Together AI signs $240 million IBM Cloud pact for massive Nvidia GPU deployment

Together AI announced a $240 million multi-year contract with IBM Cloud to host its open-weights inference service on a substantial Nvidia GPU cluster. IBM will supply a “large” deployment of Nvidia’s HGX B300 boxes, each containing eight B300 GPUs, slated to become active in Q1 2027. The provider cited IBM and Nvidia’s product roadmaps and rapid capacity delivery at the lowest token cost as decisive factors.

While the firm primarily rents compute from cloud and neocloud operators, it has recently begun placing its own GPU hardware in datacenters in Maryland, Memphis and Sweden. This partnership demonstrates how AI service companies are willing to work with competing cloud vendors to secure the compute needed for inference, fine-tuning and training workloads.