IBM will deploy an NVIDIA HGX B300 cluster on IBM Cloud in a USD$240 million multi-year deal to power cost-effective, enterprise-grade open-source AI inference for Together AI.
On 11 August 2026, IBMs and Together AI announced the start of the partnership. The deal includes the USD $240 million multi-year commercial agreement. The collaboration targets provide high-performance infrastructure optimized for scaling open-source AI model inference for enterprise customers to increase their efficiency and token economics.
In addition, IBM will configure a large-scale computer cluster that will utilize the NVIDIA HGX B300 to run the AI model. The system will use NVIDIA Spectrum-X Ethernet interconnected system to enable high-performance communication between nodes. The cluster is scheduled to be generally available in Q1 2027 and will be delivered through IBM Cloud. This deployment marks the first dedicated, large-scale inference cluster constructed on IBM Cloud leveraging NVIDIA’s HGX B300 hardware platform.
According to Together AI CEO Vipul Ved Prakash, the goal is to provide enterprise users with performance comparable to closed frontier models without the high costs of proprietary APIs, lowering the cost per token for production workloads. According to NVIDIA, HGX B300 systems are designed to deliver up to 30 times more AI factory output compared to prior-generation GPU architectures. Together AI will utilise this infrastructure to serve real-time open-source model inference, fine-tuning, training, and agentic workflows for enterprise clients.
Founded in 2022, Together AI currently powers over one million developers and serves approximately 400 trillion tokens monthly on its AI-native platform. The deal follows Together AI’s USD $800 million Series C financing round, which valued the company at USD $8.3 billion. This deployment expands on ongoing technical collaborations between IBM and NVIDIA, spanning GPU-native data analytics, unstructured data extraction, consulting services, and hybrid cloud implementations.
