Guide

Fab versus OSAT: the capital ladder priced from India's own approval book

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

India's twelve approved semiconductor projects span a capital range of nearly eight hundred to one, and that ratio explains the sequence in which every country builds this industry.

The ladder, priced from a single source

Comparisons of fab and assembly economics usually run on global averages assembled from consultancies with incompatible definitions. India offers something better: a single government, using one scheme and one set of accounting conventions, has published approved capital and approved capacity for twelve projects spanning every rung of the ladder. That makes the comparisons below unusually clean, because the figures come from one book rather than from twelve press releases.

The figures used throughout this page come from the annexure to a Lok Sabha reply submitted by the Minister of State for Electronics and Information Technology on 1 April 2026 and published by the Press Information Bureau, from the Fiscal Support Agreement of 5 March 2025 for the Dholera fab, and from the Cabinet releases of 12 August 2025 and 5 May 2026. Where this page derives a ratio, the derivation is shown and labelled as this record's arithmetic on published figures. Approved capital is not spent capital, and approved capacity is nameplate rather than achieved, both of which are stated here once and assumed throughout.

The headline ratio is the one in the fact block. Rs 91,526 crore against Rs 117 crore is about 780 to 1, and against the Rs 3,307 crore approved for Kaynes Technology India at Sanand it is about 28 to 1 DURABLE. The Dholera fab alone accounts for about 56 percent of all approved semiconductor investment in India, leaving roughly Rs 72,800 crore spread across the other eleven projects. A programme that looks diversified in project count is heavily concentrated in capital.

What the fab number is actually buying

A wafer fab's capital divides into process equipment, the facility and cleanroom with their utility systems, and everything else, which covers land, engineering, commissioning and working capital. Equipment dominates. Lithography, deposition, etch, implant, planarisation and metrology tools are individually expensive and bought by the hundred, and they are almost entirely imported into India, since no domestic supplier makes front-end process equipment at production grade. Our cost answer works through the global anatomy.

Two consequences matter for reading any Indian fab headline. First, a very large investment figure is mostly an equipment order placed abroad, so the share of the number that lands in the local economy as construction and services is far smaller than the number implies. Second, the tool base is what depreciates, and depreciation is the dominant line in fab operating economics. A fab that runs at low utilisation still depreciates at full rate, which is why utilisation, not price, is the variable fab managers protect.

The building is not cheap either. Fab structures carry vibration-controlled slabs, enormous mechanical plant, cleanroom envelopes, ultrapure water production, gas yards, chemical distribution and abatement. This is where domestic contractors genuinely participate, and it is also why fab construction runs on a longer critical path than a comparable industrial building. The Dholera project's own record illustrates the point: civil work stood at about 50 percent as of mid-2026 with cleanroom fit-out under way, more than two years after ground was broken, with commercial production guided to mid-2028 TARGET.

What an assembly and test plant buys instead

An outsourced assembly and test plant buys a different machine population entirely. Probers and wafer sort equipment, grinders and dicing saws, die attach and wire bonders or flip chip placement, moulding presses, marking systems, and above all automated test equipment. Testers are the capital concentration in the back end, and test time is the operating cost that decides margin. The building is closer to a high-specification industrial facility than to a fab: clean, controlled and serious, but without fab-class vibration isolation, without ultrapure water at fab volumes and without the same gas and chemical infrastructure. Our definition answer covers the process sequence, and the packaging industry file covers the Indian projects as a group.

The consequences are structural rather than incidental. Back-end plants reach revenue in roughly two to three years against five to seven for a greenfield fab. They can be financed without a national programme behind every rupee. Their workforce profile, precision assembly and equipment maintenance with test engineering above it, is learnable at industrial scale in a way that front-end process integration is not. And their ceiling is lower: value captured per wafer in the back end is a fraction of the front end, conventional packaging is price competitive, and a country whose whole position is assembly still imports the wafers.

Capital per unit of output, worked properly

Ratios of capital to capacity are more revealing than headline capital, and India's approval book supports them. The arithmetic below is this record's own, performed on the government's published figures, and it carries an obvious caveat: a packaged discrete, a DRAM module and a processed 300mm wafer are not comparable products, so these ratios measure product mix at least as much as they measure efficiency.

• Dholera fab. Rs 91,526 crore for 50,000 wafer starts a month, which is 600,000 wafers a year, works out to about Rs 15 lakh of approved capital for each wafer of annual capacity.
• Jewar bumping and probe. Rs 3,706 crore for about 20,000 wafers a month, which is 240,000 a year, works out to about Rs 1.5 lakh per wafer of annual capacity, roughly a tenth of the front-end figure for work performed on wafers that arrive already fabricated.
• SiCSem. Rs 2,066 crore for 60,000 silicon carbide wafers a year is about Rs 3.4 lakh per wafer of annual capacity, and that figure includes a packaging line. Compound wafers are smaller and the tool set is different, which is why a fab can exist at this capital scale at all.
• Kaynes. Rs 3,307 crore for more than 6.33 million chips a day, about 2,310 million a year, is roughly Rs 1.4 crore of capital per million units of annual capacity.
• Jagiroad. Rs 27,120 crore for 48 million units a day, about 17,520 million a year, is roughly Rs 1.5 crore per million units of annual capacity, close to the Kaynes ratio despite being eight times the project.
• Micron Sanand. Rs 22,516 crore for about 14 million units a week, roughly 728 million a year, is about Rs 31 crore per million units of annual capacity, some twenty times the Kaynes and Jagiroad ratios. Memory packaging units are not equivalent objects to discrete chips, which is precisely the point about mix.
• CDIL Mohali. Rs 117 crore for about 158.38 million units a year is roughly Rs 0.74 crore per million units, the lowest ratio in the portfolio, because a brownfield expansion inherits its building, utilities, workforce and customer qualifications.

The pattern worth carrying away is that the two greenfield conventional assembly plants land within about 10 percent of each other on capital per unit, which suggests the underlying economics of that rung are stable and knowable, while the front-end and specialty numbers sit in a different world. Suchi Semicon cannot be placed on this scale, because the government published a combined figure of about Rs 3,936 crore for the pair approved on 5 May 2026 without splitting it, and this record does not manufacture splits.

Who actually pays: the subsidy structure underneath every number

None of these figures is a private capital number. Under the India Semiconductor Mission, central fiscal support covers 50 percent of eligible project cost, and states stack their own incentives on top. Micron's Sanand plant is the clearest published example: a total project cost of about 2.75 billion dollars, of which 825 million dollars comes from Micron, with central support at 50 percent of project cost and Gujarat contributing incentives equivalent to 20 percent, for a total outlay reported above Rs 22,500 crore DURABLE. The private share in that structure is under a third.

For the Dholera fab, the Fiscal Support Agreement of 5 March 2025 cites Rs 91,526 crore with central support at 50 percent of eligible cost, disbursed against milestones rather than paid up front DURABLE. The mechanics are covered in the incentives file. The relevant economic point for this page is that subsidy changes the ladder's shape. It compresses the private capital gap between a fab and a packaging plant, since both receive the same headline percentage, but it does not compress the operating gap: depreciation, utilisation risk and yield learning remain entirely the operator's problem once the plant is built. A subsidy makes a fab financeable. It does not make it profitable.

Semicon 2.0, approved by the Union Cabinet on 15 July 2026 with a budget outlay of Rs 1,27,500 crore, extends support across design, manufacturing, research, materials, packaging and talent DURABLE. Whether the next tranche of approvals shifts the portfolio's capital mix toward the front end or deepens the back end is the single most informative thing to watch about India's position on this ladder.

Payback, and the honest asymmetry between the rungs

Back-end plants convert capital to revenue faster and more predictably. Their equipment is redeployable across customers, their qualification cycles are shorter, and their output can be sold to multiple buyers. Their weakness is margin: conventional packaging competes on price against established Asian capacity, and a merchant plant without anchor customers runs the risk of building capacity it cannot fill.

Front-end fabs invert every one of those properties. Capital is enormous and largely irreversible, the tool set is process-specific, and revenue arrives only after yield learning and customer qualification, which is why new fabs characteristically lose money before they earn it. What a fab buys in exchange is durability: a qualified mature-node process with qualified customers can run profitably for a decade or more after its equipment is depreciated, which is the economic engine behind the entire mature-node business. Our mature node file covers that durability argument on its own terms.

The Indian record already shows the timing asymmetry. The India Semiconductor Mission's own site, last updated 19 August 2026, records Micron's Sanand plant in commercial production from February 2026 and Kaynes Semicon from March 2026, both shipping while the Dholera fab is still in cleanroom fit-out DURABLE. Commercial production at Dholera is guided to mid-2028 TARGET. Back end first is not a policy preference. It is what the capital and construction arithmetic produces.

Jobs per rupee, with the definitional warning attached

Employment claims are the least comparable numbers in this subject, because direct, indirect and induced jobs are counted differently by every promoter. With that stated, the published figures give a rough sense of scale. The Cabinet release of 5 May 2026 puts expected employment for the Crystal Matrix and Suchi Semicon pair at 2,230 skilled professionals against a combined investment of more than Rs 3,900 crore, which is about Rs 1.8 crore of approved capital per job TARGET. SiCSem's promoters claimed 5,000 direct and indirect jobs against about Rs 2,066 crore, roughly Rs 0.4 crore per job, on a broader definition REPORTED. Tata Electronics' 2024 communication on the Dholera project spoke of more than 20,000 direct and indirect jobs against Rs 91,526 crore, roughly Rs 4.6 crore per job on that same broad definition REPORTED.

Those three ratios differ by more than a factor of ten, and most of that spread is definitional rather than real. The safe conclusion is narrow and still useful: semiconductor manufacturing is capital-intensive employment by any measure, packaging generates more direct jobs per rupee than wafer fabrication, and anyone comparing employment claims across projects without checking the definitions is comparing nothing.

Why countries start where India started, and how the ladder breaks

Every industrialising economy that has entered semiconductors from a standing start has entered through the back end, and the reason is arithmetic rather than ambition. A packaging plant costs one to two orders of magnitude less than a fab, reaches revenue in half the time, needs utilities an industrial economy already has, and trains a workforce that the front end will later need. India's portfolio follows that pattern precisely: nine of twelve approved units are packaging plants.

The ladder breaks in two directions. A country can stay on the bottom rung permanently, assembling imported dies while capturing a thin slice of value, which is the outcome several assembly economies have lived with for decades. Or a country can build a fab without the surrounding ecosystem, leaving it dependent on foreign back-end capacity and foreign materials while carrying front-end depreciation. India is currently attempting both rungs at once, with nine packaging units, one silicon fab and two compound fabs under a single programme, and the honest read is that this is a deliberate sequencing bet whose result will be visible from about 2028 onward. The capital ledger keeps the approved and the spent in separate columns, which is the only way to score it as it happens.