MiniMax raises Alibaba Cloud spending ceiling 220% to $1.2 billion through 2028
The AMW Read
A 220%/$1.2B multi-year cloud ceiling increase and a mid-year budget overrun update MiniMax's compute trajectory as a scaling CN frontier lab, exemplifying segment-wide compute-cost pressure without resolving an open debate.
MiniMax raises Alibaba Cloud spending ceiling 220% to $1.2 billion through 2028
MiniMax has raised the three-year purchase ceiling on its cloud computing deal with Alibaba Group by 220%, to US$1.2 billion, after the Shanghai-based AI company burned through two-thirds of its 2026 cloud budget by the end of June. This year's spending cap on Alibaba Cloud services jumps to roughly $300 million, nearly triple the original $115 million limit. Under the amended agreement, annual caps rise further in 2027, from $125 million to $400 million, and in 2028, from $135 million to $500 million. MiniMax also expanded its 2026 API-service budget with Alibaba more than tenfold, from $650,000 to $7.5 million.
The increase lands months after MiniMax closed a roughly $2.04 billion funding round in July and shipped two open-weight flagship releases in quick succession: M3 in June, a 1M-context multimodal model pitched against closed-source frontier systems, and H3 in August, a video-generation model that topped global benchmarks and now runs locally on consumer RTX 4090 hardware. Serving that pace of releases, plus API traffic on Token Plan pricing, evidently outran the original cloud budget faster than planned. AMW's news pipeline logged 21 MiniMax-related items in the last 90 days versus 13 in the prior 90 (per the AI Market Watch index; name-matched, pipeline-ingested coverage, not a census), consistent with a company scaling faster than its own infrastructure commitments.
For builders and investors, the read is less about the model releases and more about unit economics: locking in $1.2 billion of committed Alibaba Cloud spend through 2028, on top of a fresh funding round, signals MiniMax expects sustained training and inference load rather than a one-off compute spike. Enterprises evaluating MiniMax's API or self-hosted H3/M3 deployments should watch whether the expanded capacity improves availability and latency; investors tracking Chinese frontier labs' capital intensity now have a concrete data point — a mid-year budget overrun large enough to force a multi-year contract renegotiation.
#MiniMax #AlibabaCloud #ComputeEconomics #FoundationModels #ChinaAI #CloudSpending
