Moonshot AI’s latest open-source model was reportedly trained using thousands of Nvidia chips supplied through Alibaba, underscoring the growing importance of large-scale computing infrastructure while raising fresh questions over access to advanced AI hardware.
Chinese artificial intelligence startup Moonshot AI has emerged as one of the country's fastest-growing AI companies after reportedly using a massive computing cluster of around 20,000 Nvidia chips provided through Alibaba to develop its latest Kimi K3 large language model.
The arrangement highlights the scale of computing resources now required to build next-generation AI systems and illustrates how Chinese technology firms are continuing to expand their AI capabilities despite tightening US restrictions on exports of advanced semiconductor technologies.
Unveiled on July 17, Kimi K3 has been introduced by Moonshot AI as a 2.8-trillion-parameter open-source model. The company claims the model delivers performance comparable with leading AI systems developed by global competitors, including OpenAI and Anthropic. The launch has significantly strengthened Moonshot's position in China's rapidly evolving AI landscape.
The company's rapid growth has also attracted investor interest. Market estimates now place Moonshot AI's valuation between $20 billion and $35 billion, up sharply from about $4.8 billion earlier this year. Reports also suggest the company is evaluating the possibility of a public listing in Hong Kong.
Alibaba plays a strategic role
Alibaba has become one of Moonshot AI's most important strategic partners, serving not only as a major investor but also as the cloud infrastructure provider supporting the company's AI development. Following its participation in Moonshot's funding round in early 2024, Alibaba holds an estimated 36% stake in the startup.
Industry reports indicate that a substantial share of Moonshot's AI training infrastructure comes from Nvidia GPU clusters operated through Alibaba's cloud platform. The cloud giant has traditionally encouraged companies within its investment portfolio to build and operate workloads on its infrastructure.
Interestingly, reports suggest Kimi K3 has recently outperformed Alibaba's own Qwen AI models on several industry benchmarks, despite both teams having access to similar computing resources. The development is said to have triggered internal discussions within Alibaba regarding the competitive positioning of its AI initiatives.
Focus turns to advanced AI chips
Attention has also shifted to the hardware used to train Kimi K3. Reports indicate the computing cluster primarily relies on Nvidia's H200 Hopper processors, which have been widely deployed for AI model training before the introduction of Nvidia's newer Blackwell architecture.
However, questions have emerged over whether Moonshot AI may also have accessed Blackwell processors despite US export controls on advanced AI chips. Allegations from US officials suggest the company may have obtained access to newer Nvidia hardware through intermediaries in Southeast Asia, although no public evidence has confirmed how such processors, if used, were acquired.
The developments highlight two broader industry trends: the enormous computing power required to train frontier AI models and the increasing complexity of enforcing export controls as global cloud infrastructure and semiconductor supply chains continue to evolve.
See What’s Next in Tech With the Fast Forward Newsletter
Tweets From @varindiamag
Nothing to see here - yet
When they Tweet, their Tweets will show up here.
