Inside L&T Vyoma’s Plan To Build India’s AI Infrastructure Giant

Larsen & Toubro (L&T) is consolidating its data centre, cloud and AI infrastructure businesses under newly incorporated subsidiary Vyoma.AI as it bids to build a larger play in India’s fast-growing AI infrastructure market. 

Earlier this week, L&T transferred its data centre and cloud services business to Vyoma for ₹1,400 Cr. Today, the conglomerate said Vyoma’s AI infrastructure subsidiary, LTN Compute, has secured an order from US-based AI cloud platform Together AI to deploy an AI factory at L&T’s Chennai data centre. 

The facility is expected to house 10,000 NVIDIA B300 GPUs, with 8,000 GPUs forming one cluster under its agreement with Together AI and another 2,000 GPUs being deployed as a separate cluster to serve customers. 

The two developments are in line with the conglomerate’s envisioned scale of ambitions for Vyoma, which is looking to move beyond conventional data centre operations into a full-stack digital infrastructure business spanning colocation, cloud and AI factories. 

Speaking to Inc42, Vyoma MD Prashant Chiranjive Jain outlined the company’s strategy to build out this stack as demand for compute, power and connectivity accelerates.

From Data Centres To AI Factories

Being a construction behemoth, L&T’s initial data centre strategy was built around owning and operating colocation infrastructure. The company, however, soon realised that it needed to move up the AI value chain. For this, adding public, private and sovereign cloud services to its offerings seemed the pressing need.

Founded in 2024, Vyoma was initially incorporated as L&T Cloudfiniti — an incubated new-age business subsidiary. It was rebranded as Vyoma.AI a year later as L&T broadened its focus towards AI-enabled digital infrastructure.

Today, Vyoma’s business is built around three pillars — colocation, cloud and AI factories. Its cloud business includes public, private and sovereign cloud offerings, while the company is also developing its own cloud IP, Jain said.

The company has already begun building a sovereign AI infrastructure stack, aimed at giving enterprises a secure environment to deploy CPU and GPU compute without moving sensitive data to public AI platforms.

L&T’s data centre journey began with a 2 MW pilot facility in Panvel, followed by a 30 MW facility in Chennai. The Chennai campus currently has 150 MVA of live power and can support up to 250 MW of capacity.

L&T is now converting a large part of that infrastructure for AI workloads. Of the existing 30 MW facility, around 6 MW will continue to operate as a colocation facility, while the remaining 24 MW is being converted into what the company calls an ‘AI mega factory’.

L&T expects the B300-powered AI factory to go live by November, while continuing to operate NVIDIA H100, H200 and B200 GPUs at its Chennai facility. The company plans to deploy next-generation Vera Rubin and Blackwell platforms at subsequent facilities.

The scale of the planned infrastructure is significant. Jain said the current NVIDIA-linked AI factory represents around 23 MW of AI infrastructure. “The capex investment we have set for the current setup and the expansion is roughly ₹10,000-12,000 Cr earmarked, which includes the existing data centre and kickstarting the expansion,” he said.

The company is also looking beyond Chennai. With land, power and permits already available, L&T is evaluating the next phase of expansion at the campus. 

Meanwhile, L&T has broken ground on a potential second data centre campus in Mahape, Maharashtra. Jain said the company is evaluating two facilities with capacity of around 40 MW each, with the overall campus potentially scaling up to 100 MW. 

L&T is also in discussions with multiple state governments to secure additional land parcels for future data centre development. However, Jain said land availability is not the biggest constraint at present. 

“The constraint is deploying sovereign scale technology and getting it up and running. That is what we are focusing on,” Jain said.

For context, AI data centres require higher power density and more sophisticated cooling and networking infrastructure than conventional facilities. 

For instance, L&T’s existing Chennai facility was initially designed for rack densities of around 12.5 kW. As the company began deploying H100 and H200 GPUs, it had to retrofit parts of the facility to accommodate higher-density AI workloads.

Jain sees the emerging AI infrastructure market as an ecosystem rather than a winner-takes-all contest, with different players operating across the infrastructure, compute and software layers.

“At the bottom is the physical infrastructure layer — land, power, permits, cooling and storage. Above that sits the compute layer, comprising CPUs and GPUs, followed by networking, cybersecurity and other software layers,” he said.

This layered architecture also allows Vyoma to target different customer segments. While Hyperscalers and neoclouds can use its physical infrastructure to deploy their own compute, other customers can check out the underlying cloud or AI stack. Some can also opt for a fully managed solution.

“Some are deploying an asset-light model and would want somebody else to provide the data centre. Somebody would need just the stack, and somebody needs the solution wrapped,” Jain said.

E2E Strategic Investment

L&T is also building out its AI infrastructure business through strategic investments and partnerships. 

In 2024, the company announced an investment of approximately ₹1,407 Cr to acquire a minority stake of up to 21% in listed AI and hyperscale cloud platform provider E2E Networks. To note, Jain sits on E2E’s board as a non-executive, non-independent director. 

The partnership combines L&T’s data centre footprint with E2E’s GPU-powered cloud infrastructure. E2E is also operating the 2,000 B200 GPUs deployed at the Chennai facility. 

Jain said these GPUs are being operated by E2E for client workloads, while the larger 10,000-GPU B300 deployment will be operated by Vyoma as its own AI factory.

L&T’s Larger Pivot To Emerging Tech

Vyoma’s incorporation comes as L&T looks to expand beyond its traditional infrastructure and engineering businesses into technology-led growth areas.

Jain, who has been with L&T since 2024, was initially roped into L&T to work on future technology businesses and is responsible for the innovation fund, data centre business and electronics products and systems (EPS). 

As the data centre and EPS businesses moved out of incubation and received capex approvals, Jain handed over the innovation fund and took charge of the data centre business as MD, while continuing to head EPS.

For Vyoma, the immediate priority is to scale the infrastructure already under development. The longer-term ambition, however, is to combine L&T’s traditional strengths in infrastructure, construction and project execution with the growing demand for compute, cloud and AI.

[Edited by Akshit Pushkarna]

The post Inside L&T Vyoma’s Plan To Build India’s AI Infrastructure Giant appeared first on Inc42 Media.