In the race for artificial intelligence, the United States holds the lead not because of individual algorithms, but through total control over the key elements of the technological pyramid: chips, cloud platforms, data centers, and underwater data highways. This is a fundamentally different level of competition that is often overlooked when comparing specific models like those from OpenAI, Anthropic, or China's DeepSeek and Moonshot.
Nvidia stands out as a critically important asset here. The company controls roughly 85% of the global market for graphics processors designed to train advanced neural networks. However, its dominance is not limited to hardware. The CUDA software ecosystem has become an industry standard, creating enormous barriers to switching to alternative solutions. Around this core, a holistic "American AI stack" is taking shape: from equipment manufacturers (Broadcom) to hyperscalers (AWS, Google Cloud, Azure) and leading model developers, including Meta and Alphabet.
Defense framework and budget flows
A significant advantage for the United States also comes from the close linkage between the technology sector and defense structures. The largest AI infrastructure providers simultaneously serve both the commercial market and the military-industrial complex. For example, the Defense Autonomous Warfare Group (DAWG), created by the U.S. Department of Defense and responsible for autonomous military systems, is included in the budget request for fiscal year 2027 with funding of $54.6 billion.
The dynamics of federal AI contracts are also telling. In 2026, their potential volume for the Pentagon reached $90.7 billion, although the actually reserved funds currently amount to a modest $7.2 billion. This gap demonstrates the scale of planned expansion, not current spending.
Digital geography: cables and jurisdiction
A separate front in the struggle is underwater fiber-optic lines. American giants are increasingly building their own highways, gaining leverage over global data flows. A telling example is the Pacific Light Cable Network: the segment between the United States and Taiwan was approved, but a direct connection to Hong Kong was blocked due to national security concerns. This restriction directly affects Chinese laboratories as well, which, even when training models within the PRC, remain dependent on the global internet for data collection and servicing external APIs.
Nor should we forget legal instruments such as the CLOUD Act of 2018. This law gives American authorities the legal right to request data from local cloud providers, no matter where the servers are physically located—outside the country or within it.
Achilles' heel and forecast
Nevertheless, absolute dominance does not exist. The production of key components—chips and servers—depends on Asian giants: TSMC, SK Hynix, Samsung, Foxconn. However, Taiwan and South Korea are deeply integrated into the American technological chain, making it unlikely they would use this leverage against Washington.
In the long term, we will likely witness the formation of two isolated AI ecosystems—American and Chinese—with their own standards and rules. This is not merely technological rivalry but a fundamental geopolitical fault line.
My analysis: the focus on infrastructure rather than models is an accurate diagnosis. While China tries to circumvent chip restrictions, the United States is building an impenetrable ecosystem where every element—from CUDA to underwater cables—strengthens the overall defense. However, dependence on Asian manufacturing remains the most vulnerable point of this strategy, and it is precisely here that the next phase of confrontation may unfold.