Key Takeaways

  • MiniMax increased its three-year Alibaba Cloud purchase ceiling by 220% as training and inference demand accelerated.
  • The revised agreement allows spending of up to $300 million in 2026, $400 million in 2027 and $500 million in 2028.
  • Rapid enterprise revenue growth suggests MiniMax increasingly needs computing capacity for commercial workloads, not only model development.

MiniMax has substantially expanded its cloud computing arrangements with Alibaba Group Holding, offering a fresh measure of how quickly infrastructure costs can rise when generative AI products move from development into widespread commercial use.

The Shanghai-based AI developer raised the three-year purchase ceiling for Alibaba Cloud services by 220% to $1.2 billion, according to Hong Kong stock exchange filings. The revised agreement covers the period through 2028.

For 2026, MiniMax now plans to spend as much as $300 million on Alibaba Cloud, nearly three times its original $115 million cap. The change came after MiniMax had already consumed two-thirds of that earlier budget by the end of June.

The annual increases become even larger over the next two years. MiniMax lifted its 2027 spending limit from $125 million to $400 million, while the ceiling for 2028 rose from $135 million to $500 million.

Those figures are ceilings rather than confirmed expenditures. Still, revising them so sharply indicates that MiniMax expects demand for accelerated computing, storage and network capacity to remain elevated. It also gives Alibaba Cloud greater visibility into the possible infrastructure requirements of one of China’s prominent generative AI developers.

Training a foundation model is only one part of the bill. Once customers begin using an AI service, inference workloads can become a recurring and potentially fast-growing expense. Video generation is particularly demanding because each request can involve substantial processing, memory and data movement.

MiniMax’s product portfolio spans its M-series large language models, the H3 video-generation model and the consumer-facing Hailuo AI application. Supporting these products requires infrastructure for model training, evaluation and live inference, with demand fluctuating according to product launches and user activity.

A separate change to MiniMax’s application programming interface arrangement with Alibaba points to growing commercial distribution. MiniMax raised its 2026 API service budget from $650,000 to $7 million, an increase of more than tenfold. Its three-year API spending ceiling grew to nearly 20 times the previous limit.

APIs allow models and related capabilities to be integrated into external applications and enterprise workflows. Higher API expenditure can therefore reflect increased usage by business customers, more services moving into production, or both.

MiniMax reported first-half revenue of $116.6 million, up 283%. Enterprise sales increased 700%, providing an important commercial backdrop to the cloud agreement. More enterprise adoption can bring steadier demand, but it also raises expectations around latency, availability and capacity planning.

The expansion lands amid a broader surge in infrastructure spending. Figures reported by Synergy Research Group put worldwide cloud infrastructure service revenue at about $419 billion in 2025. Q4 2025 accounted for roughly $119 billion, representing 30% year-over-year growth.

Meanwhile, forecasts compiled from Gartner and IDC show the widening role of AI in cloud investment. Gartner estimated public cloud end-user spending at $723.4 billion in 2025 and forecast $850 billion in 2026, including about $211.9 billion for infrastructure as a service. IDC put 2025 AI infrastructure spending at $318 billion and projected approximately $487 billion to $497 billion in 2026, with spending expected to exceed $1 trillion by 2029.

Most AI-centric infrastructure expenditure goes toward servers, particularly accelerated systems built around GPUs and other specialized processors. But the surrounding stack matters too. High-bandwidth networking, storage and Kubernetes-based orchestration increasingly determine whether costly processors can be used efficiently. The Cloud Native Computing Foundation’s 2026 survey found that 82% of container users run Kubernetes in production, illustrating how central orchestration has become to modern workloads.

That said, MiniMax is also taking on concentration and cost-management considerations by reserving substantially more capacity from Alibaba Cloud. Procurement ceilings do not reveal utilization efficiency, unit pricing or the split between training and inference. Those details will shape the economics.

For Alibaba Cloud, the agreement strengthens its position in China’s expanding AI infrastructure market, where hyperscale providers are competing to supply specialized clusters and managed services. For MiniMax, the revised limits provide more room to scale. The next test is whether revenue growth can keep pace with the computing appetite behind it.