AI infrastructure is shifting from headline model launches to the less glamorous constraints that decide whether adoption scales: GPU capacity, power delivery and cost visibility. IBM and Together AI have signed a $240 million, multi-year agreement to build a large NVIDIA HGX B300 inference cluster on IBM Cloud, expected in Q1 2027. The practical implication is straightforward: open-model providers are buying dedicated infrastructure to compete on enterprise token economics, not just benchmark scores. Together AI says it already serves 400 trillion tokens a month, making this a capacity bet with operational consequences. Read IBM’s announcement of the Together AI agreement.
What else matters
NVIDIA turns AI power into a systems problem
NVIDIA is pushing an 800 VDC architecture for AI factories, with a hybrid power rack designed to fit existing AC facilities and deliver higher-density power to compute rows. The company says a row power centre supporting up to 2 megawatts is expected in 2027. That makes electrical design—not merely GPU supply—a near-term scaling decision for data-centre operators. Read NVIDIA’s 800 VDC roadmap.
GitHub exposes the tokens behind Copilot credits
GitHub’s updated AI usage report now breaks down input, output, cache-read and cache-write tokens by model, alongside the AI credits consumed. The detail is available to Copilot Business and Enterprise admins and individual users. It gives engineering and finance teams a usable way to identify expensive prompts, caching opportunities and model-specific spend. Read GitHub’s usage-report announcement.
IBM and CHRIST University open an enterprise-AI centre
IBM and CHRIST University have launched an innovation centre in Bengaluru covering generative AI, agents, large language models, RAG, automation and responsible AI. Students, faculty and researchers will work with watsonx.ai, watsonx Orchestrate and IBM Bob, backed by training and industry projects. The signal is strategic: enterprise vendors are treating university partnerships as a route to shape the next AI talent pipeline. Read the IBM–CHRIST University announcement.
Worth your time: the infrastructure papers behind 800 VDC
NVIDIA says its approach builds on joint Open Compute Project work with Google and Microsoft, including a March 2026 white paper and a July 2026 solid-state transformer specification. Read NVIDIA’s linked OCP specifications.
