Nvidia's Huang defends Chinese open models after Kimi K3 rout

Nvidia's Jensen Huang is telling the market it read the last two weeks wrong. In an exclusive interview with Axios, Huang said Wall Street 'misunderstood the impact of Kimi again this time,' the 'again' being a nod to the DeepSeek panic that hit the same stocks earlier in the cycle. His plain-English pitch, as reported: 'Free AI should be great for hardware. Free AI should be great for chips. Free AI should be great for data centers.'

The backdrop is that Moonshot AI's Kimi K3, released in mid-July, packs 2.8 trillion parameters and a 1 million-token context window, and semiconductor stocks are reportedly off more than 20% from their June 2026 peaks. Nvidia itself briefly lost its position as the world's most valuable company. Huang's counter-framing, as covered by Cryptobriefing, is that cheaper and more capable open-weight models will pull more workloads onto GPUs, not fewer, and that American companies should 'absolutely' be allowed to use Chinese models rather than have them banned.

He named names. Huang pointed to open models from DeepSeek, Alibaba, Tencent, MiniMax and Baidu as evidence China's open-source ecosystem is competitive at the frontier, saying 'these Chinese models are excellent' and that 'open-source models that are excellent should be used.' That reads as a direct rebuttal to White House tech adviser David Sacks, who warned last week that Kimi K3 shows the US losing the AI race and pushed for tighter policy.

The honest caveat is that Huang is talking his book. Nvidia sells the chips whose demand curve he is defending, and the Axios piece is a single-source CEO interview, so take the 'misread' framing as his argument, not settled fact. What the reporting doesn't give you is whether Huang directly engaged the administration's reported push to restrict Chinese models, or how Nvidia's own China export posture shifts if K3 becomes the default open baseline.

If Huang is right, this month reads as a gift to buyers of the semis and to any enterprise that now has a roughly $3-per-million-token frontier alternative to the closed labs. If he is wrong, we get to find out how much of the AI capex story was priced on the assumption that only American models mattered.