IBM Signs $240 Million Deal with Together AI for Nvidia-Powered AI Cluster

IBM has entered into a $240 million, multi-year agreement with San Francisco-based startup Together AI to build a large-scale AI cluster powered by Nvidia systems, reflecting surging enterprise demand for open source AI models.

Together AI provides infrastructure that enables companies to train and deploy AI workloads using open models, including DeepSeek, MiniMax, and Kimi. The startup, which has received backing from Nvidia's venture capital arm, has positioned itself as a key provider of managed infrastructure for organizations that want the capabilities of frontier models without depending on proprietary API ecosystems.

Announced on Tuesday, the proposed cluster is designed to deliver inference capabilities — the process of running trained AI models to generate responses — for open source AI models. The infrastructure will be hosted on IBM Cloud and built around Nvidia's HGX B300 systems, which feature Nvidia's latest Blackwell processors, paired with Nvidia Spectrum-X Ethernet networking. Blackwell succeeds the Hopper-generation chips, such as the widely deployed H100 and H200, that underpinned the first wave of large-scale AI infrastructure buildouts.

Nvidia states that the platform can deliver 30 times more AI factory output compared with previous generations.

The initial U.S.-based deployment will include approximately 2,000 Nvidia Blackwell chips and is expected to become available in the first quarter of 2017.

For IBM Cloud — which trails AWS, Microsoft Azure, and Google Cloud in overall market share — the deal represents a targeted effort to compete as a destination for high-performance AI compute workloads, particularly for enterprises with regulatory and data-sovereignty requirements.

The agreement comes at a time of heightened scrutiny regarding the capabilities and security risks of increasingly autonomous AI systems, following a series of cybersecurity incidents involving models from Anthropic, OpenAI, and Meta.

Interest in open models — which are publicly accessible for modification and generally less expensive than closed-source alternatives — has grown in response to those concerns. Deploying open models on dedicated infrastructure also gives enterprises greater control over data handling, a consideration driving adoption in regulated industries such as finance and healthcare.

"Enterprises want the performance of the best frontier models without the closed-model price tag, and that only works if the infrastructure underneath is fast and reliable at scale," said Vipul Ved Prakash, CEO of Together AI, in a statement. "This cluster…[is] a big step in our push to make open source AI the obvious choice for enterprises."

The deal also marks the latest chapter in an ongoing partnership between IBM and Nvidia, as the two companies continue to expand AI infrastructure and software offerings for enterprise customers.

In March, IBM and Nvidia announced an expanded collaboration aimed at helping "enterprises operationalize AI at scale." That initiative spans GPU-accelerated data analytics, unstructured data processing, regulated AI infrastructure, and cloud computing and consulting services.