Overview
- IBM and Together AI announced the $240 million multi‑year partnership on Tuesday to deploy a large inference cluster on IBM Cloud that Together AI will operate for customers.
- The cluster will use Nvidia HGX B300 systems with Blackwell processors and Spectrum‑X networking, is reported to include about 2,000 Blackwell chips in an initial U.S. build, and companies target general availability in Q1 2027.
- Together AI plans to run open‑source models on the cluster and says its platform already processes roughly 400 trillion tokens per month, using the deal to give customers reserved, production‑grade inference capacity.
- Markets gave a modest vote of confidence after the announcement with IBM and Nvidia shares edging higher, but analysts say the partnership must prove performance on live production workloads before it changes IBM’s broader growth trajectory.
- The deal ties into broader industry trends where inference demand is outpacing training needs and where IBM’s recent moves, including Project Lightwell and Red Hat AI offerings, position it to sell regulated, enterprise‑grade infrastructure for open models.