According to Bloomberg’s investigation, Moonshot AI signed an agreement in August 2026 to rent 32 servers manufactured by Super Micro Computer, each equipped with Nvidia’s B300 GPUs.
The financial documents reviewed by Bloomberg reportedly valued the servers at 432 million yuan, equivalent to approximately $64.8 million.
The documents were submitted to a Chinese financial records database by Beijing Zhongguancun Bank, which financed the server purchase through an arrangement involving a company called Lingmeixin (Beijing) Technology.
The financing agreements were signed on August 4, 2026, and submitted to the database two days later.
The servers were reportedly located in Qingyang, a city in China’s Gansu province that is part of the country’s national computing infrastructure development programme.
However, Moonshot disputed the accuracy of the documents.
A company spokesperson stated that Moonshot had no business relationship with Lingmeixin and had neither participated in nor authorised the financing or guarantee arrangements.
After Bloomberg contacted the companies involved for comments, the documents reportedly stopped identifying Moonshot by name.
This leaves important questions about the exact nature of the arrangement and whether Moonshot ultimately obtained access to the hardware as described.
Why Nvidia B300 GPUs Are Important for AI Development
Nvidia’s B300 processors belong to its Blackwell Ultra generation of AI accelerators.
They are designed to handle demanding artificial intelligence workloads, including training large language models and running inference operations at scale.
AI companies require enormous computing resources to develop increasingly sophisticated models. These processors help accelerate the calculations needed to train neural networks, process large datasets and generate responses to user queries.
The B300 is particularly relevant for companies developing frontier AI models because of its memory capacity and computing performance.
Access to advanced GPUs can influence how quickly an AI company develops new models, handles larger workloads and serves millions of users.
For Moonshot AI, additional computing capacity could support the development of future Kimi models and improve the company’s ability to compete with leading AI developers in the United States.
However, access to powerful processors alone does not guarantee better AI performance. Model architecture, training data, software optimisation and research capabilities remain equally important.
How Moonshot AI Could Access Restricted Nvidia Chips
The reported arrangement highlights how Chinese AI companies are exploring alternative ways to obtain computing resources despite restrictions on direct purchases of advanced American processors.
The United States has introduced export controls targeting high-performance semiconductor technology to China.
These restrictions are intended to limit access to computing hardware that could support advanced AI development and other strategically sensitive applications.
However, companies can sometimes obtain computing capacity through third-party infrastructure providers and overseas data centres.
Instead of purchasing GPUs directly, an AI developer can rent access to servers containing powerful processors.
Such arrangements can allow companies to use advanced computing infrastructure without directly owning the hardware.
Earlier reports also indicated that Moonshot was exploring access to Nvidia’s Blackwell GPUs through overseas data centres.
In November 2026, reports described the company’s interest in leasing large computing clusters in Europe, primarily for inference workloads, while also exploring capacity for model training.
These developments illustrate how demand for advanced AI computing is creating increasingly complicated international infrastructure arrangements.
The $64.8 Million Server Deal Raises Export Control Questions
The reported server arrangement has attracted attention because US export regulations restrict the transfer of certain advanced Nvidia processors to China without appropriate authorisation.
The 32 servers were reportedly valued at approximately $64.8 million.
Bloomberg’s investigation indicated that the financing documents placed the equipment in China, potentially raising questions about compliance with US export restrictions.
However, Bloomberg could not determine whether the necessary export licence had been obtained.
Nvidia stated that the entities involved in the reported transaction were not its customers and that it did not provide them with replacement products, services or support.
Super Micro also said it did not conduct business with the entities identified in the investigation and was committed to complying with applicable US export and re-export control laws.
Moonshot, meanwhile, denied having a business relationship with Lingmeixin.
The allegations therefore remain subject to important factual and regulatory uncertainties. The reported documents do not independently establish how the hardware was originally sourced or whether any export-control violation occurred.
China Continues Depending on Nvidia Despite Domestic AI Chip Development
The reported Moonshot arrangement highlights a broader challenge facing China’s artificial intelligence industry.
Beijing has been encouraging domestic technology companies to reduce their dependence on American semiconductor manufacturers.
Companies such as Huawei are developing alternative AI processors intended to support China’s growing demand for computing infrastructure.
However, Nvidia continues to play an important role in the global AI hardware market.
Its GPUs are widely used for training and deploying large language models, partly because of their computing performance and mature software ecosystem.
Chinese AI developers have also built substantial experience using Nvidia’s CUDA software platform, making a complete transition to alternative hardware more complicated.
Developers must consider not only processor performance but also software compatibility, development tools, infrastructure availability and operational costs.
As a result, Chinese companies continue exploring different sources of advanced computing capacity while domestic semiconductor manufacturers work to improve their products.
Moonshot AI’s Kimi Models Increase Demand for Computing Power
Moonshot AI has become one of China’s prominent artificial intelligence startups through its Kimi family of large language models.
The company attracted international attention with the development of models capable of handling complex reasoning, coding and long-context tasks.
Its Kimi K3 model, introduced in July 2026, further strengthened its position in China’s competitive AI market.
The model reportedly uses a 2.8-trillion-parameter architecture and was positioned as a major competitor to leading international AI systems.
Developing and operating models of this scale requires significant computing resources.
Training involves processing enormous datasets, adjusting model parameters and repeatedly evaluating performance.
Inference also requires substantial computing capacity because every user request consumes processing resources.
For a company serving millions of users, securing reliable access to GPUs can become a major competitive advantage.
The reported B300 server arrangement could therefore be important for Moonshot’s future computing requirements if the company ultimately receives access to the hardware.
Earlier Reports Linked Moonshot to Thousands of Nvidia GPUs
The reported B300 arrangement is not the first time Moonshot’s access to Nvidia processors has attracted attention.
In July 2026, Bloomberg reported that Moonshot had access to approximately 20,000 Nvidia chips through a computing agreement with Alibaba.
According to the report, the arrangement involved Nvidia’s earlier Hopper-generation processors.
Alibaba is one of Moonshot’s major investors and has its own extensive cloud computing infrastructure.
However, Alibaba denied claims that it supplied H200 processors to Moonshot, calling the allegations groundless.
Separately, reports indicated that Moonshot was exploring access to newer Blackwell processors through overseas computing infrastructure.
The latest reported B300 server arrangement adds another dimension to the discussion about how Chinese AI developers are obtaining the computing resources needed to compete internationally.
The Bigger Picture
The Moonshot AI development highlights the growing difficulty of controlling access to advanced AI computing hardware through export restrictions alone. As demand for Nvidia processors increases, international data centres, server rental arrangements and third-party infrastructure providers are creating additional pathways for companies seeking computing capacity. This makes enforcement more complicated because regulators must monitor not only direct chip sales but also the movement, ownership and use of servers across international supply chains.
For Nvidia, the situation also illustrates the tension between its global commercial ambitions and US national security restrictions. Meanwhile, Chinese AI developers face pressure to secure powerful computing resources while expanding domestic alternatives. The outcome could influence the competitive balance of the global AI industry, particularly as companies invest heavily in developing increasingly capable models.
Looking Ahead
The next important development will be determining whether Moonshot actually obtained access to the 32 reported B300-equipped servers and whether the arrangement complied with applicable export regulations. Statements from the companies involved and further regulatory scrutiny could clarify the ownership, location and operational control of the hardware. The situation may also encourage US authorities to examine how third-party server rental and financing arrangements are being used to access restricted AI processors.
For China’s AI industry, access to advanced computing will remain an important factor in developing competitive models. Moonshot and other Chinese developers are likely to continue exploring different sources of computing infrastructure while domestic chipmakers work to close the technology gap. How effectively the United States enforces its semiconductor restrictions, and how quickly China develops alternative processors, could shape the next phase of the global AI competition.
SEO Metadata
- SEO Title: China’s Moonshot AI Has Access to Nvidia’s Top AI Chips
- Meta Description: China’s Moonshot AI reportedly rented 32 Nvidia B300-powered servers worth $64.8 million, raising questions about US chip export restrictions.
- Permalink: china-moonshot-ai-access-nvidia-b300-chips
- Focus Keyword: Moonshot AI Nvidia chips
- Secondary Keywords: Nvidia B300 GPUs, China AI chip restrictions, Moonshot AI Kimi, Nvidia Blackwell chips, US semiconductor export controls, Chinese AI industry
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
