Briefing

The Rs 2.5 lakh crore question

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

A round number is doing heavy lifting in a thousand sales pitches. We went looking for its ledger.

Where the big number comes from

Add every announcement touching Dholera, the reported Adani data centre hub, Tillman's plan, L&T Vyoma, Tsingshan, Jabil, the solo Vedanta claim, slices of Reliance's state-wide MoU, and round numbers of that size assemble easily. Our own ledger's REPORTED rows alone sum past four lakh crore of stated intent. The arithmetic is real. The category error is the word committed.

The three-bucket honest version

• Confirmed and building: Rs 91,000 crore (the fab) plus the operational solar and the cleared Cabinet additions above Rs 3,900 crore DURABLE.
• Government-channel MoUs: plans carried by official channels, like L&T Vyoma via newsonair, real engagements at MoU stage REPORTED.
• Press-sourced announcements: everything else, including the largest numbers, some single-sourced and flagged as such REPORTED.

Why the distinction is the whole game

Cities are financed by bucket one, marketed on bucket three. Collapsing the buckets into one number is how buyers get sold certainty that does not exist yet, and it is also unnecessary: Dholera's honest bucket one, a Rs 91,000 crore fab physically rising plus an open expressway plus live solar, is already the strongest confirmed story any Indian greenfield city can tell. The truth here does not need inflation.

Rebuilding the number, row by row

The only honest way to interrogate an aggregate is to take it apart and see which rows survive contact with a primary document. Build the ledger from the bottom up, admitting only items where an instrument exists: a Cabinet approval, a signed agreement, a budget line, a commissioned asset. On that basis the Dholera ledger as of August 2026 looks like this.

• Tata Electronics and PSMC wafer fab, Rs 91,000 crore, Union Cabinet approval 29 February 2024 DURABLE, Fiscal Support Agreement signed 5 March 2025 DURABLE.
• Ahmedabad (Sarkhej) to Dholera semi high-speed double line, Rs 20,667 crore, CCEA approval 13 May 2026 DURABLE, target completion 2030-31 TARGET.
• Dholera International Airport Phase 1, approximately Rs 1,305 crore, CCEA approved June 2022 DURABLE.
• INOX Air Products electronic specialty gas hub, Rs 500 crore, construction commenced October 2025 DURABLE.
• NICDIT allocation under Budget Estimates 2026-27, Rs 3,000 crore DURABLE, though this is a national corridor allocation and not a Dholera-only figure.
• Gujarat Budget 2025-26 Dholera line items: social infrastructure Rs 170 crore, fire station Rs 25 crore, trunk infrastructure augmentation Rs 20 crore DURABLE.

Those rows total roughly Rs 1.17 lakh crore, and even that figure is generous, because two of the largest entries are not Dholera-specific in any strict sense. A railway line is a linear asset that serves everything along its alignment and is justified on corridor economics rather than on one industrial estate. The NICDIT allocation covers the national industrial corridor programme. Strip both out and the site-bound committed capital falls to something in the region of Rs 93,000 crore, of which the fab is the overwhelming majority. The Crystal Matrix compound semiconductor unit approved by the Union Cabinet on 5 May 2026 DURABLE belongs in the ledger by instrument, but no capital figure for it sits in the verified record, so it is entered at zero rather than estimated. That is the correct treatment. An unpublished number is not a small number, but it is not your number to invent either.

What the Rs 91,000 crore line actually contains

The single largest row deserves to be opened up, because it behaves differently from the way a reader instinctively assumes. First, it is not private capital. The Government of India committed fiscal support of 50 per cent under the Fiscal Support Agreement of 5 March 2025 DURABLE, and Gujarat's Semiconductor Policy 2022-27 adds 40 per cent of the centrally approved capex assistance on top DURABLE. The headline therefore describes total project cost, not promoter outlay, and a substantial share of it is a public transfer, ordinarily disbursed against milestones rather than sitting as a lump sum in an escrow account. Announcement aggregates almost never disclose this split, which means a reader adding headline numbers is silently adding subsidy to investment and counting the same rupee under two labels.

Second, the internal composition of any wafer fab budget is heavily weighted towards tools. Across the industry, process equipment routinely accounts for the majority of greenfield fab capital cost, commonly cited at somewhere around three quarters, with the building shell, cleanroom, and facilities systems taking the remainder. That ratio matters here for a specific reason. The publicly stated node range is 28, 40, 55, 90 and 110 nanometres DURABLE, and Bloomberg reported on 17 July 2026 that the fab will open mostly at 90nm rather than 28nm REPORTED, a sequencing Tata described as always intended and PSMC's Eric Tang framed as the normal pattern for technology transfer REPORTED. That account sits awkwardly beside the Tata Sons chairman's own statement, in the annual report for the year ended March 2025, that the company had chosen to start its chip journey at the 28nm node REPORTED, and the tension between the two is itself worth noting. Lithography, metrology and etch tool sets for 90nm are materially cheaper than for 28nm. A fab that ramps on mature nodes first spends its equipment budget later and differently from one that starts at the tightest node it has licensed. So the Rs 91,000 crore figure is not a single dated commitment at all. It is a spending envelope whose drawdown schedule shifts with node sequencing, and mid-2028 commercial production guidance TARGET pushes the heaviest tool spending well past the point at which the headline was first published in February 2024.

Third, the observable evidence of that spending is thin in the public record. There is no publicly confirmed main EPC or civil contractor, and no publicly valued equipment purchase order DURABLE. What is confirmed is that Fugro, Cengrs and Geo Dynamics were engaged in late 2025 to redesign the main fab foundations after soil testing found the ground unsuitable REPORTED, and that Malaysia's IAQ Group was onboarded to design and install the cleanroom REPORTED. Those are real engineering events. They are also early-stage ones.

The announcement tier, and why it dwarfs the committed tier

Alongside the instrument-backed ledger sits a second, larger stack of numbers carrying no instrument at all. Adani's reported AI-ready data centre hub of up to 7.5 GW is discussed at roughly Rs 3 lakh crore REPORTED. Tillman Global's hyperscale plan is reported at about Rs 83,000 crore REPORTED from a single source, which is a serious weakness, and the same group separately signed a Rs 15,000 crore memorandum with Andhra Pradesh for capacity at Visakhapatnam REPORTED. L&T Vyoma's 250 MW green data centre is put at Rs 25,000 crore REPORTED. Tsingshan's steel and EV battery proposal is reported at about Rs 21,000 crore REPORTED and Jabil's silicon photonics memorandum at about Rs 1,000 crore REPORTED.

Those five items alone come to roughly Rs 4.3 lakh crore, comfortably more than the round aggregate figures that circulate for Dholera and nearly four times the committed ledger. This is the arithmetic that should unsettle anyone quoting a single headline sum. There is no defensible construction that lands precisely on any of the round figures in circulation. You reach one only by taking all of the committed tier, some of the announcement tier, and stopping. Which announcements you admit, and which you leave out, is a discretionary act that the final figure conceals entirely. A number that can be assembled several different ways from the same evidence is a range wearing the costume of a measurement.

Twelve plots, and what the delivery report is really saying

Set that against the government's own reporting. The NICDC and DPIIT Delivery Monitoring Unit report dated 31 May 2026 records 12 plots allotted at Dholera, of which 436 acres are industrial DURABLE, with Tata Chemicals named as anchor investor DURABLE. It states plainly that industrial and commercial or residential land remains readily available for allotment DURABLE. That last clause is the tell. It is written as a supply assurance, and it functions as a demand disclosure.

Convert the units and the contrast sharpens. 436 acres is about 1.76 square kilometres. The Phase I Activation Area as described in the same report is 22.5 square kilometres with trunk infrastructure works complete DURABLE, so allotted industrial land is roughly eight per cent of the serviced area, and about two tenths of one per cent of the 920 square kilometre DSIR DURABLE. The SEZ notified for Tata Semiconductor Manufacturing on or around 16 April 2026 covers 66.16 hectares DURABLE, which is roughly 163 acres, or something close to 38 per cent of all industrial land allotted at Dholera to date. One project accounts for well over a third of the take-up. Broaden the frame and the picture holds: PIB reporting on the Economic Survey 2025-26 records 350 industrial plots collectively allotted across all Phase-I corridor cities DURABLE, of which Dholera's 12 is about three per cent.

Two caveats keep this honest. The delivery report's 22.5 square kilometre activation figure sits alongside the official DSIRDA and DICDL portal describing Phase I activation of 153 square kilometres across TP1 and TP2 DURABLE, with construction initiated on 2,250 hectares DURABLE. The 2,250 hectare figure converts to exactly 22.5 square kilometres, which suggests the larger number is a planning boundary and the smaller one the built footprint, though the sources do not say so explicitly. And plot count is a crude metric: one 163 acre allotment to a fab operator carries more industrial weight than fifty small parcels. The point is not that 12 is a failing grade. It is that the announcement ledger and the allotment ledger are measuring different things, and only one of them requires a signature on a land document.

How cumulative figures are constructed, and why they drift upward

Aggregate investment figures are built by accretion, and the mechanism has four systematic biases, none of which requires anyone to lie.

The first is that announcements are added but never retired. When the Vedanta and Foxconn joint venture collapsed in 2023 after Foxconn withdrew REPORTED, no widely quoted aggregate was publicly revised downward to remove it. Cumulative totals have no journal entry for reversal, so failure is invisible in the running sum while success and mere intent both register at full face value.

The second is geographic drift. Reliance's Gujarat-wide renewables and green hydrogen commitment of roughly Rs 5.95 lakh crore REPORTED is a state-level figure covering an entire energy programme, and the Viksit Gujarat Data Centre Policy 2026-29 sets a statewide target of about Rs 6 lakh crore against 7.5 GW TARGET. Both are legitimate numbers about Gujarat. Neither is a Dholera number. Aggregates assembled by keyword search rather than by site verification pull state and national figures into district-level totals routinely.

The third is nominal accumulation across time. A total spanning announcements from 2022 to 2026 adds rupees of different vintages without deflation, and spreads an implied spending horizon that runs well into the 2030s across a headline that reads as though it describes the present. The rail line targets 2030-31 TARGET. The 700 MW balance of the solar park targets March 2027 TARGET. Commercial fab production is guided to mid-2028 TARGET. A single sum flattens a decade of staged capital into one apparently current stock.

The fourth is double counting between capex and incentive. State policy commitments, land subsidies at 75 per cent on the first 200 acres for a fab in Dholera Semicon City, power support at Rs 2 per unit for ten years DURABLE, and central fiscal support are frequently tallied alongside the project costs they defray rather than netted against them.

A reusable method

The test that survives all of this is boring and repeatable. For any figure attached to any industrial site, ask five questions in order. What instrument backs it, ranked from commissioned asset, to signed agreement or Cabinet approval, to budget allocation, to memorandum, to statement of intent? Who bears the cost, and how much of the headline is public support rather than promoter capital? Over what period is it spent, and in which year's rupees? Where does the land come from, and does an allotment, lease or notification exist for it? And is there a counterparty on the other side, a contractor appointed, an order placed, a payment recorded?

The land question is the sharpest of the five, because industrial capital cannot be deployed on paper. A gigawatt-scale data centre campus needs hundreds of acres, grid interconnection and, given that roughly a megawatt of IT load consumes on the order of 25 million litres of water a year under conventional cooling REPORTED, a settled water source. Dholera currently operates about 20 MLD of desalination capacity, with a tender issued on 9 July 2026 to appoint consultants for a 200 MLD seawater plant DURABLE, which is the consulting stage of a plant, not the plant. So when an announcement tier of Rs 4.3 lakh crore sits above an allotment record of 436 industrial acres, the constraint is not scepticism about intent. It is that the physical prerequisites for the announced capital have observable, datable milestones, and most of them have not happened yet.

Applied consistently, the method does not produce a smaller number so much as a differently shaped one: a firm figure of roughly Rs 1.17 lakh crore with instruments behind every row, a much larger conditional figure with none, and a set of named events that would convert the second into the first. A published EPC award for the fab. A valued equipment purchase order. A land allotment to a data centre operator. Those are the entries worth waiting for, and each of them, when it lands, will be a document rather than a headline.