IBM and Together AI Sign $240 Million Deal for Open-Source AI Inference on IBM Cloud

IBM and Together AI have signed a multi-year, $240 million agreement to deploy a large cluster of NVIDIA HGX B300 systems on IBM Cloud, dedicated to open-source AI inference. The infrastructure, expected to become available in the first quarter of 2027, is designed to meet soaring enterprise demand for running open models like Llama at production scale.

Key Takeaways

IBM and Together AI have signed a multi-year, $240 million agreement to deploy a large cluster of NVIDIA HGX B300 systems on IBM Cloud, dedicated to open-source AI inference.
The infrastructure, expected to become available in the first quarter of 2027, is designed to meet soaring enterprise demand for running open models like Llama at production scale.

AI synthesis3 sources6

IBM and Together AI have signed a multi-year, $240 million agreement to scale open-source AI inference, with IBM deploying a large cluster of NVIDIA HGX B300 systems on IBM Cloud. The dedicated infrastructure is expected to become available in the first quarter of 2027.

The deal reflects surging enterprise demand for running open-weight models at production scale. Together AI, one of the best-funded 'neocloud' startups — it raised an $800 million round at an $8.3 billion valuation in July — operates GPU infrastructure specialized in serving open-source models. Under the agreement, IBM positions the NVIDIA-powered cluster as a managed offering for customers who want open models without standing up their own fleets.

IBM has been repositioning its cloud and watsonx platform around open-source AI, betting that enterprises will standardize on open-weight models for flexibility, data control and cost. The partnership gives IBM Cloud access to Together AI's serving stack and gives Together AI additional geographic reach and enterprise credibility.

Availability in Q1 2027 would arrive amid a broader boom in AI infrastructure spending, with 'neoclouds' like Together AI, CoreWeave and others racing to secure GPUs and data center capacity to serve both open-source and frontier model workloads.