The answer desk

How much does a semiconductor fab cost?

Published 25 August 2026. Facts verified to 25 August 2026 unless dated otherwise.

A leading edge logic fab now costs tens of billions of dollars. A mature node 300mm fab of the kind being built at Dholera costs in the region of ten billion dollars. The Dholera project carries a headline of Rs 91,000 crore, and the Fiscal Support Agreement of 5 March 2025 cites Rs 91,526 crore, for a 300mm plant designed for up to 50,000 wafers a month across 28 to 110nm technologies. Most of that money is process equipment rather than building, which is standard for the industry, and the India Semiconductor Mission covers half of the eligible project cost.

The short answer, and why the range is so wide

There is no single price for a semiconductor fab, and anyone who quotes one without conditions is quoting an average of things that are not comparable. The honest answer is a range with the variables attached. A mature node 300mm wafer fab of serious capacity sits in the region of ten billion dollars, as the sourced comparisons below show. A leading edge logic fab with extreme ultraviolet lithography sits in the tens of billions. A multi fab campus built over a decade reaches into the hundreds of billions as a programme rather than as a building.

Five variables move the number more than anything else. The process node, because finer geometries need more mask layers, more exposure passes and far more expensive scanners. The capacity, because a fab is a fleet of tools and capacity is bought in units of tool. The geography, because construction labour, land, power, seismic requirements and regulatory timelines differ enormously. Whether the site is greenfield or an extension of an existing campus, since a brownfield expansion inherits utilities, roads, substations and an experienced workforce. And what is counted inside the fence, since some announced figures cover a single fab shell and others cover packaging plants, research facilities and multi phase programmes stretching over twenty years.

That last variable causes more confusion in coverage than the first four combined. Two headline numbers can differ by a factor of ten and both be accurate, because they are counting different things over different periods.

The Dholera anchor, stated precisely

The figure this record uses for the Tata Electronics and Powerchip fab is Rs 91,000 crore, the number attached to the Union Cabinet approval of 29 February 2024, with the Press Information Bureau release on the Fiscal Support Agreement of 5 March 2025 citing Rs 91,526 crore. DURABLE The plant is specified at up to 50,000 wafers a month on 300mm substrates across technologies from 28nm to 110nm. Under the India Semiconductor Mission, the central government supports 50 percent of the eligible project cost, and the state incentive layer sits on top of that; the mechanics are set out in our guide to how fab incentives work and the capital stack is treated separately in the answer on who funds the fab.

Three qualifications belong with the number every time it is used. It is a full project figure covering land preparation, structure, cleanroom, utilities, abatement and process equipment together, not a construction contract and not an annual spend. It is spread across a multi year build and ramp rather than committed at once. And no breakdown between equipment and facility has been published by either party, so how much of it sits in the tool fleet is not something anyone outside the project can state. The commonly used dollar rendering of about 11 billion depends entirely on which exchange rate is applied on which date, and this desk does not silently convert between currencies: figures below appear in the currency in which they were announced.

Where the money actually goes

The most useful thing to understand about fab capital is that it is mostly not a building. Process equipment is the dominant share of total capital cost in a modern wafer fab, with the structure, cleanroom and utility systems making up the remainder and land, engineering, commissioning and working capital accounting for the rest. A headline investment figure is therefore mostly an equipment order, largely placed with suppliers abroad, and regional economic expectations should be calibrated to that fact rather than to the size of the number. Our capital breakdown page treats the split as an engineering subject.

Within the equipment share, the distribution is uneven in a way that matters. Lithography is typically the single largest line, and it is also the line whose cost scales most violently with node: an i-line stepper suitable for 110nm layers and an immersion argon fluoride scanner suitable for 28nm critical layers are separated by a very wide margin in price, and an extreme ultraviolet system is more expensive again. Etch and deposition tools follow, then implant, cleaning, planarisation, and the metrology and inspection fleet, which is easy to underestimate because it produces no wafers and yet determines how fast the fab learns. Our comparison of the two lithography worlds explains why a fab whose published range stops at 28nm avoids the most expensive tools in the catalogue entirely.

The building is cheaper only by comparison. A fab shell carries vibration resistant structures with slab masses and damping unlike ordinary industrial construction, enormous mechanical plant, a cleanroom fit-out, an ultrapure water plant, gas yards, chemical distribution, exhaust abatement and wastewater treatment able to handle fluoride bearing streams. A large share of that scope is where a domestic contracting industry genuinely participates, which is the part of the spend a host region can realistically capture.

A normaliser that makes projects comparable

Comparing headline figures directly is close to meaningless, because a bigger fab should cost more. The standard way to make projects comparable is to divide capital by capacity, giving capital per wafer start per month. The arithmetic below is done openly on published figures and is labelled illustrative, not reported.

Taking the Rs 91,000 crore headline against the stated ceiling of 50,000 wafers a month gives roughly Rs 1.8 crore of capital per monthly wafer start. That figure assumes the entire headline is fab capital expenditure and that the full stated capacity is reached, and neither assumption is established: the headline is a full project figure, and design capacity is a ceiling rather than an opening rate. TARGET Change either assumption and the number moves, which is the point of showing the working rather than the result.

The same arithmetic applied to sourced comparators puts the Dholera figure in a normal band for a mature node 300mm fab rather than an unusual one. That is the useful conclusion, and it survives the imprecision in a way that a single derived number would not.

Sourced comparisons, in the currencies they were announced in

Two readings follow from that table. The first is that the Dholera figure is not an outlier: a mature node 300mm fab at meaningful capacity costs something in the region of ten billion dollars wherever it is built, and the Kumamoto and Dresden projects, both mature and mid range in node terms, land in the same territory. The second is that the very large American numbers are programmes rather than fabs. Comparing a single Indian fab against a multi fab, multi decade campus announcement and concluding that India is investing less is a category error, and it is one that appears constantly in comparative coverage.

Who pays, and why every government pays something

Almost every large fab announced in the last five years has arrived with public support attached, and the instruments differ more than the intent does. India's mechanism covers 50 percent of eligible project cost under the India Semiconductor Mission, disbursed against milestones rather than paid up front. DURABLE Japan's support for the first Kumamoto fab was up to 476 billion yen from its trade ministry. The European Commission cleared 5 billion euro of German support for the Dresden plant under state aid rules in August 2024. The United States awarded 6.6 billion dollars of direct funding to TSMC's Arizona project under the CHIPS and Science Act, alongside loan and tax credit instruments.

The comparison worth drawing is structural rather than moral. A capital subsidy of the Indian type reduces the entry cost of the plant and leaves operating economics untouched, which suits a first fab in a country with no existing base. Tax credits of the American type reward capital deployed over time and favour firms with taxable income. Grants of the European type are tied to specific facility commitments and negotiated conditions. None of them addresses the harder problem, which is that a fab's competitiveness after it opens depends on yield, utilisation and customer qualification, and no subsidy buys those. Our comparison of incentive regimes works through the instruments in detail, including the point that headline programme sizes are not comparable to single project support without saying which is which.

The cost that arrives after the capital

Capital is the entry ticket, not the cost of the business. A fab runs continuously, consuming power, ultrapure water, gases and chemicals, and employing a maintenance heavy organisation around the clock. Depreciation on the equipment base is the dominant line in fab economics, and it does not pause when the fab is idle, which is why utilisation discipline matters more in this industry than in almost any other manufacturing sector. A plant running well below its nameplate carries most of its fixed cost anyway.

Yield sits on top of that. A fab that produces the same wafers with fewer good dies on each one has a higher cost per shipped part with an identical cost structure, and new fabs learn yield rather than starting with it. That learning period is the reason first fabs generally lose money before they make it, and it is why the mid-2028 commercial production guidance for Dholera describes the start of the hard part rather than its conclusion. TARGET

What the rest of India's approved portfolio reveals about the ladder

The clearest way to see why a front end fab costs what it does is to set it beside the other plants approved under the same mission. Assembly, test and packaging facilities take finished wafers and turn them into packaged parts. They need cleanrooms, but of a less demanding class; they need equipment, but of a far lower unit cost; they need water and gases, but at a fraction of the load. The Tata back end plant at Jagiroad in Assam, cleared by the Union Cabinet in February 2024, carries a figure of about Rs 27,000 crore, which is large for a back end facility and still under a third of the fab. DURABLE Smaller specialised units sit further down again: the two further semiconductor units for Gujarat cleared by the Union Cabinet on 5 May 2026 carry a combined investment of about Rs 3,936 crore and about 2,230 projected jobs, per the release issued on that decision. DURABLE

Set those beside Rs 91,000 crore for one front end fab and the capital ladder of this industry is visible in a single line, running from a few thousand crore for a specialised unit, through tens of thousands for a large back end plant, to the better part of a lakh crore for a 300mm wafer fab.

That ratio explains the sequencing choice almost every entrant country makes, including India: start at the back end, where capital intensity is lower, payback is faster and the skills are learnable, and attempt the front end once demand, talent and supply chain are partly in place. The economics of that choice are set out in our fab against assembly comparison, and the confirmed money for every Indian project this desk tracks is kept on the capital ledger with its tier attached.

What is not public about the Dholera number

Three specific gaps should accompany any use of the Rs 91,000 crore figure, and stating them is the difference between a record and a repetition.

• No split between process equipment and facility construction has been published, so the share of the money leaving India as an equipment order is unknown.
• No valued equipment purchase order for Dholera has been made public by any supplier, which means the largest component of the number has no visible evidence trail as of 25 August 2026. REPORTED
• No phasing schedule has been published, so the annual capital profile, which is what actually matters for construction activity and for the disbursement of public support, is not in the public domain.

None of those gaps is unusual for a project at this stage, and none of them is a reason to doubt the headline. They are the reason this desk reports the headline as a full project figure with a date and a source attached, rather than as a construction budget, an annual spend, or a proxy for how much of India's money stays in India.