China's Kimi K3 AI Model May Trigger a DeepSeek Moment. Expect a Lot of FUD.
Tae Kim
· Key Context by Tae Kim
· July 17, 2026 at 10:27
· ⏱ 2 min read
| Read on Substack ↗
Summary
China's Moonshot released Kimi K3, a massive AI model approaching frontier benchmarks, which the author expects to trigger 'FUD' similar to DeepSeek. However, the article argues this will ultimately drive more AI compute demand, not a glut, because scaling laws remain intact and serving such a large model requires high-end Nvidia hardware.
•Kimi K3 has ~2.8–3 trillion parameters, performs well on benchmarks behind Anthropic Fable and OpenAI GPT 5.6 but ahead of Opus 4.8 and GPT 5.5.
•Moonshot plans to release full model weights by July 27, 2026, enabling open-weight customization.
•SemiAnalysis noted K3 is so large it won't fit on a single Nvidia DGX B200 and requires higher-memory servers like GB300 NVL72 and B300.
•FundaAI said K3's inference gross margins are likely well below those of Anthropic and OpenAI due to model size and low user-base utilization.
Read time2 min
Length2,723 chars
Categoryfinance
Ideas
Tae KimSenior writer, Barron's; author of The Nvidia Way
The article directly states that Kimi K3 is 'so large it won’t fit on a single Nvidia DGX B200 and requires higher-memory, higher-powered AI servers such as the GB300 NVL72 and B300' — implying increa
The article directly states that Kimi K3 is 'so large it won’t fit on a single Nvidia DGX B200 and requires higher-memory, higher-powered AI servers such as the GB300 NVL72 and B300' — implying increased demand for Nvidia's highest-end hardware as the model scales, consistent with the author's thesis that compute demand will surge.
Risk: If the FUD triggers a temporary selloff in AI stocks as with DeepSeek, NVDA may face short-term volatility before the compute-demand narrative reasserts.