Why serving Kimi K3 in the U.S. may be 2–5x cheaper, not 10x
koltregaskes · x · 2026-07-21
Emad Mostaque’s claim that U.S. providers could serve Kimi K3 at 10x lower cost than Chinese competitors is challenged as overstated.
The post argues the real gap is more likely 2–5x cheaper, not 10x. It says:
- Chinese industrial electricity can be around $0.06/kWh with subsidies in key zones, while U.S. data centers average closer to $0.09/kWh.
- That power gap matters because electricity is roughly a quarter of inference costs.
- The U.S. advantage is hardware efficiency: Blackwell systems reportedly deliver about 2x usable compute per watt versus current Huawei Ascend clusters.
- On Nvidia hardware, mature tooling has cut costs 2–5x on similar large MoE models once weights are open.
- China still has advantages in build cost and labor, but the post argues silicon efficiency usually dominates for frontier inference.
Related event: Kimi K3 Reshapes Global AI Pricing, Sparking China-US Compute Cost Debate(9 posts)→
More from Infra
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
- Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11