Alibaba Group has provided Chinese AI firm Moonshot access to a computing cluster of about 20,000 Nvidia chips [1].
The deal underscores the critical reliance of Chinese artificial intelligence developers on Western semiconductor technology to train large-scale models despite ongoing trade tensions. Access to such massive compute power is essential for the development of frontier AI capabilities.
Moonshot is utilizing the infrastructure to support its Kimi AI models [1]. The agreement allows the firm to leverage Alibaba's cloud computing capacity to handle the intense processing requirements of generative AI. This partnership enables Moonshot to scale its operations without the immediate need to purchase and maintain its own physical hardware clusters.
Reports regarding the specific hardware involved have seen some discrepancy. Some sources identified the hardware as H200 chips [2]. However, Alibaba denied the specific chip type while confirming the existence of the computing deal [2].
The arrangement highlights a growing trend in the Chinese tech sector where cloud providers act as intermediaries for high-end hardware. By renting access to chip clusters, smaller AI startups can compete with global giants while navigating the complex landscape of hardware procurement in China.
This infrastructure deal comes as the global race for AI supremacy intensifies. The ability to secure tens of thousands of high-performance GPUs is often the primary bottleneck for companies attempting to improve the reasoning and efficiency of their language models [1].
“Alibaba Group has provided Chinese AI firm Moonshot access to a computing cluster of about 20,000 Nvidia chips.”
This agreement illustrates the 'compute-as-a-service' model becoming a lifeline for Chinese AI firms facing strict import restrictions on advanced semiconductors. By utilizing Alibaba's existing cloud clusters, Moonshot can bypass some of the logistical hurdles of hardware acquisition, though it remains dependent on the availability of Nvidia's technology to remain competitive against U.S.-based AI labs.
