Moonshot использует 20 000 чипов Nvidia от Alibaba для обучения модели Kimi

Moonshot — a Chinese AI startup that developed the Kimi language model — has reached an agreement with Alibaba Group Holding Ltd. to use a cluster of approximately 20,000 Nvidia Corp. chips. Bloomberg reported this on July 31, 2026, citing sources familiar with the operations of both companies. The deal underscores the persistent dependence of leading Chinese AI laboratories on Western semiconductors — despite years of U.S. sanctions pressure.

What is known about the Moonshot and Alibaba agreement

According to Bloomberg, the parties agreed on access to computing capacity: Alibaba is providing Moonshot with GPU infrastructure based on Nvidia Corp. chips — the largest manufacturer of accelerators for AI tasks. The scale of the cluster — approximately 20,000 chips — indicates a serious computational base: large training runs of modern language models require thousands of accelerators working in parallel, and this volume places Moonshot among first-tier organizations.

• Parties: Moonshot (flagship model Kimi) and Alibaba Group Holding Ltd.
• Cluster size: approximately 20,000 Nvidia chips
• Format: computing power agreement — an agreement for access to computing resources
• Details (specific chip models, timelines, financial terms): not disclosed by Bloomberg
• Sources: individuals familiar with the operations of both companies

Moonshot was founded in 2023 and is known for the Kimi model, which stood out for supporting ultra-long context — one of the key competitive advantages at the time of launch. The startup quickly became one of the most discussed players in the Chinese AI industry and attracted major venture capital funding. Kimi competes with Alibaba's Tongyi Qianwen, Baidu's Wenxin, and the internationally known DeepSeek — and the continued growth of the computational base is essential for this competition.

Why China still depends on Nvidia chips

Moonshot is far from the only Chinese AI company building infrastructure on Western accelerators through cloud platforms. According to Bloomberg, leading local laboratories continue to rely on American GPUs, utilizing the capacity of providers that managed to stockpile supplies before the latest waves of export restrictions.

The U.S. has consistently tightened export controls since 2022: first the A100 and H100 accelerators were banned, then their versions for the Chinese market — A800 and H800. As the Bloomberg report indicates, opportunities to work with Western chips in China persist — including through agreements with cloud providers that hold existing stockpiles. Domestic alternatives, primarily the Huawei Ascend accelerator, are gradually increasing in capacity, but the gap in software ecosystem and actual performance during large-scale training runs remains.

"The situation underscores China's continued dependence on Western semiconductors for the development of artificial intelligence,"

Bloomberg reported.

For Alibaba Group, the deal with Moonshot is beneficial in two dimensions: the technology giant locks a promising AI startup into its ecosystem and monetizes its accumulated GPU fleet. Competition for AI startups among China's leading platforms — Alibaba, Tencent, Baidu, ByteDance — has long become a strategic battleground for future positioning in the technology race.

What this means

A cluster of 20,000 Nvidia chips at a single startup is clear evidence that the U.S. sanctions strategy has not yet stopped the Chinese AI race, but has merely increased its cost and organizational complexity. Moonshot maintains its position in the first tier, and the deal with Alibaba demonstrates a working model of adaptation: partnership with a platform instead of direct purchases of banned components. The infrastructure gap with domestic alternatives has not yet closed — while the pace of the race shows no signs of slowing.