Microsoft Renews OpenAI Datacenter Funding
Microsoft has provided a new round of major funding to OpenAI specifically for datacenter expansion. The investment highlights the continued capital-intensive nature of training and serving large-scale AI models, reinforcing the close ties between leading AI labs and hyperscale cloud providers.
- This investment is part of a multi-phase partnership that began with a $1 billion investment in 2019 and was followed by a multiyear, $10 billion investment in early 2023. A restructured agreement in October 2025 gave Microsoft a 27% stake in OpenAI's for-profit public benefit corporation. - Training next-generation models is a primary driver of these capital needs; for context, the compute cost alone for training GPT-4 was estimated to be over $100 million. The upcoming joint Microsoft and OpenAI supercomputer, codenamed "Stargate," is projected to cost $100 billion. - To counter the high cost of running AI models at scale, Microsoft is developing its own custom silicon. The recently announced Maia 200 AI accelerator is specifically designed for inference workloads and powers services like Microsoft 365 Copilot and OpenAI's latest GPT-5.2 models. - Microsoft claims its Maia 200 chip offers 30% better performance-per-dollar than other hardware in its fleet and is more powerful than the latest custom chips from Amazon (Trainium) and Google (TPU) on certain benchmarks. This move is part of a broader trend among hyperscalers to build their own ASICs to reduce reliance on Nvidia and control costs. - The Maia 200 is fabricated on TSMC's 3-nanometer process and is built to excel at low-precision FP4 and FP8 computing, which is critical for optimizing the cost and speed of AI inference. This focus on inference addresses the significant ongoing expense of serving models to millions of users, which can surpass the one-time training cost. - OpenAI's long-term infrastructure plans are massive, with one internal memo outlining a goal to build 250 gigawatts of computing capacity by 2033, a figure that would require a significant portion of the world's chip fabrication capacity. More recently, OpenAI has revised its 2030 compute spending target to $600 billion, down from an earlier projection of $1.4 trillion. - The immense power requirements for these large-scale AI data centers are forcing a strategic shift toward building dedicated energy infrastructure. Both OpenAI and Microsoft have indicated plans to fund new power generation and transmission resources to avoid straining local grids. - Despite the deep partnership, OpenAI is also diversifying its infrastructure providers. In September 2025, it signed a massive $300 billion, five-year cloud computing deal with Oracle to secure additional capacity for its future needs.
