
Caught Between Two Stacks: The US-China Contest for the Rest of the World's AI Future
Washington and Beijing are each writing the constitution for the age of artificial intelligence. The US-led Pax Silica, launched in December 2025, drew more than thirty governments and economies to its second summit in June. China's answer, the World Artificial Intelligence Cooperation Organization, was founded in Shanghai in July by 29 signatory states with its headquarters in the city. One club guards chips and supply chains; the other preaches openness. The rest of the world must now navigate both.
Kazakhstan is the only country so far belonging to both blocs, and it shows what hedging looks like in practice. It signed the Pax Silica declaration in Washington, joined the Shanghai organization weeks later, and now hosts American GPU clusters alongside record Chinese investment. For a middle power with cheap electricity and critical minerals, refusing to choose has been profitable.
The hedging has hardware behind it. On October 23, the US company Firebird secured an American export license for 45,000 Nvidia Grace Blackwell chips bound for Kazakhstan, part of a $10 billion plan for a cluster running 100,000 advanced GPUs. The first phase, built with Kazakhtelecom at Ekibastuz, is due to start commercial operations in the third quarter of 2027. Nvidia chief Jensen Huang made his first visit to the country for a fireside chat with its AI minister.
Indonesia, another mineral-rich power, has joined China's organization while keeping its American options open. Its economic council vice chair said Jakarta wants to be part of both and "choose the best system." The leverage is geological: Indonesia produces about 62 percent of the world's mined nickel, and nickel and copper run through everything from batteries to data-center power systems.
The dilemma is not only for the Global South. Analysts at the Carnegie Endowment for International Peace note that countries like Britain and Germany have economic and military weight but no homegrown frontier model, so access to data centers and chip deals depends on aligning with Washington — putting their AI future at the mercy of policymakers in Washington and Beijing. As one CSIS expert put it, every country dreams of AI sovereignty, but perhaps only the US and China can truly build the largest models.
Dependence has already bitten. In June, US export restrictions on Anthropic's newest models forced the company to suspend access worldwide; the curbs were lifted less than three weeks later. The episode, aired at the company's Singapore launch, showed governments exactly how fast Washington can switch off their AI — and why Chinese open-weight models look like useful insurance.
Decoupling remains more slogan than reality. A Washington think tank estimates Chinese-accessible compute will grow from about 4.6 million to 8.8 million H100-equivalents next year, with roughly 65 percent of that capacity running on US-designed chips. Export controls may hasten Huawei-style alternatives, but for now the two stacks are still entangled.
Analysts warn the era of picking freely from both menus may be ending. As rivalry hardens, countries could be forced to choose: America's most advanced chips or China's cheaper open systems. Paraguay became the 26th Pax Silica signatory on October 23; more capitals will have to pick a chair. The countries that wanted both may soon discover that both want an answer.