IBM, Together AI Sign $240 Million Blackwell Inference Deal

Competition in generative AI is shifting from model training toward inference, the computing-intensive process of serving responses from deployed models. As companies move AI applications into production, throughput, latency and the cost of each request have become central constraints. IBM’s partnership with Together AI, an AI cloud startup focused on open-source models, pairs enterprise cloud infrastructure with specialized inference software to address those economics at scale.

IBM and Together AI announced a multi-year agreement valued at $240 million on Aug. 11, 2026, to build a large-scale inference cluster on IBM Cloud. The deployment will use NVIDIA HGX B300 systems based on the Blackwell architecture, along with NVIDIA Spectrum-X networking. The companies said the cluster will deliver high-performance inference for open-source AI models, aiming to increase throughput and help enterprises reduce the cost of running production AI workloads.