MiniMax has sharply increased the cloud-computing capacity it is prepared to buy from Alibaba, providing a concrete measure of how expensive China’s leading AI model race has become. South China Morning Post reported that the Shanghai-based company raised the ceiling of its three-year Alibaba Cloud procurement agreement by 220% to US$1.2 billion through 2028. The revised arrangement covers cloud services for model training and inference, as well as API services for the company’s AI products.
The new cap is not a declaration that MiniMax will certainly spend every dollar. It is a contractual upper limit that gives the company room to buy more capacity if its business and research plans require it. That distinction is important. Nevertheless, the increase shows that MiniMax expects its original allocation to be insufficient as it serves more users, trains models, and operates inference systems.
MiniMax had set a US$115 million ceiling for Alibaba Cloud services in 2026. SCMP reported that the company had used about two-thirds of that original amount by the end of June, leading it to raise the 2026 ceiling to US$300 million. The new annual caps are US$400 million for 2027 and US$500 million for 2028. The agreement also increases the 2026 maximum for Alibaba Cloud API services to US$7.5 million from US$650,000, with a US$62.5 million ceiling for the three years.
The Revised Agreement Reflects Training and Inference Demand
Cloud spending is one of the clearest operational costs for a foundation-model company. Training requires large amounts of computing power for extended periods, while inference costs rise as users submit more queries, generate more content, and use agents or APIs in multi-step workflows. A company that grows quickly can find that the capacity it planned for at the start of a year is no longer enough.
MiniMax’s revised agreement addresses both sides of that demand. The US$1.2 billion ceiling applies to Alibaba Cloud services used for model training and inference through 2028. The increased API-services allowance provides another route for MiniMax to build or deliver products on top of Alibaba’s cloud platform. SCMP’s report does not break down how much of the future cap will be used for training versus serving existing models, so it would be wrong to describe the deal as a direct forecast for any specific product line.
The 2026 adjustment is the most immediate signal. MiniMax had a US$115 million cloud-service ceiling for the year but had used around two-thirds by June 30, according to SCMP. Raising the cap to US$300 million gives the company more room for the second half. That does not prove that all of the remaining allowance will be consumed, but it indicates a faster-than-expected need for computing services.
The agreement highlights a structural feature of China’s AI market. Model developers can build their own infrastructure, rent capacity from a cloud provider, or use a mix of the two. MiniMax’s decision reinforces its relationship with Alibaba Cloud at a time when cloud providers are competing not only to host applications but also to become the underlying computing partner for AI companies.
EastFrontier previously covered how Alibaba Cloud was expanding its overseas compute and model-serving business. The MiniMax agreement shows the domestic side of that strategy. Cloud capacity is becoming a way for large platforms to participate directly in the growth of independent Chinese model developers.
A Larger API Budget Connects Cloud Capacity to Commercial Products
The API component of the agreement is smaller than the main cloud-services cap, but it is still revealing. MiniMax’s 2026 limit for Alibaba Cloud API services rises to US$7.5 million from US$650,000. Over the three years through 2028, the cap is US$62.5 million. APIs let applications call models and related services programmatically, which makes them essential for enterprise software, consumer products, content tools, and agent workflows.
A larger API ceiling can mean several things. MiniMax may be using Alibaba Cloud’s services to expand its own products, or it may need more capacity for services that customers access through APIs. The report does not specify exactly how the budget will be allocated, so the prudent conclusion is that MiniMax expects a material increase in its reliance on the cloud platform across product development and delivery.
The deal is also a reminder that model economics are more complex than a headline valuation or benchmark score. A company can attract users and show revenue growth while still facing substantial costs for chips, cloud services, engineering, data, safety work, and sales. The balance between those costs and the revenue generated by subscriptions, enterprise contracts, or APIs will determine which Chinese AI companies can scale sustainably.
MiniMax has already been making enterprise AI a larger part of its business. EastFrontier’s report that enterprise AI became MiniMax’s largest revenue engine examined the company’s first-half operating results. The Alibaba Cloud procurement is a separate development, but it provides one explanation for why cost management will matter alongside revenue growth. Serving business customers at scale requires reliable capacity, especially if clients expect low-latency inference and access to newer models.
Alibaba Gains a Long-Term Customer in China’s AI Race
For Alibaba Cloud, the revised cap creates a larger potential customer commitment through 2028. The company has been investing heavily in AI infrastructure, model services, and overseas capacity. A multiyear procurement agreement with MiniMax supports the cloud provider’s effort to convert China’s foundation-model boom into recurring demand for computing services.
The benefit is not guaranteed. A spending ceiling does not equal booked revenue, and MiniMax’s actual purchases will depend on product demand, pricing, available hardware, and the company’s strategic choices. Alibaba Cloud also faces competition from other Chinese cloud providers and from companies that prefer to operate more of their own AI infrastructure. But the deal gives MiniMax flexibility while giving Alibaba a potential long-term workload from a prominent domestic AI developer.
The annual caps make the expected growth path visible. The 2026 cloud-service ceiling is US$300 million. For 2027, it rises to US$400 million. For 2028, it reaches US$500 million. Adding those amounts produces the US$1.2 billion maximum for the three years. MiniMax’s API-services cap follows a smaller but similarly expanded path, with US$7.5 million allocated for 2026 and a total ceiling of US$62.5 million through 2028.
This is not simply a transaction between a startup and a cloud provider. It illustrates the infrastructure burden behind China’s AI model competition. As companies build larger models and offer more services to businesses and consumers, the cost of training and inference becomes a strategic issue. MiniMax’s 220% increase in its Alibaba Cloud purchase cap is a contractual response to that pressure.
The next measure will be execution. MiniMax must turn additional computing capacity into products and revenue that justify the expense. Alibaba Cloud must deliver the capacity, tools, and pricing that make the relationship attractive over several years. For the broader market, the agreement shows that cloud contracts are becoming as important as model launches in determining which Chinese AI companies can keep scaling.