IBM and Together AI have signed a $240 million multiyear agreement to build a large Nvidia-powered AI cluster on IBM Cloud, with the infrastructure expected to come online in the first quarter of 2027 to serve inference workloads for open-source models. The deployment will use Nvidia HGX B300 systems built on Blackwell processors and Spectrum-X networking, and Nvidia says the B300 platform can deliver up to 30 times the AI factory output of previous-generation hardware as spending shifts from model training toward real-world inference. San Francisco-based Together AI says enterprises use its platform to train and run models including DeepSeek, MiniMax and Kimi at lower cost than closed systems while reassessing security incidents involving models from Anthropic, OpenAI and Meta; the startup, valued at $8.3 billion in July after raising $800 million, says it now processes roughly 400 trillion tokens a month. The deal gives IBM exposure to the neocloud market alongside companies such as CoreWeave and Nebius, extends the Blackwell push IBM and Nvidia outlined in March, and was followed by modest gains in both companies' shares.