The United States maintains its lead in the artificial intelligence race not primarily due to individual models, but rather through its command over critical infrastructure, including chips, cloud services, data centers, and submarine cables. This perspective was shared by Alex Capri, author of Techno-Nationalism, in a column for Fortune.
Capri argues that discussions about AI competition often focus on comparing models from companies like OpenAI, Anthropic, DeepSeek, Z.ai, or Moonshot. However, he emphasizes that advanced systems require a more extensive and capital-intensive foundation, including graphics processors, servers, cloud infrastructure, large data centers, and data transmission networks.
A key player in this domain is Nvidia, which Capri highlights as controlling about 85% of the global market for graphics processors used in training sophisticated AI models. He notes that Nvidia's advantage stems not just from its chips, but also from its CUDA software ecosystem, which has become the standard environment for many developers, creating substantial switching costs for those considering alternatives.
Building on this ecosystem, Capri outlines a broader "American AI stack" that encompasses hardware manufacturers like Broadcom, cloud hyperscalers such as Amazon Web Services, Google Cloud, and Microsoft Azure, along with model developers and big tech firms including OpenAI, Anthropic, Meta, and Alphabet.
Defense Framework
Capri connects the US advantage to collaboration between tech companies, the Pentagon, and intelligence agencies. He estimates that major AI infrastructure providers serve both the commercial market and contribute to systems for the defense sector.
As an example, he cites the US Department of Defense's Defense Autonomous Warfare Group (DAWG), which was established to develop autonomous military systems, and for which the Pentagon's budget request for fiscal year 2027 allocates $54.6 billion.
Another indicator of this trend is federal AI contracts. According to Brookings, the potential value of these contracts with the US Department of Defense is projected to reach $90.7 billion by 2026, with actual reserved funds in this area amounting to $7.2 billion.
Submarine Cables and Data
Another layer of infrastructure involves submarine fiber optic cables. Capri notes that major American tech firms are increasingly building their own backbone networks, thereby gaining influence over data transmission routes. He references the Pacific Light Cable Network, which in 2020 received government support for its segment between the US and Taiwan, while direct connections to Hong Kong were restricted due to concerns related to China.
Capri assesses that control over such routes is crucial not only for the United States but also for Chinese laboratories. Even if models are trained within China, they rely on the global internet for data collection, user access, and external APIs.
The author also brings attention to the CLOUD Act, a law passed in the US in 2018 that allows authorities to request data from American cloud providers through legal means, even if that data is stored outside the country.
Limitations of US Advantage
Capri acknowledges that the US dominance is not absolute. The production of key components for the AI stack is dependent on Asia, particularly companies like TSMC, SK Hynix, Samsung, Foxconn, Quanta, and Wistron.
However, he argues that Taiwan and South Korea are unlikely to leverage this situation against Washington, as they are integrated into a technology supply chain that is friendly to the US. Looking ahead, he suggests the possibility of two competing AI ecosystems—one American and one Chinese—with differing standards, infrastructure, and governance rules.
Previously, Nvidia, Meta, Microsoft, and 22 other companies urged US authorities not to restrict AI models with open weights. The signatories contended that the country's leadership depends on an open ecosystem rather than a single closed model.
It is worth noting that in November 2025, Anthropic announced plans to invest $50 billion in American AI infrastructure.