Alibaba Supplies Rival Moonshot AI with 20,000 Nvidia Chips

What happened

Alibaba has provided Moonshot AI with computing capacity equivalent to roughly 20,000 Nvidia chips to support the development and deployment of the startup's Kimi artificial intelligence models, according to multiple reports. The arrangement reportedly includes access to Nvidia's H200 chips, part of the Hopper generation designed for demanding AI training and inference tasks. Alibaba denied specifically supplying H200 chips but did not refute the broader claim that Moonshot received capacity equivalent to around 20,000 Nvidia processors. Moonshot AI launched its Kimi K3 model around July 17, a 2.8 trillion-parameter AI system positioned as a direct competitor to leading Western AI models. The model reportedly performed comparably to top U.S. AI systems on benchmarks, despite being far cheaper to train. Alibaba holds approximately 36% of Moonshot AI, making it the startup's largest strategic backer and critical infrastructure provider. Moonshot's valuation has climbed to as much as $35 billion following the K3 launch, and the company has raised $3.5 billion in recent funding. Discussions around a potential IPO in Hong Kong are reportedly underway. The arrangement has drawn attention because Alibaba and Moonshot are both competitors in the Chinese AI sector. Reports indicate that Moonshot's Kimi K3 model has outperformed Alibaba's Qwen model on some measures, even while running on Alibaba's infrastructure.

Why it matters

This deal highlights a complex strategic dynamic for Alibaba. On one hand, providing computing capacity to Moonshot AI generates revenue for Alibaba's cloud computing division and strengthens its position as an AI infrastructure supplier in China. On the other hand, Alibaba is effectively enabling a direct competitor to its own Qwen AI models. The arrangement demonstrates how large internet companies can profit from backing startups that may one day threaten their own products. For Moonshot AI, access to this computing power is critical. A cluster of 20,000 H200 chips represents a massive concentration of computing resources, enough to train models that rival anything coming out of Silicon Valley. The K3 model's 2.8 trillion parameters put it in the same weight class as frontier models from OpenAI and Anthropic. The deal also raises questions about U.S. export controls. The H200 and advanced Hopper chips have been under strict U.S. export restrictions against China since 2022-2023, yet Chinese AI companies continue to access high-end Nvidia hardware through various channels. Moonshot is reportedly already pursuing Nvidia's next-generation Blackwell chips for its anticipated K4 model.

Bigger picture

The Moonshot-Alibaba arrangement underscores how aggressively Chinese AI companies are scaling up, even as Washington attempts to keep the most advanced semiconductors out of their hands. Chinese AI labs have demonstrated the ability to train frontier models using restricted Nvidia hardware, raising fresh doubts about the effectiveness of U.S. tech restrictions. DeepSeek, another major Chinese AI player, reportedly trained its latest model using Nvidia's Blackwell chip in Inner Mongolia despite existing U.S. bans. If Chinese companies continue to access top-tier hardware regardless of export controls, it could reshape the global AI competitive landscape. The practice has been difficult to monitor and will likely remain so for the foreseeable future, according to industry analysts. For the broader AI infrastructure market, the deal signals strong demand for cloud computing services capable of supporting trillion-parameter models. Nvidia's stock rose 2.9% following the reports, while Alibaba's American depositary receipts jumped 5.1%, suggesting investors see value in both the chip maker and the infrastructure provider.

What to watch

Investors will be watching whether Alibaba extends comparable computing agreements to other AI startups and how those deals might impact its competitive position. The speed at which Moonshot translates its computing capacity into more advanced AI products will also be key, particularly the development of the anticipated K4 model. Moonshot's potential IPO in Hong Kong could provide further clarity on the company's financial performance and growth trajectory. On the regulatory front, any U.S. government response to continued Chinese access to advanced Nvidia chips could reshape how and where AI compute gets allocated globally. If Washington follows through with actual penalties against intermediary access points, it could disrupt the supply chains that currently enable Chinese AI labs to acquire restricted hardware. Finally, the performance gap between Moonshot's Kimi models and Alibaba's Qwen models will be an important indicator of whether Alibaba's dual strategy as both investor and competitor is sustainable.