Larsen & Toubro has secured a contract worth between ₹10,000 crore and ₹15,000 crore — roughly $1.05 billion to $1.57 billion — to build India's largest single-cluster artificial intelligence infrastructure for Together AI, the San Francisco-based open-source AI cloud provider. The deal, executed through L&T's data centre subsidiary Vyoma.AI and its compute arm LTN Compute, will see 10,000 Nvidia B300 'Blackwell Ultra' GPUs deployed in a single NVLink-connected cluster at L&T's Chennai campus.
The distinction that separates this deal from India's prior AI infrastructure announcements is architectural rather than purely financial: this is not a distributed deployment spread across independently networked servers, but a unified single-cluster of 10,000 GPUs joined by a shared NVLink fabric — effectively making it one very large machine rather than many smaller ones working in parallel. That design choice matters enormously for the kind of large-scale AI model training and inference workloads Together AI intends to run on it.
Vyoma's Chennai data centre campus is designed as a gigawatt-scale AI infrastructure site, with Phase 1 built for 250 megawatts of capacity and power infrastructure readiness of 150 MVA. The integrated AI Factory combines hyperscale data centre infrastructure, accelerated computing, high-performance networking, ultra-low-latency interconnects, high-throughput parallel storage and AI infrastructure operations under a single roof — a full-stack approach that L&T has positioned as central to what it calls its Gigawatt AI Infrastructure Mission.
L&T Chairman and Managing Director S N Subrahmanyan said the deployment reinforces the company's commitment to making India a global hub for next-generation AI infrastructure, framing the order as a milestone within L&T's broader ambitions in the space. Those ambitions extend well beyond this single contract: L&T has separately outlined plans to scale Nvidia GPU clusters at its Chennai campus up to 30 megawatts, alongside a new 40 MW facility under development in Mumbai, positioning the conglomerate as one of the anchor infrastructure players in India's rapidly forming sovereign AI computing stack.
At the rack scale, Nvidia's GB300 NVL72 systems — which pack 72 Blackwell Ultra GPUs into a single liquid-cooled rack — deliver dramatically higher throughput per megawatt and substantially lower cost per token compared to Nvidia's previous-generation Hopper H100 platform for agentic AI workloads, according to data cited by Nvidia. Each individual B300 GPU operates at a thermal design power of 1,400 watts, nearly double that of the H100, underscoring both the computational intensity of the new hardware and the scale of the cooling and power infrastructure required to run it reliably at the cluster level Vyoma is targeting.
The Together AI order lands amid a broader wave of GPU infrastructure investment across India, anchored by the government's IndiaAI Mission. Nvidia has positioned itself as central to that push, supplying GPUs for sovereign data centres through partners that include Yotta — which has committed $2 billion to deploy more than 20,000 Blackwell Ultra GPUs at its Greater Noida campus — alongside L&T and cloud provider E2E Networks, with India targeting more than 100,000 GPUs deployed nationally by the end of 2026. Nvidia is also working with Indian government-backed initiatives such as BharatGen, a 17-billion-parameter model targeting agriculture, public services and cultural preservation, on sovereign language models tailored to local languages and regulatory priorities.
For Together AI, the Chennai deployment extends a company that had already crossed $1.15 billion in bookings earlier this year, giving it dedicated large-scale compute capacity in one of the world's fastest-growing AI markets outside the United States and China. For India, the deal represents a concrete example of the country's push to build domestic computing capacity capable of supporting advanced AI training and inference — a capability policymakers and technology companies alike have increasingly argued the country needs if it wants to participate meaningfully in the global AI buildout rather than remain solely a consumer of foreign-hosted AI services.
The Vyoma.AI order is likely to be closely watched as a bellwether for how India's engineering conglomerates — traditionally associated with construction, defence and heavy industry — are repositioning themselves as infrastructure providers for the AI economy. With L&T now running parallel gigawatt-scale build-outs in Chennai and Mumbai, and international AI labs increasingly looking to India for compute capacity outside the congested US and China supply chains, this single-cluster deployment may prove to be the first of several similarly structured deals in the country's AI infrastructure pipeline over the coming year.
