India's Largest AI Factory Is Being Built in Chennai

Larsen & Toubro said on Thursday it has won an order worth between Rs 10,000 crore and Rs 15,000 crore to build India's largest single-cluster AI infrastructure facility, at its data centre campus in Chennai. L&T hasn't given an exact figure. The range comes from its own order classification, where a "mega order" falls in that band.

The customer is Together AI, the US company whose cloud platform is used for training, fine-tuning and serving open models. The builder is L&T's own AI stack: LTN Compute, its AI infrastructure arm, working with Vyoma.AI, the group's AI and data centre business.

What is actually being built

Ten thousand NVIDIA B300 GPUs in one integrated cluster, hosted at Vyoma's Chennai campus.

The single-cluster part is what makes this different. Plenty of Indian operators have GPUs. Far fewer have them wired together tightly enough to run one large job across the entire fleet. That gap is the difference between renting hardware and running an AI factory.

The campus is built for more than this one deployment. L&T describes a gigawatt-scale site with Phase 1 capacity of 250 MW and power infrastructure readiness of 150 MVA — the first factory sized to what the grid connection carries today, with headroom already engineered for what comes after.

What surrounds the GPUs matters as much as the GPUs. L&T is assembling hyperscale data centre infrastructure, accelerated computing, high-performance networking, ultra-low-latency interconnects and high-throughput parallel storage, plus the operations layer that keeps all of it running. The pitch to customers is one end-to-end stack instead of a parts list they have to integrate themselves.

The workloads it is meant to carry: large-scale inference, fine-tuning and training.

Why Chennai

Tamil Nadu has spent two decades building the unglamorous prerequisites — power, land in industrial corridors, water, and engineers who already live there. Strip away the branding and an AI factory is a heavy industrial facility. Where you put one is a power and cooling decision long before it is a software decision.

It also shifts what the city sells. Chennai has spent thirty years exporting IT services. This is compute production, which is a different business with different multipliers.

The part worth arguing about

An infrastructure company just entered the compute business. L&T builds ports, metros and refineries. Its move into AI factories is a statement that the binding constraint in Indian AI is no longer model access or talent — it is megawatts, land, and finishing a hard build on schedule. That happens to be exactly what an EPC company is good at.

Then there's the tenant. Together AI is American, and its platform will run on Indian soil serving global demand. Most Indian AI infrastructure gets framed as sovereign capacity for domestic workloads. This is the other pattern: India selling compute into the world market rather than buying it.

LTN Compute's stated ambit runs well past this contract — hyperscale AI data centres, sovereign cloud platforms, AI factory services, GPU-as-a-service and managed AI platforms, sold to governments, enterprises, cloud providers and AI companies. One order does not make a programme, but L&T is clearly treating it as the first proof point for what it calls its Gigawatt AI Infrastructure Mission.

"Artificial Intelligence is becoming foundational to every industry and AI Factories will power this transformation," said chairman and managing director S N Subrahmanyan, describing the Together AI deployment as a milestone in that mission. Together AI co-founder and CEO Vipul Ved Prakash put the demand side more bluntly, calling the work of making AI globally accessible the biggest infrastructure build-out in human history.

What to watch

This is an order, not a running facility. The questions that decide whether it becomes a landmark are industrial ones. When does Phase 1 energise? What does the power mix look like? How do you cool 250 MW in Chennai's climate? And does a second factory follow quickly enough to make "gigawatt-scale" more than a slide?

For Indian enterprises the near-term consequence is narrower and more useful. Serious training and inference capacity is being built in-country, at a scale that moves the latency, data residency and cost conversations. Worth watching even if you never buy an hour of it.

Source: L&T Bags Rs 10,000-Crore Order To Build India's Largest AI Factory In Chennai — NDTV