Dean Ball on Kimi, Open Weights, and the Deceleration Thesis

Dean Ball, reacting to the release of the Kimi open-weight model, lays out a thesis that open-weight models are inherently decelerationist. He grants they accelerate diffusion — getting AI into more hands — but argues they decelerate frontier development by raising the cost of capital for AI infrastructure and undercutting the investment case for frontier labs.

He finds it surprising that China permits open-sourcing models as capable as Kimi (which he rates on par with top Q1 2026 models for agentic coding). He attributes this mostly to Chinese leadership not being "AGI-pilled" about the risks, and partly to export controls constraining Chinese inference compute paired with an export-driven strategy.

Ball predicts an open-weight-dominant world leads to AI becoming state-provided digital public infrastructure — "full AI communism" — which he considers dystopian. He claims every open-weight advocate he's met ultimately concedes this endpoint. On U.S. policy, he expects the Trump Administration to inject regulatory uncertainty around Chinese open-weight models through agency soft law rather than an outright ban.

Notable pushback came from Nathan Lambert (who questioned whether anyone actually expects the DPI endpoint), Neil Chilson (who argued open weights can still be accelerationist on real-world deployment speed), and Tim Sweeney, who dismissed the thread as the idle speculation of someone outside the lab world.