Moonshot Gains Access to 20,000 Nvidia Chips Through Alibaba Cloud Deal

The Computing Arrangement

Chinese artificial intelligence firm Moonshot has secured access to approximately 20,000 Nvidia chips through a computing agreement with Alibaba Group, according to people familiar with the arrangement. The chip cluster forms a substantial portion of the computing infrastructure powering Moonshot's Kimi AI models.

Alibaba, which ranks among Moonshot's largest investors, maintains a practice of encouraging portfolio companies to utilize its cloud services. The arrangement gives Moonshot access to Nvidia's Hopper generation processors, the chip architecture that dominated AI training and deployment before Blackwell's introduction.

An Alibaba spokesperson denied specific aspects of the arrangement, stating that "the alleged supply of H200 chips to Moonshot from us is completely groundless." The company did not address the broader computing relationship.

The Export Control Landscape

U.S. export restrictions on advanced semiconductors have transformed computing access into a defining constraint for Chinese technology companies. Washington's controls specifically target cutting-edge Nvidia chips, forcing firms to seek alternative pathways to the processing power required for large-scale AI development.

Moonshot is reportedly accessing Nvidia's newer Blackwell processors through Southeast Asian channels and actively pursuing additional chip capacity to train its next-generation AI system. The strategy reflects a broader pattern among Chinese AI developers navigating the bifurcated semiconductor supply chain.

The computing bottleneck has intensified as model sizes expand. Moonshot unveiled Kimi K3 earlier this month, a 2.8-trillion-parameter system the company described as the world's largest open-weight AI model. Moonshot claimed K3's performance approaches that of Anthropic's frontier Fable model, though independent verification remains limited.

Performance and Scrutiny

Kimi's trajectory has drawn attention within China's AI ecosystem. The model surpassed Alibaba's own Qwen system on key performance benchmarks despite access to comparable training resources, a development that reportedly disappointed some employees at the e-commerce giant.

The achievement arrives alongside mounting scrutiny from U.S. officials. Michael Kratsios, the White House's top technology adviser, stated last week that the U.S. government possesses information suggesting Moonshot distilled Anthropic's Claude Fable 5 model to develop Kimi K3. Model distillation involves training a smaller system to replicate the behavior of a larger, more capable model, a technique that can transfer capabilities without full access to the original training data or architecture.

Neither Moonshot nor Anthropic has publicly addressed the distillation allegation. The U.S. Commerce Department and Nvidia declined to comment.

The Investor-Infrastructure Dynamic

The Moonshot arrangement illustrates how investment relationships in China's AI sector increasingly intertwine with infrastructure access. Cloud computing has become a strategic lever for large technology conglomerates, enabling them to extend influence over portfolio companies while monetizing surplus data center capacity.

For startups like Moonshot, these arrangements solve immediate resource constraints but create dependencies that may shape long-term strategic flexibility. The computing power required to train frontier models continues to escalate, and access to sufficient chip capacity has emerged as a competitive advantage in itself.

The geographic workarounds Chinese firms are employing to access restricted chips add complexity and potential vulnerability to supply chains. Southeast Asia has become a critical node in these arrangements, though the durability of such channels remains uncertain as U.S. export enforcement evolves.

Moonshot's next model will test whether the company can maintain its development pace under tightening constraints. The firm's ability to secure additional Blackwell-class chips, whether through Alibaba or alternative channels, will likely determine its position in China's increasingly competitive generative AI landscape.