pausing U.S. AI progress until misuse and misalignment risks are fully resolved could compromise the country’s competitive position relative to adversaries who do not apply the same constraints. The overarching national security interest for the United States is advancing each of these three objectives in tandem, without sacrificing any one for the sake of another.

About This Paper

This paper does not seek to resolve any particular debate about how AI markets will evolve, nor to make direct causal arguments that certain market structures will be better or worse for U.S. national security interests. In fact, in most scenarios, there will likely be a mix of competing incentives as firms experiment with different ways to achieve profitability. By charting a range of plausible commercial scenarios and then mapping out how these scenarios may implicate U.S. national security interests in a variety of ways, the goal of the paper is to inform decision-makers in the private and public sector as they seek to align commercial incentives with national security interests. A richer understanding of these market dynamics will help U.S. policymakers and private sector actors seeking to shape these markets to appropriately calibrate their interventions to avoid unintended consequences.

The analysis and recommendations in this paper are informed by extensive desk research by the authors, as well as dozens of interviews conducted with industry, civil society, and government representatives in the San Francisco Bay Area, Washington, D.C., and the United Arab Emirates. Interview subjects included companies making AI chips, building AI data centers, and developing foundation models and AI applications; enterprises integrating AI into their business operations; venture capital firms and other firms investing in AI; AI safety experts; academic economists; and government officials regulating AI and developing strategies to integrate AI into their economies. Interviews were conducted confidentially to allow participants to share candid views.

The next section examines potential variations in AI companies’ commercialization strategies and assesses to what extent these commercial incentives align with U.S. national security interests. The subsequent section zooms out from the company level to the sector level, assessing how market structure and concentration may evolve across layers of the AI stack and the national security implications of these trends. The final section outlines a high-level framework for strengthening policy interventions to better align commercial incentives with national security interests.