China didn’t close the AI gap - it vaulted over it. Moonshot AI’s Kimmy K3, a 2.8 trillion parameter model, surged to the top of coding and data analytics benchmarks, surpassing OpenAI and Anthropic offerings despite US chip bans.
The model’s rise is not just technical - it’s economic. While US labs charge $50 per million tokens, Kimmy K3’s open-source version costs about a dollar. This 50x price gap exposes a structural flaw in Silicon Valley’s strategy: massive compute spending without efficiency gains. As Salim Ismail notes, the shelf life of a top model is now weeks, not months.
US restrictions may have accelerated China’s leap. Emad Mostaque argues that export controls pushed Chinese engineers to optimize around H-800s and domestic chips from Huawei and Alibaba. The result: a lean, high-performance model that runs on accessible hardware. "They used our playbook," said Alex Weiser Gross, "but skipped the R&D tax."
"US labs pay a premium while China builds equivalent intelligence for pennies."
- Alex Weiser Gross, Moonshots with Peter Diamandis
The cost advantage is reshaping global developer behavior. Rabbit Hole Recap highlighted that Chinese-backed models are drawing open-source contributors away from US ecosystems. If training shifts to Huawei-optimized stacks, American hardware and software influence erodes by default.
Silicon Valley is split. Big labs lobby for bans on Chinese models, while developers warn protectionism hides stagnation. The White House is weighing restrictions, but Marty Bent warns this plays into Beijing’s hands: "If the US shuts down open weights, the world adopts China’s."
"Banning the competition just hides the fact that US models have become too slow and too expensive."
- Saagar Enjeti, Breaking Points with Krystal and Saagar
The AI race is no longer about who computes first - it’s who computes best. With Kimmy K3, China has proven that innovation under constraint can beat abundance. The era of US dominance is over, not with a crash, but with a quiet, efficient takeover.




