IBM, Together AI ink $240 million deal for Nvidia-powered AI inference cluster

IBM and Together AI have signed a multi-year deal worth $240 million to construct a large-scale AI cluster on IBM Cloud utilizing Nvidia systems, the companies announced on Tuesday. This cluster will offer inference services for open-source AI models, which have emerged as a cost-saving alternative for businesses concerned about cybersecurity risks associated with models from companies like Anthropic, OpenAI, and Meta.

Inference, the process of executing trained AI models to produce responses, has emerged as a significant catalyst for demand in computing resources, prompting cloud service providers and chip manufacturers to invest heavily in expanding AI infrastructure. The cluster on IBM Cloud will leverage Nvidia's HGX B300 systems, which incorporate the company's latest Blackwell processors, and its Spectrum-X Ethernet networking equipment.

Nvidia has indicated that the Blackwell chips are specifically engineered for AI inference tasks. Together AI, based in San Francisco, provides a platform that enables businesses to train and deploy AI workloads on open models such as DeepSeek, MiniMax, and Kimi at reduced costs compared to proprietary systems. The startup was last valued at $8.3 billion in July.