Company file

Adani at Dholera

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

Adani-linked Dholera plans: a reported AI data centre hub of up to 7.

Status tier: REPORTED

Two plans, one tier

The data centre number is enormous and, precisely because it is enormous, this record treats it with care: REPORTED, verify before treating a rupee of it as committed. The Embraer MoU is similarly announcement-stage, and its around-2028 horizon depends on an airport that is itself still counting down to first operations, targeted for September to October 2026.

How to read announcement-stage giants

The honest frame is optionality: large groups reserve positions in places they believe will matter. Reserved positions sometimes convert into steel and payroll, and sometimes quietly expire. The capital ledger exists so that conversions and expiries both get recorded, which is the part the brochure economy never does.

How to read a headline number that has not yet been spent

The figures attached to the reported Adani data centre plan at Dholera, a hub of up to 7.5 GW at an indicated outlay near Rs 3 lakh crore REPORTED, are large enough that most readers stop at the adjective and skip the arithmetic. The arithmetic is where the useful information sits. Rs 3 lakh crore spread across 7.5 GW implies roughly Rs 40 crore of capital per megawatt. That is the right order of magnitude for hyperscale construction, though at the low end of it, and it already sits below the construction-only benchmark set out next. A number that is internally consistent has usually been through a concept-level cost model rather than being invented at a press conference, which is worth knowing. It is equally worth knowing that internal consistency proves nothing about funding, sequencing or intent.

Set it against an external benchmark and the picture sharpens. Turner and Townsend put Mumbai data centre construction at US$6.64 per watt, the second lowest among fifty-two global markets it surveyed. Applied literally to 7.5 GW, that implies construction alone of the order of Rs 4 lakh crore at recent exchange rates, before land, before grid connection, before the IT hardware that usually costs more than the building that houses it. So either the announced envelope refers to a different capacity definition, most commonly total connected load rather than critical IT load, or it assumes a materially cheaper build than a Mumbai comparable, or it is a first-phase number wearing a full-programme headline. All three are ordinary in this industry. None of them is disclosed.

The second discipline is to compare the number with the market it would join rather than with itself. India's operational data centre stock is roughly 1.8 GW of IT capacity, with 258 MW added in the first half of 2026, a rise of about three-fifths year on year, on Savills' count. Knight Frank puts the national pipeline at 8.33 GW, of which only 322 MW is actually under construction. A single 7.5 GW campus would therefore be roughly four times everything India has built to date and close to the entire declared national pipeline. That is not a reason to dismiss it. It is a reason to treat it as a ceiling that a programme might grow into over a decade and more, and to watch first-phase sizing instead. The separately reported L and T Vyoma agreement for a 250 MW green data centre at Dholera at about Rs 25,000 crore REPORTED is a more instructive shape: a phase that a single balance sheet can actually reach.

Why power, not land, is what a group like this actually brings

Wood Mackenzie's assessment of the Indian market makes a point that reframes every siting announcement: reliable, cost-competitive power has overtaken land and capital as the defining variable in data centre site selection. Land in India is cheap in absolute terms and abundant outside the metros. Capital, at least for investment-grade sponsors, is available. Firm electricity at a predictable long-run price is neither, and a data centre is among the least forgiving loads on any grid, running near flat at high utilisation, twenty-four hours a day, with no tolerance for interruption.

That is the strategic logic behind an infrastructure conglomerate with generation, transmission and renewable assets entering colocation at scale. It is not selling floor space. It is selling a bundled power position, and it can internalise margin at several points along the chain that a pure-play operator has to buy at market. The group's reported ambition, a group-wide investment programme of around US$100 billion to 2035 with roughly 5 GW of data centre capacity in view REPORTED, is best read as a power-and-digital-infrastructure thesis rather than a real estate one.

The engineering caveat matters more than the strategy. A solar park cannot serve a data centre on its own, because the load is flat and the generation is not. Round-the-clock renewable supply is assembled, not bought: solar oversized against nameplate demand, wind for a complementary diurnal and seasonal profile, storage to shift a few hours, and a grid connection with firm contracted capacity underneath the whole arrangement as the true reliability backstop. Every one of those layers adds cost per unit delivered, and the honest version of a "green data centre" claim is a contracted percentage of matched renewable energy on an annual or hourly basis, not a slogan. At Dholera, 300 MW of the 1,000 MW solar park, built by Tata Power Solar, is commissioned DURABLE, with the remaining 700 MW under development by ReNew, SJVN, Vena and TEQ Green against a March 2027 target TARGET. That is a useful adjacency. It is not, by itself, a supply solution for a gigawatt-class campus.

Two hard physical limits sit under any Dholera data centre plan

The first is water. Published rules of thumb for data centre water use differ by a factor of about three, one commonly cited figure putting a 100 MW facility near 2 million litres a day, another implying closer to 7 million litres a day once you annualise at roughly 25 to 26 million litres per megawatt per year. That spread is not sloppiness. It is the honest signature of cooling design, because an evaporative system trades water for electricity and a closed-loop or air-cooled system does the reverse, spending several percentage points of extra power to spend almost no water. Whichever figure you take, the scale problem is stark: 7.5 GW would imply somewhere between roughly 150 and over 500 million litres a day. Dholera currently has a desalination capacity of about 20 MLD DURABLE, and a tender was issued on 9 July 2026 to appoint consultants for a 200 MLD seawater desalination plant DURABLE, which is a consultancy appointment rather than a plant. The arithmetic tells you that a large campus here would almost certainly have to be designed as a low-water, high-electricity facility from the first drawing, and that this is a design decision with a permanent cost attached.

The second is connectivity. Mumbai holds more than half of India's operational data centre inventory, on CBRE's count, and it holds it for a physical reason: subsea cables land there, and latency to the international internet is measured from the landing station, not from the state boundary. Gujarat's stated plan for two cable landing stations REPORTED is the single most important line item for anyone assessing Dholera's long-run suitability for latency-sensitive workloads. Absent that, a Dholera campus is best suited to training, batch inference and archival workloads, where a few milliseconds cost nothing, and poorly suited to the low-latency edge tier. This distinction, training-class versus latency-class capacity, is the one most announcements omit and the one that most determines which customers can actually be signed.

What final assembly actually means in aerospace

The reported Embraer and Adani memorandum for civil aircraft final assembly at Dholera, indicated around 2028 REPORTED TARGET, deserves precision about the word "assembly", because in aerospace it is a term of art with a narrow scope. A final assembly line receives major structural assemblies that were built elsewhere: fuselage barrel sections, wings, empennage, doors, landing gear, engines and the large systems packages. The line's work is to join them in a defined sequence on hard tooling, mate the wing to the body, install and functionally test the hydraulic, electrical, fuel and avionics systems, fit the interior, paint the aircraft, run ground tests, then fly it, and finally hand it over on a customer acceptance flight.

That is demanding and highly skilled work, and it is a minority of an aircraft's value. The bulk of the money in a modern airliner sits in the engines, the avionics suite and the primary structures, all of which are made by a small global set of suppliers under type certificates that do not move because an assembly hall does. A second final assembly line is therefore, in the first instance, an industrial and market-access instrument: it localises jobs, satisfies domestic content expectations, shortens the delivery chain into a growing market, and creates a base from which supplier localisation can later be negotiated. Depth in the supply chain follows years afterwards, if it follows at all.

Certification is the other reality. The aircraft type certificate stays with the original design organisation. What a new line needs is regulatory approval of the production organisation itself: an approved quality system, controlled tooling and processes, trained and authorised inspection staff, and the extension of the manufacturer's production approval to a second site, with the airworthiness authority of the country of manufacture and the authorities of importing countries all satisfied that an aircraft built at the new line is identical in conformity to one built at the parent line. That process is measured in years, not quarters, and it is entirely invisible from the outside until the first conformity aircraft flies.

Why the aerospace plan cannot move before the airport does

Finished aircraft leave a factory by air. That single fact makes the Dholera assembly proposal strictly dependent on the airport, and it is why the sequencing question is more informative than the announcement. Dholera International Airport Company Ltd, a joint venture of the Airports Authority of India, the Government of Gujarat and NICDIT, is building a runway of 3,200 metres to code 4E DURABLE, which is ample for the regional jet class. Whether it can also take the outsize freighters that move large aerostructures is a separate question that the published scope does not answer, since that depends on wingspan category and pavement strength rather than length alone. The first AAI trial landing took place on 4 June 2026 DURABLE, and the Civil Aviation Minister said on 14 July 2026 that the project was around 80 per cent complete with operations likely in September or October 2026 REPORTED TARGET.

An operating passenger airport is necessary but not sufficient for a final assembly line. Such a line also needs a dedicated production apron with taxiway access from the assembly building, a paint facility, a flight test and delivery hangar, secure fuel supply, calibrated compass and weighing facilities, and customs arrangements that allow major assemblies to enter under bond and leave as an exported aircraft. Nothing in the published Phase 1 scope, a 25,000 square metre terminal, a cargo terminal of 2,500 square metres and an approximate cost of Rs 1,305 crore DURABLE, describes any of that. No aerospace apron, hangar or bonded facility scope for Dholera has been published, and it would be wrong to assume it exists because a memorandum mentions the site. The realistic reading is that airport commissioning is the first gate, a dedicated aerospace precinct is a separate and later project, and the assembly date follows both.

Optionality is not dishonesty, and it is not commitment either

The most useful frame for both the data centre hub and the aircraft line is the financial one: a memorandum of understanding is a cheap option, not a purchase. The premium is small, some executive time, a signing ceremony, a reputational commitment that can be quietly allowed to lapse. What the option buys is real: an early claim on land in a planned industrial zone before allotment tightens, standing in a state incentive regime, a seat in the policy conversation, and the right to scale quickly if demand materialises. Gujarat's data centre policy announced on 9 July 2026 DURABLE makes the calculation sharper still, since it offers a capital subsidy specifically for projects inside Dholera SIR, an interest subsidy, a power tariff subsidy for twenty years, twenty years of electricity duty reimbursement and full stamp duty exemption on land, with a minimum 150 MW IT load to qualify. Announcing early is how a group positions itself inside a regime like that. It is rational corporate behaviour and it is not deception. It simply is not the same thing as a decision to build.

The distinction is observable, and the reader does not need inside information to track it. Commitment leaves a paper trail in a fixed order: a land allotment or registered lease deed for a specific plot area, environmental clearance, a grid connectivity agreement with contracted capacity in megawatts, financial close with named lenders, an award to a main civil or EPC contractor, orders for long-lead equipment such as transformers, chillers and switchgear, and, for a data centre, an anchor customer contract that turns the campus from speculative to underwritten. For the aerospace line, the equivalent markers are a definitive agreement replacing the memorandum, a production organisation approval process opened with the regulator, and a published apron and hangar scope at the airport.

Against that checklist, the honest position on both Adani proposals as of August 2026 is that they sit at the first stage. The public record for Dholera as a whole is more advanced than the record for these two specific plans: the NICDC delivery report of 31 May 2026 records twelve plots allotted, of which 436 acres are industrial, with Tata Chemicals named as anchor investor DURABLE, trunk infrastructure complete across the 22.5 square kilometre Phase I activation area DURABLE, and the Ahmedabad to Dholera expressway inaugurated on 31 March 2026 DURABLE. The allotment register, not the announcement calendar, is the scoreboard worth reading. When a plot at Dholera is allotted to a named Adani entity for a stated area and use, the option will have started to convert. Until then, the correct description of a 7.5 GW hub REPORTED and a 2028 aircraft line REPORTED TARGET is that they are announcement-stage possibilities held open by a group that can afford to hold many of them at once.