Together AI and Larsen & Toubro set up AI plant with 10,000 B300 processors in India

Together AI announced a partnership with Larsen & Toubro to build the largest artificial-intelligence cluster in India, based on 10,000 Nvidia B300 processors. The cluster is intended to run inference, development and training at full scale for the local open-model ecosystem and to provide sovereign compute infrastructure that does not rely on external API interfaces.

The 10,000 B300 processors represent a capacity designed for production-level large-model workloads, not just experiments. The B300 architecture, with expanded HBM3E memory and improved bandwidth, is built specifically for continuous-training loads and real-time inference on the same hardware. Together AI supplies the software and orchestration layer, while Larsen & Toubro is responsible for physical integration and site construction.

The move reflects a clear trend: nations and companies are seeking independent AI capability without dependence on American cloud providers. Open models running on locally hosted infrastructure enable data control, compliance with local regulation and flexibility for language and niche requirements. Together AI’s announcement explicitly states that the real story is the shift to open models on sovereign compute, not the size of the cluster itself.

Performance metrics, exact construction timelines and details of the first models to be run were not disclosed. The business model—whether a managed service, hardware leasing or a long-term partnership with government bodies—remains undefined. Together AI also did not indicate whether the cluster will serve only Indian customers or be globally available through its platform.