How a $1.57 billion order to build India's largest Nvidia AI factory turns a storied engineering conglomerate into an AI landlord — and pushes the compute buildout beyond America's borders.

Larsen & Toubro builds bridges, refineries, and metro lines. Its newest project is a 10,000-chip Nvidia AI factory in Chennai — and it signals that the geography of AI compute is finally spreading past the usual handful of countries.

On August 13, India's Larsen & Toubro said it had secured an order worth up to 150 billion rupees — about $1.57 billion — to build an AI data center for the U.S. cloud platform Together AI, powered by Nvidia's high-end chips. Reuters reported the value lands in a range of roughly $1.05 billion to $1.57 billion. In L&T's own press release, it billed the project as India's largest Nvidia B300 AI factory. The company's shares rose as much as 1.2% on the day.

Two things make this more than a regional infrastructure contract. First, it marks a storied industrial conglomerate's entry into a brand-new business. Second, it's a concrete data point that the AI compute buildout — long concentrated in the United States — is going global in a way that will reshape where models are trained, served, and governed.

What L&T and Together AI are building

The specifics come from L&T and Reuters. The order was won by LTN Compute, L&T's AI-infrastructure subsidiary, and the facility will be hosted at the company's Vyoma.AI unit on its campus in Chennai, in southern India. It is designed to house 10,000 Nvidia B300 chips, which Business Standard reported makes it the largest such deployment in the country. The capacity will support Together AI's cloud platform for AI inference, fine-tuning, and training.

For L&T, this is a first. The company describes the deal as its foray into the AI factory business — a notable pivot for a firm whose decades-spanning résumé runs to ports, metro systems, defense hardware, and heavy engineering. It is bringing that industrial competence to a new kind of project: the large-scale, power-dense, precisely engineered buildings that modern AI runs inside.

For Together AI, the arrangement secures dedicated capacity in a fast-growing market without the company having to become a builder itself. It supplies the demand and the cloud platform; L&T supplies the construction, the site, and the operational muscle.

Why the buildout is going global

For most of the AI era, the physical infrastructure clustered in a familiar map: U.S. hyperscaler regions, a few European hubs, pockets of East Asia. The Chennai project is a marker that the map is widening, and the reasons are structural rather than incidental.

Demand for AI capacity is outrunning what any single geography can supply. Power, land, cooling, and skilled construction are all constraints, and no one country has an unlimited stock of them. Spreading the buildout across more regions relieves that pressure. At the same time, serving users in a given market is often cheaper and faster with compute located nearby, and a growing thicket of data-sovereignty and localization rules pushes workloads toward local soil. India, with its scale, engineering talent, and policy push toward domestic AI infrastructure, is a natural next node.

Nvidia sits at the center of this diffusion. Every new AI factory, wherever it rises, is another anchor customer for its chips and its software ecosystem. A B300 deployment in Chennai extends Nvidia's platform into a new market and deepens the dependence of that market's AI ambitions on Nvidia's hardware. The globalization of compute is, in practice, the globalization of Nvidia's franchise.

The stakeholders

For L&T, the deal is a strategic reinvention dressed as a construction order. The company is applying its core strength — building large, complex physical things on budget and on schedule — to one of the highest-growth demand curves in the world. If it can turn AI factories into a repeatable line of business, it converts a one-off contract into a durable franchise, and hedges a legacy engineering portfolio against slower-growth infrastructure. The risk is that AI data centers are a specialized, fast-evolving discipline; execution missteps on power density, cooling, or timelines are punishing, and the technology underneath keeps changing.

For Together AI, the project is capacity insurance in a market it wants to serve. By partnering with a proven builder rather than constructing itself, it gets scale quickly and offloads the physical risk. The trade-off is dependence: its Indian footprint now rests on a partner's execution and on a single large site.

For India, the factory is a step toward hosting, not just consuming, frontier AI. Domestic compute capacity matters for sovereignty, for serving local users with low latency, and for building an ecosystem of companies that can train and fine-tune models on home soil rather than renting everything from abroad. A 10,000-chip cluster is a foundation other local AI efforts can build on.

For Nvidia, each such deal is compounding advantage. The more countries anchor their AI ambitions on B300-class hardware, the wider and stickier its ecosystem becomes. Global diffusion of AI infrastructure is, for Nvidia, global diffusion of demand.

What to take from it

The first lesson is that the AI boom is an industrial boom, and it rewards industrial competence. Building an AI factory is closer to erecting a refinery than to writing software — it demands land, power, precision engineering, and project management at scale. Firms that already do that work, in regions that already have the sites, are finding an unexpected on-ramp into the AI economy.

The second is that compute is decentralizing, and with it, leverage. As AI factories rise outside the traditional hubs, the countries and companies that host them gain a measure of control over where AI is trained, served, and governed. The map of AI power is being redrawn one data center at a time — and it no longer runs through only a few zip codes.

The zoom-out

It's easy to picture AI as something that happens in California and a handful of cloud regions. A $1.57 billion order to stand up 10,000 Nvidia chips in Chennai is a reminder that the technology's foundation is being poured in far more places than that, by companies most people would never associate with artificial intelligence.

The globalization of compute changes the strategic calculus for everyone. For nations, it's a chance to host rather than merely rent the infrastructure of the AI era. For legacy industrial firms, it's a rare doorway into a high-growth market that plays to their oldest strengths. And for the chipmakers at the center, it's the surest sign yet that demand for their hardware is not a coastal phenomenon but a planetary one. The AI factory is becoming a global asset class — and the builders, not just the model makers, are cashing in.

Frequently Asked Questions

What did Larsen & Toubro announce?

On August 13, 2026, L&T said it had secured an order worth up to 150 billion rupees (about $1.57 billion, in a range from roughly $1.05 billion) to build an AI data center for U.S. cloud platform Together AI. The facility, which L&T calls India's largest Nvidia B300 AI factory, will host 10,000 Nvidia B300 chips in Chennai.

Who is building it and who will use it?

L&T's subsidiary LTN Compute won the order and will build the facility at the company's Vyoma.AI unit in Chennai. It will support Together AI's cloud platform for AI inference, fine-tuning, and training. The project marks L&T's entry into the AI factory business.

Why is this deal significant beyond India?

It signals that the AI compute buildout, long concentrated in the U.S. and a few other hubs, is spreading globally. Rising demand, power and land constraints, low-latency needs, and data-sovereignty rules are pushing AI infrastructure into new regions like India, reshaping where models are trained and served.

How does this benefit Nvidia?

Every new AI factory anchored on Nvidia's B300 chips extends its hardware and software ecosystem into a new market. As more countries build AI infrastructure on Nvidia's platform, their AI ambitions become more dependent on its technology, compounding the company's competitive advantage globally.

Editor's note — sources: Reuters (via Yahoo Finance); Larsen & Toubro press release; Business Standard.

Share this article: