IBM-Together AI to Build Inference Cluster
IBM and Together AI announced a multi-year, $240 million partnership to build a massive AI inference cluster on IBM Cloud using Nvidia HGX B300 systems with Blackwell GPUs and Spectrum-X networking, slated to become operational in Q1 2027. The cluster will run open-source models and process about 400 trillion tokens per month, with platforms such as DeepSeek, MiniMax, and Kimi. The deal positions IBM to expand its neocloud presence and provide enterprise-grade inference, integrated with Red Hat OpenShift and IBM’s watsonx governance stack. Together AI plans to scale open-source infrastructure across enterprises, and investor reaction showed modest gains for IBM stock and a bump for Nvidia. The emphasis shifts from training to inference, promising faster and potentially cheaper deployment of AI workloads on IBM Cloud.
Where do you stand?

