Moonshot AI's Kimi K3 open-source model drives $35B valuation and hardware race

Moonshot AI Kimi K3 open-source AI model valuation Nvidia chips

Chinese AI startup Moonshot AI announced on 27 July that its new language model Kimi K3, with 2.8 trillion parameters, is fully open‑source—the largest open‑weight model by size. The release includes the complete model weights, a technical report and three core infrastructure technologies, allowing developers worldwide to download and run the model locally.

The launch sparked a surge in investor interest, pushing Moonshot AI’s valuation to $35 billion after a funding round that raised $3.5 billion, far exceeding its original $1.5‑2 billion target. Bloomberg reports the model’s performance rivals leading systems from OpenAI and Anthropic, prompting concerns in the United States about China narrowing the AI gap.

To train Kimi K3, Moonshot secured roughly 20 000 Nvidia chips from Alibaba, highlighting the role of U.S. hardware in Chinese AI development. Analysts note that deploying the model in‑house would require about 64 high‑end GPUs (e.g., Nvidia B200), costing roughly 30 million RMB and demanding substantial power and cooling infrastructure.

Separately, Moonshot AI introduced a coding‑focused model, Kimi K2.7 Code, which became available in Microsoft Foundry on 1 July 2026, targeting multi‑file software engineering tasks with a 256 K context window.

Entities: Alibaba · Kimi K3 · Microsoft · Moonshot · Moonshot AI · Nvidia

Claims

What the coverage asserts, and how well corroborated each claim is across sources.

[● 2 SOURCES] Kimi K3 is fully open‑weight, with model weights, a technical report, and three infrastructure technologies available for download. (Moonshot AI)
[● 2 SOURCES] Kimi K3’s performance is comparable to leading models from OpenAI and Anthropic. (Bloomberg)
[○ 1 SOURCE] The funding round originally aimed to raise $1.5‑2 billion but raised $3.5 billion. (Bloomberg)
[○ 1 SOURCE] Deploying Kimi K3 locally requires about 64 high‑end GPUs (e.g., Nvidia B200), costing roughly 30 million RMB. (Ti Media)
[○ 1 SOURCE] Alibaba supplied approximately 20 000 Nvidia chips to support training of Kimi K3. (Bloomberg)
[● 2 SOURCES] Moonshot AI released Kimi K3, an open‑source model with 2.8 trillion parameters, the largest open‑source model by parameter count. (Moonshot AI)
[○ 1 SOURCE] Moonshot AI’s valuation reached $35 billion after a funding round that raised $3.5 billion. (Moonshot AI)
[○ 1 SOURCE] Moonshot AI’s Kimi K2.7 Code model became available in Microsoft Foundry on 1 July 2026. (Moonshot AI)