The numbers come first. Nvidia controls roughly 85% of the global market for graphics processing units — the chips required to train frontier AI models — according to Alex Capri, author of Techno-Nationalism and a lecturer at the National University of Singapore. That single figure explains more about the AI power balance than any model leaderboard.
Capri's analysis, published in Fortune, argues that the global debate over AI has been distracted by model performance. Systems from China's DeepSeek, z.ai, and Moonshot have sharpened that focus, suggesting a narrowing gap with U.S. leaders. But the real contest, he writes, is being fought at the infrastructure layer: hyperscale data centers, cloud computing, AI servers, and the undersea fiber-optic cables that bind them together.
The ecosystem behind the chip. Nvidia's advantage is not only in silicon. Its CUDA software layer has become the default development environment for AI, creating high switching costs for any competitor. That software moat connects hardware manufacturers like Broadcom to cloud hyperscalers — Amazon Web Services, Google Cloud, and Microsoft Azure — which in turn power foundational AI developers including Anthropic, OpenAI, Meta, and Alphabet.
This network of firms is, in Capri's framing, the heart of a new U.S. AI industrial complex, one with deep ties to America's defense and intelligence establishment. The Department of Defense has signed standard operational agreements tapping major providers — including Google, OpenAI, Microsoft, Amazon Web Services, Oracle, and Nvidia — to deploy frontier AI tools on classified military networks. Separately, OpenAI, xAI, and Google have signed binding contracts granting the Pentagon 'all lawful use' for defense-related purposes, including autonomous weapons and mass surveillance, according to Capri. The Pentagon's FY2027 budget earmarks more than $54 billion for autonomous warfare and drone systems through a newly formed Defense Autonomous Warfare Group.
Federal AI contracting surged by $90.7 billion in 2026 alone, according to the Brookings Institution as cited by Capri.
Cables on the ocean floor. The infrastructure advantage extends underwater. American cloud hyperscalers account for 70% of usable undersea cables in 2026, yet Washington retains the right to restrict where those networks go and who has access. In 2020, U.S. regulators blocked the Hong Kong segment of the Pacific Light Cable Network — a project backed by Google and Meta — over Chinese espionage concerns, leaving some 13,000 kilometers of already-laid cable unused on the ocean floor.
Meta is now planning an around-the-world subsea fiber-optic cable covering 40,000 kilometers at a projected cost of $10 billion. Google, through its Pacific Connect Initiative, will spend over $1 billion to further connect Japan to the South Pacific.
Even China's frontier labs, training on domestically hosted infrastructure, remain dependent on American undersea cables wherever their models touch the global internet, Capri notes.
CEO Times reads it this way. Capital rewards clear rules, and right now the rules of the AI infrastructure game are written in American — in CUDA code, in Defense Department contracts, and in the legal authority Washington holds over privately owned global data pipelines. The emergence of low-cost Chinese models is a genuine competitive signal, but software cleverness does not substitute for the physical stack underneath it. Free enterprise built this lead; the federal contracts now reinforcing it are a force multiplier, not the foundation. The question for the next administration is whether Washington can sustain the regulatory clarity and capital environment that made the stack possible in the first place — or whether bureaucratic overreach will do what Beijing's engineers have not.
