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
- SkyPilot exits stealth with $20M to unify fragmented GPU compute across five clouds — skypilot_org · 2026-07-22
- SkyPilot exits stealth with compute orchestration for fragmented AI fleets — skypilot_org · 2026-07-22
- Production AI budgets include retries, routing, caching and observability—not just token prices — arx-go · 2026-07-22
- NVIDIA briefs analysts on Vera CPU and doubles down on monolithic agentic design — BenBajarin · 2026-07-22
- NVIDIA unveils Vera Rubin platform with claims of 10x better performance per watt — nvidia · 2026-07-22
- Why a 1GW Chinese AI data center may be plausible after all — teortaxesTex · 2026-07-22