Kazakhstan Shows How Nations Hedge Between US and China AI

Kazakhstan's AI minister, Zhaslan Madiyev, flew to Washington in June to join the American-led group working on AI supply chains. Three weeks later he was in Shanghai, signing a ceremony that made his country a founding member of China's rival initiative. According to reporting by the Associated Press, Astana did not need to explain the contradiction. This is what AI diplomacy now looks like for most of the world.

What happened

The two blocs have hardened into named institutions. The US-led Pax Silica gathers allies around advanced chips, data center deals and trusted-supply rules. China's World AI Cooperation Organization has 28 other founding members alongside Kazakhstan, including Pakistan and Russia.

Washington is explicit that membership is not meant to be shared. "Countries can't be trusted partners in Pax Silica while substantively engaging in initiatives that advance a competing vision for super intelligence," the State Department said. Beijing frames its project as the opposite: "AI should not be owned by major countries," China's embassy in Washington said. President Donald Trump put the stakes bluntly last month: "whoever wins AI, wins."

The hedging playbook

Kazakhstan is the clearest case because it is cashing in on both sides. It signed an investment deal worth up to $10 billion involving Nvidia to build a "data center valley," while Chinese direct investment hit a record $2.8 billion in 2025. Indonesia, which holds nickel and copper, joined China's organization while still courting US money. Singapore is weighing Beijing's group even as Anthropic opens its newest Asia office there.

"Everyone is hedging," said Dana Masalimova, a former Kazakh diplomat now at Harvard. "We don't want this binary choice."

Why it matters for builders

For teams deploying AI, the bloc competition is already a procurement question. Countries that join both camps need vendors that run on either stack, which pushes demand toward portable inference layers, multi-provider routing and open-weight models that do not depend on one export licence.

China's strategy is priced for that market. Beijing has funded AI-related projects in at least 77 low- and middle-income countries, according to AidData at William & Mary. That builds familiarity with Chinese open-weight tooling years before an enterprise contract is signed.

The counterweight is hardware: Pax Silica membership buys access to Nvidia-class compute, which is why Japan keeps US alignment on security while still allowing cheaper Chinese models for "lower-risk applications," as professor Takahiro Tsuchiya of Kyoto University of Foreign Studies put it.

The practical lesson: assume the customer's country may belong to both blocs, and architect for a model layer that can be swapped without rewriting the product.

The window is narrowing

"Every country in the world in some way or another is thinking about AI sovereignty," said Aalok Mehta of the Center for Strategic and International Studies, even if only the US and China can build frontier-scale models. For now, hedging works. The question is how long both clubs tolerate it.