US and China control 90% of global AI compute, panel says

The number that keeps surfacing in the question of who actually owns AI is 90%. At a Rest of World panel held during New York Tech Week on June 9, 2026, Sam Winter-Levy of the Carnegie Endowment for International Peace put the figure on stage: the US and China between them control 90% of global computing power and attract 70 to 80% of global AI investment. He called it "primarily a two-horse race," and the rest of the discussion was about what everyone else is supposed to do about that.

The framing that stuck with me came from Aditya Vashistha, the Cornell assistant professor leading the Cornell Global AI Initiative. He noted that current systems are designed by and for what researchers call WEIRD societies, which represent only 14 to 15% of the world's population, and asked the obvious question: "what about the 85%?" His examples were ChatGPT producing homogenized depictions of religious groups and struggling to accurately portray the roughly one billion people worldwide with disabilities.

Winter-Levy laid out three options he sees for middle powers. Pool resources into coalitions. Build sovereign models, which he flagged as very hard at current scale and cost. Or align with one superpower and negotiate durable access guarantees. Countries holding chokepoints on the supply chain (the Netherlands, Taiwan, Japan, South Korea) or with valuable inputs like India's data and talent or Ukraine's battlefield data have something to trade. His warning was blunt: "If you have a model that is okay but not great, and you're going up against a state that has access to the best models, you're just going to be totally outcompeted" in the sectors that matter.

Peter Micek, Access Now's general counsel, brought the geopolitical pressure into the room with a concrete story. RightsCon 2026 was cancelled after Chinese pressure on Zambia's government over expected Taiwanese participants and LGBTQ content discussion. He also flagged AI's slide into "direct control over life or death decisions" in military and humanitarian contexts.

The honest caveat is that this was a panel, not an investigation, so the headline numbers come without methodology and the strategy menu is conceptual. What the reporting does not give you is which middle-power coalitions are actually forming, how Anthropic's managed-access Mythos tier is gated in practice, or what binding outcomes (if any) came out of the February 2026 India AI Summit Vashistha referenced. The forward-looking part worth watching is whether small language models keep closing the gap with frontier ones, because that is the variable that decides whether sovereign AI for the other 85% is a real option or just a talking point.