IBM and San Francisco–based Together AI have agreed a multiyear, $240 million contract to build a large-scale artificial intelligence infrastructure, Reuters reported on Tuesday. The project aims to create a cloud-hosted network intended primarily to run open-source AI models.
Technical foundation and components
The planned network will be built on Nvidia systems. Key elements of the solution include Nvidia’s HGX B300 platform, which links the company’s newer Blackwell-family processors, and Nvidia’s Spectrum-X networking technology. These components are designed to accelerate inference workloads — the computation that produces model outputs in response to user queries.
Focus of the investment and why it matters
The investment centers on inference, the phase when already-trained AI models generate responses to user requests. Demand for the computational capacity needed for inference is rising rapidly, prompting cloud providers and chipmakers to invest billions in dedicated AI infrastructure.
Together AI’s platform allows companies to run and train open models; Reuters specifically mentions DeepSeek, MiniMax and Kimi. After a funding round in July, the startup was valued at $8.3 billion.
Context and implications
The agreement underscores continued large-scale spending by major technology and cloud firms on AI infrastructure, particularly hardware and networking optimized for inference. Such investments are intended to expand the scalability and commercial availability of open-source models.
Reuters indicated that further details on deployment timelines and specific service terms may be released by the parties in subsequent announcements.
