AI infrastructure business Lambda has secured a multibillion-dollar deal with Microsoft to deploy tens of thousands of NVIDIA GPUs across multiple years, in what marks one of the largest AI infrastructure agreements disclosed this year.
The San Francisco startup will supply Microsoft with NVIDIA's latest GB300 NVL72 systems as part of a major cloud computing expansion, extending an eight-year partnership into significantly larger territory.
The agreement positions Lambda as what it calls an "at-scale partner" for deploying infrastructure behind leading AI products, as hyperscalers increasingly rely on specialist builders to meet unprecedented compute demand.
The deal underscores surging demand for GPU capacity as enterprises race to deploy AI assistants and language models at scale.
Microsoft has been scrambling to expand its Azure cloud infrastructure following overwhelming usage of GPT-4 through its OpenAI partnership.
"It's great to watch the Microsoft and Lambda teams working together to deploy these massive AI supercomputers," Lambda CEO Stephen Balaban said in a statement released Sunday.
Lambda positions itself as "The Superintelligence Cloud," building gigawatt-scale data centres specifically for AI training and inference workloads.
Neither party disclosed specific GPU quantities or contract values, though "tens of thousands" of processors suggests hardware expenditure running into hundreds of millions of dollars at current pricing.
The NVIDIA GB300 NVL72 represents the chipmaker's newest rack-scale architecture, engineered for large language model operations at higher density and efficiency than previous generations.
Lambda faces stiff competition from CoreWeave, Together AI, and traditional cloud providers, including AWS and Google Cloud - all aggressively expanding GPU capacity.
Microsoft itself has committed billions to OpenAI whilst simultaneously building out Azure's own infrastructure.
Microsoft shares traded flat following the announcement, whilst privately-held Lambda does not disclose financial performance.
