Jensen Huang: Kimi K3 and DeepSeek Are Not Threats — They Boost AI Demand

What is Kimi K3, and who released it?

Chinese AI startup Moonshot AI (月之暗面) released a new large language model called Kimi K3 in July 2026, according to both CNYES and TechNews. CNYES reported that "Chinese company Moonshot AI released the new-generation large language model Kimi K3 in July" (E2), and TechNews separately described Kimi K3 as "a large language model released by a Chinese AI startup in July 2026" (E6). Both outlets place the launch in the same month, with reporting on the reaction dated 2026-07-22.

What does the market get wrong about Kimi K3?

According to TechNews, Jensen Huang said the biggest misunderstanding surrounding Kimi K3 is the belief that a low-cost model will reduce AI investment demand. His stated view is the opposite: "The market's biggest misunderstanding about the latest Chinese AI model Kimi K3 is thinking that a low-cost model will reduce AI investment demand, but the truth is the exact opposite — the smarter and cheaper an AI model gets, the more users it attracts, which will ultimately drive even greater AI computing demand" (E5).

How does Huang respond to fears that cheap models will cut AI investment?

CNYES reported that Huang was responding directly to concerns that Kimi K3 signals AI over-investment, and called that reasoning backwards: "This kind of thinking is completely backwards" (E1). He tied this rebuttal to the same logic reported by TechNews — that a smarter, cheaper model draws in more users and therefore more computing demand rather than less (E5). The two outlets, reporting on the same 2026-07-22 remarks, describe one consistent argument: falling model cost is presented as a demand driver, not a demand killer.

Why does Huang draw a parallel with DeepSeek?

According to TechNews, Huang said the market had made the same mistake before: "The market also misunderstood DeepSeek in the past. In fact, DeepSeek was a positive development for the entire AI industry, and Kimi is the same" (E7). CNYES quoted a closely related version of the same comparison: "Many people completely misunderstand Kimi K3, just as they misunderstood DeepSeek — the more users adopt it, the better it actually is for the development of the AI industry" (E3). Both outlets frame Kimi K3 as a repeat of the DeepSeek episode rather than a new kind of threat.

Why would wider adoption of smarter, cheaper models boost AI industry demand?

The demand argument rests on the same two data points cited above: CNYES's report that more users adopting a model like Kimi "is actually very good for the development of the AI industry" (E3), and TechNews's report that DeepSeek's wider adoption was "a positive development for the entire AI industry" (E7). Taken together, both outlets attribute to Huang the same underlying mechanism — user growth around a cheaper, smarter model translates into industry-wide demand growth, rather than a diversion of spending away from AI infrastructure.

What role do open and closed models both play in Huang's vision of the market?

CNYES quoted Huang describing a market that needs both categories of model: "We need ready-to-use closed models like those from Anthropic and OpenAI, but we also need excellent open models like Kimi" (E4). TechNews reported a parallel statement on the same point: "The future AI market needs commercial services like those from Anthropic and OpenAI, and it also needs excellent open-source models like Kimi, to jointly drive continued growth in the AI industry" (E8). Both reports credit Huang with the same closed-plus-open framing rather than treating the two categories as competing alternatives.

What this means

Across both CNYES and TechNews' 2026-07-22 reporting, Huang applies one consistent lens to two separate models released roughly a year apart: DeepSeek was previously misread as a threat and, in his account, turned out to be a net positive for the AI industry (E7); Kimi K3, released by Moonshot AI in July 2026 (E2, E6), is now being read through the same misunderstanding, according to Huang (E3). The specific fear he addresses — that a cheaper model lowers AI investment demand — is the one he rejects most directly, calling it "completely backwards" (E1) and arguing smarter, cheaper models expand the user base and the resulting computing demand instead of shrinking it (E5). That demand argument sits alongside a separate, market-structure claim in the same reporting: Huang describes the AI market as needing both closed commercial models from Anthropic and OpenAI and open models like Kimi (E4, E8) — a framing that treats Kimi's openness as complementary to, rather than competing with, closed-model providers.