IBM and AI infrastructure company Together AI have agreed to a $240 million multi-year deal to develop a large AI inference cluster on IBM Cloud, as demand for computing power to run artificial intelligence applications continues to accelerate.
The planned cluster will be built around NVIDIA HGX B300 systems, which link the chipmaker’s newer Blackwell processors and NVIDIA Spectrum-X Ethernet networking technology. The infrastructure is expected to be available from the first quarter of 2027.
Together AI intends to use the cluster to deliver AI inference services for customers running open-source and other advanced AI models. Its platform supports models from developers including DeepSeek, MiniMax and Kimi.
The agreement comes as the AI industry shifts from a focus primarily on training large models toward expanding the infrastructure needed to operate them at scale. Inference — the stage where trained models process requests and generate outputs — is becoming an increasingly important source of demand for high-performance computing.
NVIDIA’s latest-generation systems are designed to handle demanding AI workloads, while high-speed networking is becoming increasingly important as data centers deploy larger clusters of accelerators.
For IBM, the agreement adds to its efforts to expand AI infrastructure and cloud services for enterprise customers. For Together AI, the additional capacity is expected to support the company’s efforts to provide faster and more scalable access to AI models.
The deal also reflects the broader investment taking place across the AI infrastructure sector, where cloud providers, chipmakers and AI companies are committing billions of dollars to computing capacity as adoption of generative AI expands.
With the new cluster scheduled to come online in 2027, the partnership is expected to add significant computing capacity for AI inference workloads and further strengthen IBM’s position in the growing AI infrastructure market.
Leave a comment