IBM inks multi-year deal to run Together AI on NVIDIA HGX B300 systems in IBM Cloud, highlighting how big customers are building global AI stacks

International Business Machines is betting on a familiar foundation for today’s generative AI build-outs: NVIDIA’s data center GPU systems. On August 11, IBM announced a multi-year agreement worth $240 million with Together AI, a company known for providing access to large language model infrastructure. Under the deal, the two companies will deploy a large cluster of NVIDIA HGX B300 systems on IBM Cloud.

The announcement frames the arrangement as part of IBM’s wider effort to support customers running AI workloads on a managed cloud platform. For Together AI, which provides a route to deploying and operating large language models, the key point is the availability of NVIDIA accelerated infrastructure hosted within IBM Cloud rather than requiring every deployment to be assembled from separate hardware and hosting arrangements.

NVIDIA’s HGX B300 is part of the company’s broader family of accelerated computing platforms designed for AI training and inference. In plain terms, HGX systems bundle NVIDIA GPUs and related networking and software components into a pre-integrated cluster configuration so that data center operators can scale AI performance more quickly than if they were to assemble equivalent GPU capacity from individual parts.

The IBM-Together AI agreement also raises a competitive question that sits behind the market coverage: if NVIDIA is setting the performance standard for AI accelerators, is there room for other technology providers, including IBM, to compete meaningfully at the application and platform layers? The reality, as the deal suggests, is that enterprises often do not view AI competition as a pure hardware substitute game. Instead, large organizations are frequently choosing NVIDIA for the compute core while emphasizing differentiation through cloud management, security controls, compliance, and integration services.

IBM’s announcement ties the value proposition to the cloud experience, not to a rival accelerator architecture. The company did not, in the information referenced by the Yahoo Finance write-up, spell out every technical parameter that typically matters in these arrangements, such as the exact number of HGX B300 systems in the cluster, the target performance envelope, or how the capacity will be allocated across Together AI’s customer workloads.

NVIDIA’s role in these partnerships is likely to remain central as more companies move from experimenting with AI to running it in production. GPU demand has been driven by the scale required for model training and by the compute required for inference at volume. But for IBM, the opportunity is to position IBM Cloud as the operational home for AI, where customers can run models with managed services rather than managing the infrastructure stack end-to-end themselves.

For investors and industry watchers, the $240 million headline matters less as a standalone number than as a announcement of where enterprise AI deployments are heading. IBM is committing capital to a multi-year capacity relationship with an AI infrastructure partner, and it is doing so with NVIDIA systems as the acceleration layer. That dynamic suggests that even when “catch-up” conversations focus on one vendor’s dominance, buyers often respond by assembling best-in-class components into a single production platform.

What remains unclear, and what IBM did not disclose in the referenced reporting, is the commercialization timeline and how the deployed capacity will scale over the contract term. The announcement also does not provide granular details on whether the deployment will be limited to specific model types or inference scenarios, or whether additional NVIDIA configurations beyond the cited HGX B300 cluster will be added later. The next watch item is whether IBM and Together AI will publish more implementation details that clarify capacity ramp, service availability, and how customers will be able to consume the platform.