L&T Is Building a 10,000-GPU AI Factory in Chennai

Larsen & Toubro's Vyoma.AI is putting 10,000 NVIDIA B300 GPUs into a single cluster at its Chennai data centre campus, with Together AI — the US cloud platform — as the customer. L&T calls it the largest single-cluster AI facility in India, though the numbers underneath that claim are more interesting than the claim itself.

The power figures are the real story

Phase one is designed for 250MW, with power readiness of 150 MVA, on a site L&T describes as positioned for gigawatt-scale expansion.

Read those in order. 250MW for a first phase is not a data centre in the sense most enterprises use the word — it is an industrial power consumer. The gigawatt framing says L&T is not treating this as a one-off build but as the opening phase of a compute utility.

S N Subrahmanyan, L&T's Chairman and Managing Director, described AI factories as powering an industry-wide transformation, and called the Together AI deployment a milestone in what the company brands its Gigawatt AI Infrastructure Mission.

The corporate move behind it

L&T is transferring its data centre and cloud services business into Vyoma.AI for ₹14 billion, roughly $167 million.

That restructuring deserves as much attention as the GPU count, because it turns a services line inside an engineering conglomerate into a standalone compute business with its own balance sheet. Infrastructure companies becoming AI infrastructure companies is now a pattern rather than an exception — the overlap in required skills is larger than the software industry tends to admit. Land, power, cooling, project finance, multi-year builds: none of that is a software competence.

Why Chennai

The cluster runs through LTN Compute and targets large-scale inference, fine-tuning and training, supported by high-performance networking and parallel storage.

Chennai's pull is unglamorous — land, grid access, an existing industrial base. It also places a very large share of India's AI capacity in Tamil Nadu, which alongside the state's semiconductor and quantum announcements this week starts to look deliberate rather than coincidental.

What to watch

This is a deployment announcement, not a finished build. Three things decide whether it matters: whether the 250MW phase energises on schedule and where that power actually comes from, whether Together AI's demand holds long enough to justify phase two, and whether anyone else in India commits at gigawatt scale or L&T ends up alone at this tier.

Ten thousand B300s is a real number. Whether it becomes a utility or an expensive single-customer deployment depends entirely on what follows it.

Source: L&T's Vyoma.AI to construct NVIDIA B300 AI Factory in India — World Construction Network