MiniMax Triples Alibaba Cloud Spending Cap to $1.2 Billion as AI Compute Demand Surges
MiniMax, the Chinese artificial intelligence startup, has sharply raised its cloud computing procurement ceiling with Alibaba Cloud, tripling the three-year spending cap to $1.2 billion (approximately RMB 80.66 billion yuan) from the original $375 million — a 220% increase — as surging demand for AI model training and inference pushes the company well beyond its previous resource limits.
Under a supplemental agreement announced on August 26, MiniMax set annual procurement caps with Alibaba Cloud at $300 million for 2026, $400 million for 2027, and $500 million for 2028, up from prior ceilings of $115 million, $125 million, and $135 million respectively. The revision was triggered in part by the pace of actual spending: MiniMax had already consumed $75.6 million worth of cloud services from Alibaba Cloud in just the first half of 2026, equivalent to 65.7% of its original full-year limit. In full-year 2025, the comparable figure was $75.9 million — meaning the company is on track to roughly double its annual cloud expenditure this year.
The scale of the revision underscores the capital intensity of frontier AI development. MiniMax disclosed that its research and development expenses in 2025 totaled approximately $252.8 million, of which Alibaba Cloud purchases accounted for around 30%. With a planned acceleration in infrastructure investment — MiniMax intends to deploy approximately $1.621 billion, or about 80% of net proceeds from a share placement and convertible bond issuance completed in July 2026, toward AI infrastructure and model development by end-2027 — the company projects average monthly R&D-related new spending of roughly $92.6 million going forward. That spending trajectory formed a key basis for the expanded procurement cap.
Beyond raw compute, the revised agreement also signals a deepening strategic integration between the two companies. A separate adjustment to MiniMax's API service arrangement with Alibaba Group raises the three-year ceiling for MiniMax-supplied model APIs nearly nineteenfold, from $3.15 million to $62.5 million. Annual limits climb from $650,000 to $7.5 million in 2026, from $1 million to $22 million in 2027, and from $1.5 million to $33 million in 2028. As of the end of June 2026, MiniMax had already used 77.2% of its original 2026 API cap, with actual transactions reaching $501,700.
The expanded API scope covers MiniMax's audio and voice models for Alibaba's digital media and interactive applications, as well as its text, reasoning, and logic models for integration into Alibaba's broader software ecosystem. The two companies also plan to co-develop AI solutions targeting specific vertical industries — a move that would further entrench MiniMax's models within Alibaba's commercial infrastructure.
The relationship carries a financial dimension beyond the commercial. Alibaba currently holds an indirect stake of approximately 11.36% in MiniMax through affiliated entities, classifying the transactions as connected-party dealings under Hong Kong Stock Exchange rules. Because the transaction ratios exceed the exchange's 5% threshold, the revised agreement requires a formal announcement, a circular to shareholders, and approval from independent shareholders before taking effect.
The deal illustrates the broader dynamic reshaping China's AI industry: as model developers scale up training runs and inference workloads, cloud providers with large GPU clusters are becoming essential — and increasingly captive — infrastructure partners. For Alibaba Cloud, the contract represents a meaningful revenue commitment from one of China's most closely watched AI startups, while for MiniMax, locking in compute capacity at defined price ceilings provides cost predictability as it accelerates its buildout.
Related Coverage:
MiniMax’s ARR Tops $800M, but Margin Squeeze Tests Its AI Growth Model