Company file

L&T Vyoma data centre

Published 3 August 2026. Facts verified to 27 July 2026 unless dated otherwise.

L&T's Vyoma plan is a reported Rs 25,000 crore, 250 MW AI-ready green data centre campus at Dholera SIR with operations around 2028, confirmed via Government of India newsonair.

Status tier: REPORTED

The data centre thesis

Dholera's pitch to data centre capital is simple: land at scale, green power in build, fibre-era planning and a semiconductor anchor next door. The L&T Vyoma MoU is the most credibly sourced of the data centre plans because a government channel carries it. The far larger reported plays, Adani's up to 7.5 GW hub and Tillman Global's Rs 83,000 crore plan, sit further down the confidence scale and are labelled accordingly on the ledger.

The verified record

• Scale: ~Rs 25,000 crore, 250 MW, AI-ready, green power focus. REPORTED
• Timeline: operations around 2028. TARGET
• Confirmation channel: Government of India newsonair. REPORTED at MoU stage.

The market L&T Vyoma is stepping into, in numbers

India's operational data centre stock is somewhere near 1.8 GW of IT capacity, with 258 MW added in the first half of 2026, a 59 per cent year-on-year increase, according to Savills in July 2026. CBRE counted more than 1,700 MW at the end of 2025 after 440 MW of additions during that year, and expected roughly 500 MW to land in 2026. Invest India's figure is more conservative at about 1.5 GW operational. The spread between these numbers is less sloppiness than a definitional problem. Published Indian figures can mean contracted IT load, commissioned capacity that may not yet be leased, or gross facility power including cooling and losses, and the basis is often not stated. Anyone comparing a single project's megawatt figure against a national total needs to know which measure is being used on each side.

The forward projections are steeper than the installed base. Wood Mackenzie models growth from 2.2 GW in 2025 to 12 GW by 2030, a compound rate near 40 per cent, with the AI-dedicated slice moving from 275 MW to 6,546 MW over the same window. On the electricity side the same work puts data centre demand at 10 TWh in 2025 rising to 191 TWh by 2040, close to 7 per cent of national consumption. Knight Frank counts a pipeline of 8.33 GW, of which 322 MW is actually under construction and 2,920 MW is committed. The gap between 8.33 GW of pipeline and 322 MW of steel in the ground is the single most useful statistic in the Indian data centre conversation, and it is the right frame for reading any newly announced campus.

Geographic concentration is extreme. Mumbai holds more than half of operational inventory, with 766.6 MW live on Knight Frank's count, and Mumbai together with Chennai, Delhi-NCR and Bengaluru accounts for roughly 90 per cent of tier-I capacity. The Observer Research Foundation's framing of the imbalance is that India generates about 20 per cent of the world's data while holding 3 to 4 per cent of global data centre capacity. Announced investment intent over five years runs to US$60 to 70 billion on Invest India's tally, and the IndiaAI Mission has allocated more than 38,000 GPUs with a further 20,000 planned.

Why power, not land, now decides where a campus goes

Wood Mackenzie's conclusion is blunt: reliable, cost-competitive power has overtaken land and capital as the defining site-selection factor. The mechanism is straightforward once the cost structure is laid out. Land is a one-off purchase of a few hundred acres and, even at generous valuations, a minor line in a project running to tens of thousands of crores. Electricity is a recurring input for twenty or thirty years, and at hyperscale it is typically the largest single operating cost. A tariff difference of one rupee per unit, sustained, will dwarf any plausible difference in land price between two candidate sites.

The harder constraint is not price but availability and timing. A campus of a few hundred megawatts needs a dedicated extra-high-voltage connection, usually at 220 kV or above, with its own substation, redundant feeders from separate sources so that a single line fault cannot black out the site, and evacuation capacity on the upstream network that was not planned with this load in mind. Grid interconnection studies, transmission licensee approvals and transformer procurement all run on multi-year clocks, and large power transformers and switchgear have been on extended global lead times through the current build cycle. In practice the interconnection queue, not the construction programme, tends to set the earliest possible energisation date.

Renewables complicate rather than solve this. A data centre is a flat round-the-clock load with a load factor typically above 80 per cent. Solar generation is a daytime curve with no output overnight and little at either end of the day. Co-locating a campus beside a solar park does not, by itself, make it solar-powered. It draws from the grid and settles the renewable attribute through open access, banking or a power purchase agreement, and genuine round-the-clock renewable supply requires either storage, a wind and solar hybrid, or firming from the grid, each of which raises the delivered cost above the headline solar tariff. Gujarat's underlying position is nonetheless strong: 69 GW of installed capacity of which 47 GW is renewable. At Dholera specifically, 300 MW of the planned 1,000 MW solar park has been commissioned by Tata Power Solar DURABLE, with the remaining 700 MW under development by ReNew, SJVN, Vena and TEQ Green against a March 2027 target TARGET.

What 250 MW actually means

Set the 250 MW figure reported for the L&T Vyoma agreement REPORTED against the national picture and the scale becomes clear. It is close to a seventh of the entire operational Indian stock of roughly 1.8 GW. It is marginally less than the 258 MW the whole country added in the first six months of 2026. It is about a third of Mumbai's live capacity and roughly 8 to 9 per cent of Knight Frank's committed national pipeline. A single campus of this size, if built out, would rank as a material share of the country's installed base rather than an incremental addition.

Two qualifications matter. First, the announcement does not specify whether 250 MW refers to IT load or to total grid draw REPORTED. The difference is not cosmetic. At a power usage effectiveness of 1.3 to 1.4, a realistic band for a modern facility in a hot climate using efficient cooling, 250 MW of IT load implies something in the region of 325 to 350 MW at the meter. Comparisons against the national figures above, which are IT-capacity based, only hold if the project figure is also IT load. Second, nothing at this scale arrives in one piece. Hyperscale campuses are typically delivered in data halls of roughly 5 to 20 MW, energised in sequence over several years as tenant demand and grid capacity allow. The megawatt number in an announcement is an eventual design capacity, not a commissioning event.

The water arithmetic is worth doing in the same breath, with the caveat that published intensities vary enormously with cooling architecture. On an intensity of roughly 25 to 26 million litres per megawatt per year, a 250 MW facility using evaporative cooling would imply something above 6 billion litres a year, in the order of 17 million litres a day. A commonly cited alternative rule of thumb, that a 100 MW facility draws around 2 million litres a day, would put the same campus nearer 5 million litres a day. The gap between those two answers is itself the point: without a published cooling design, the water figure cannot be pinned down. Dholera currently has a desalination plant of about 20 MLD DURABLE, and a tender was issued on 9 July 2026 to appoint consultants for a 200 MLD seawater desalination plant REPORTED. On either intensity, a campus of this size would be a visible fraction of present local supply, which is one reason closed-loop and air-cooled designs, which trade water for electricity, are attractive for inland sites. Which architecture this project would use has not been published.

Reading the Gujarat data centre policy properly

The Viksit Gujarat Data Centre Policy 2026-29 was announced on 9 July 2026 by Chief Minister Bhupendra Patel REPORTED, with a stated ambition of around Rs 6 lakh crore of investment and 7.5 GW of capacity TARGET. The qualifying threshold is a minimum 150 MW IT load, which is a deliberate filter rather than an administrative detail: it puts enterprise, edge and all but the largest colocation facilities out of reach and writes the policy for hyperscale. The incentive stack comprises a 2.5 per cent capital subsidy for projects located in Dholera SIR specifically, an interest subsidy of up to 4 per cent, a power tariff subsidy of Rs 1 per unit for twenty years, 100 per cent reimbursement of electricity duty for twenty years, and full exemption from stamp duty and registration charges on land DURABLE.

The relative weight of these instruments is not obvious from the list, and the arithmetic is instructive. Take a facility drawing 250 MW at a 90 per cent load factor. That is roughly 1.97 billion units a year, so a subsidy of Rs 1 per unit is worth in the region of Rs 197 crore annually and close to Rs 3,950 crore across twenty years on an undiscounted basis. Against the Rs 25,000 crore investment figure reported for this project REPORTED, the 2.5 per cent capital subsidy would be worth around Rs 625 crore, assuming the whole investment qualifies, which the published policy summary does not confirm. The operating subsidy is therefore several times larger than the capital subsidy in nominal terms, though its present value depends on discount rate and on the campus actually reaching full load. The interest subvention of up to 4 per cent is the other heavyweight, since it applies to the debt portion of a capital-intensive build. The stamp duty exemption, at Gujarat's 4.9 per cent plus 1 per cent registration, is real but small relative to the other three because land is a minor share of a data centre's capital stack.

The competing geographies

Mumbai remains the incumbent and is difficult to displace. It carries more than half the operational inventory, hosts most of the country's subsea cable landings, has the deepest carrier ecosystem, and, on Turner & Townsend's benchmarking, is the second cheapest of 52 global markets to build in at US$6.64 per watt. What it lacks is land and headroom on power and water within reasonable distance of the landing stations. Chennai is the second cable gateway and carries much of the southbound and eastbound traffic. Delhi-NCR and Bengaluru complete the four markets that hold roughly 90 per cent of tier-I capacity between them, which leaves every secondary market sharing the residual tenth.

The genuinely new competition is not from these established metros but from greenfield power-led sites. Google broke ground on a gigawatt-scale AI hub at Visakhapatnam on 28 April 2026, and Tillman Global signed a Rs 15,000 crore memorandum with Andhra Pradesh for 300 MW at the same location. Visakhapatnam is an established port city on the eastern seaboard with existing international connectivity, which is the structural advantage Dholera does not have today. Closer to home, Meta and Reliance announced a 168 MW renewable-powered facility at Jamnagar on 9 June 2026, cooled with desalinated seawater. Jamnagar is the sharpest comparator, because it competes for the same Gujarat renewable power and the same coastal water, but does so inside an existing industrial complex with an operating workforce already in place. Above all of this sit the balance-sheet announcements: Adani at US$100 billion to 2035 targeting 5 GW, Microsoft at US$17.5 billion in India across 2026-29, and AWS at US$8.3 billion in the Mumbai region.

Dholera is not even a single-project location on paper. An AdaniConneX hub of up to 7.5 GW has been reported REPORTED, and a Tillman Global hyperscale plan of roughly Rs 83,000 crore has circulated from a single source and should be treated with real caution REPORTED. Announced gigawatts at this location already exceed, by a wide margin, anything under construction anywhere in India.

The connectivity gap nobody prices in the press release

Intercontinental capacity reaches India through subsea cables that come ashore at a small number of coastal landing stations, historically clustered at Mumbai and Chennai. Every inland campus therefore depends on terrestrial backhaul to reach that international capacity, and the economics of backhaul, not the physics, are what hurt. Pure propagation delay is modest: light in fibre travels at roughly 200,000 km per second, so a few hundred kilometres of route adds only a small number of milliseconds on a round trip. For AI training, which is throughput-bound and relatively tolerant of latency, that penalty is close to irrelevant, and this tolerance is a large part of why power-rich inland sites have become viable for AI workloads when they were never viable for latency-sensitive interactive or financial workloads.

The real handicaps are route diversity, transit cost and carrier neutrality. A mature market like Mumbai offers many physically distinct fibre paths, dense carrier presence and established peering, so bandwidth is cheap and a single cable cut is survivable. A greenfield campus starts with neither. It needs at least two genuinely diverse long-haul routes built or leased, and it pays metro and long-haul transit charges that an incumbent market has already amortised. Gujarat's stated intention to develop two cable landing stations REPORTED would narrow this gap rather than close it, since a landing station still requires onward metro fibre, cross-connect infrastructure and a reason for carriers to establish points of presence. None of the published material specifies where such stations would land, on what timeline, or how a Dholera campus would connect to them.

What has not been published, and what to watch

The L&T Vyoma project is presently a state government agreement for a 250 MW green AI-ready data centre at Dholera with a stated value of Rs 25,000 crore, reported by News on AIR on 20 February 2026 REPORTED. Beyond that, the record is thin. There is no published land parcel or allotment, no phasing schedule, no clarification of whether 250 MW is IT load or gross draw, no named power purchase counterparty or interconnection point, no cooling architecture, no anchor tenant, no engineering contractor and no evidence of financial close. No construction start has been reported. That is not a criticism of the project so much as a description of where it sits on the development curve, which is early.

The signals worth tracking are physical and administrative rather than promotional. Energisation of a dedicated extra-high-voltage substation and the transmission licensee's connection agreement are the first hard evidence that a campus is real, because no developer pays for that infrastructure speculatively. Completion of the remaining 700 MW of the Dholera solar park against its March 2027 target TARGET determines how much of the local renewable claim can be substantiated. Appointment of consultants and then a contractor for the 200 MLD desalination plant REPORTED sets the ceiling on water-cooled capacity in the area. And an announcement of a named hyperscale tenant, rather than a state agreement, is what converts a memorandum into a construction programme.