Open Kimi, Shut Nvidia: Moonshot's Free 2.8T-Param Model Spooks Chip Stocks 12.5% 🧠

Moonshot, an Alibaba-backed AI startup, released Kimi K3 on Thursday, a 2.8 trillion-parameter open-weight model it says matches Anthropic's Claude Fable and OpenAI's GPT-5.6, with free downloads beginning July 27. The launch triggered a broad sell-off across semiconductor equities: the Philadelphia Semiconductor Index (XSOX) fell 12.5% on the week, its worst weekly drop in over 15 months, even as the index remains up more than 60% year-to-date. Nvidia ($NVDA), Broadcom ($AVGO), Taiwan Semiconductor ($TSM) and AMD all posted sharp declines in the same window. Hong Kong-listed AI names also slid, with Zhipu down 28% and MiniMax down 16% in Friday trade.

The episode revived memories of January 2025, when China's DeepSeek release erased $589 billion in Nvidia market value in a single day — what CNBC called the largest single-day market-cap loss on record. CNBC host Jim Cramer, citing that precedent, framed the issue as one of cross-border trust. "There will always be another Deep Seek. But the one thing that has become consistent is the ingrained lack of trust between the two countries. I don't care who you hire–CrowdStrike, Palo Alto, you have to be a first class idiot to let China have any of your data, even your name," Cramer posted on X on July 17, 2026.

Kimi K3's reception has been measured in benchmark wins and pricing rather than promises. The model topped Arena's coding leaderboard this week with 1,679 points, displacing Claude Fable 5 into second place. Moonshot is charging $3 per million input tokens against Fable 5's $10, and the model handles a one-million-token context window, large enough to ingest an entire codebase in a single prompt. Founder and CEO Zhilin Yang outlined Moonshot's scaling approach in a recent session, citing three levers: token efficiency, context length and parallel agent swarms.

The market reaction has spilled into derivatives. Bernstein noted that crypto-style structured products tied to AI computing power are now trading, with CME Group preparing the first compute futures in partnership with Silicon Data and Intercontinental Exchange (ICE) announcing GPU-linked contracts with Ornn. The activity signals that traders are seeking direct exposure to the underlying compute resource rather than to the chip makers that supply it.