Madiha Yaqoob
Indian engineering group Larsen and Toubro has secured an order worth up to $1.57 billion to build what it describes as India’s largest single-cluster AI infrastructure deployment. The AI factory, built around roughly 10,000 NVIDIA B300 chips, will be hosted at the company’s Vyoma.AI campus in Chennai and will serve US-based cloud provider Together AI. L&T said the order, placed with its AI infrastructure subsidiary LTN Compute, falls in a range of about 10,000 crore to 15,000 crore rupees.
What Is Being Built
The project is structured as a strategic partnership rather than a straight construction contract. LTN Compute will build and operate the facility, and Together AI will use the capacity to run inference, fine-tuning and training workloads on its cloud platform. NVIDIA B300 accelerators form the compute layer, and L&T has described the deployment as the largest of its kind in India.
The announcement was published by Larsen and Toubro on August 13, 2026 and reported by Business Standard. It marks L&T’s entry into the AI factory business, a category it had not operated in before.
Why India, and Why Now
AI compute has become a location decision as much as a hardware decision. Power availability, land, cooling, construction capacity and regulatory clarity all determine where large clusters can realistically be built, and demand has outrun supply in the established markets. India offers scale on several of those inputs, and a domestic engineering group like L&T brings the construction and power infrastructure capability that a pure software company does not have.
For Together AI, building capacity in India adds geographic diversity to a compute footprint that most AI cloud providers currently concentrate in a handful of regions. It also puts inference capacity closer to a large and growing user base in South Asia, which matters for latency-sensitive workloads.
A New Kind of Infrastructure Contract
What is notable here is the type of company winning the work. L&T is a heavy engineering and construction group, not a technology firm. The fact that an AI compute contract of this size lands with a builder rather than a cloud provider says something about where the bottleneck now sits, which is in physical construction, power connection and thermal management rather than in software.
Several established industrial companies have moved in this direction as AI data centre demand has grown. It creates an unusual situation where the constraint on AI progress is increasingly measured in megawatts and construction timelines rather than in model architecture.
Why It Matters for Readers
For technology professionals in South Asia, this is a meaningful signal about where local opportunity is heading. A facility of this size needs data centre engineers, power and cooling specialists, network operations staff, site reliability engineers and security teams. Those are durable, well-paid roles, and they are being created regionally rather than only in North America and Europe.
For freelancers and business owners, more regional compute capacity generally means better access and more competitive pricing for AI services in the region over time, though that effect is gradual rather than immediate. It also strengthens the case for building AI-dependent products locally, since the infrastructure supporting them is no longer entirely offshore.
The figures here come from L&T’s own disclosure and are stated as a range rather than a fixed contract value, which is normal for projects of this type where final scope is still being settled. No completion date has been published.
