Together AI Builds Dedicated B300 Inference Cluster on IBM Cloud

What the collaboration covers

The infrastructure is a dedicated cluster of NVIDIA B300 GPUs, purpose-built for inference and running on IBM Cloud. The cluster is backed by NVIDIA Spectrum-X Ethernet networking.

Together AI describes a clear split of roles. It operates the inference layer, IBM provides the cloud, and NVIDIA delivers the silicon and networking. The company also says it is planning ahead as token demand goes "parabolic", according to its blog post.

• NVIDIA: B300 GPUs and Spectrum-X Ethernet networking, which Together AI says are engineered for high-throughput inference.
• IBM: decades of experience running mission-critical infrastructure for the world's largest enterprises, per Together AI.
• Together AI: the inference platform for running open models in production.

Why Together AI says demand is rising

Together AI says it started the company on the belief that the future of AI should not be owned by a handful of closed labs. The company claims that bet is paying off faster than it expected, citing hundreds of trillions of tokens served per month to over a million developers, with demand still climbing.

In its view, enterprises and AI-native companies choose open models for two reasons. Their sovereign data stays theirs, and they get frontier-level performance at a fraction of the cost of closed models. Together AI argues that the more usage grows, the better that cost comparison looks.

What it means for enterprise users

Together AI says the combination of the three partners should deliver enterprise-grade inference at massive scale, meaning more production-grade tokens with the reliability, security and guardrails enterprises expect. The company calls its own platform the fastest and most efficient way to run open models in production.

The post does not give pricing, capacity figures, the size of the cluster or a date for when it will be broadly available. What it does make clear is the direction: Together AI wants open-source AI to run everywhere, at scale, and to be as fast and reliable as closed alternatives. It describes this collaboration as "a big step toward exactly that."

For enterprises weighing open models against closed ones, the announcement signals added dedicated GPU capacity for open-model inference on a major enterprise cloud. Details on how customers can access it are not part of the source.