Guide

Where wafers and fab materials actually come from

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

A fab is a customer before it is a producer, and the first thing it buys is a disc of silicon made by an industry almost nobody can name.

Two industries that share a word

People say a fab makes silicon. It does not. A fab buys silicon, in the form of polished discs made by a completely separate industry with its own factories, its own customers and its own economics, and then prints circuits on them. The two get collapsed into one another constantly in coverage of India's build-out, and the confusion matters, because it inflates what a country is understood to have achieved when it opens a first commercial fab.

The wafer business is a materials business. Its customers are chipmakers, its product is a substrate specified in flatness, resistivity, oxygen content, particle count and crystal defect density, and its competitive advantage is decades of process control that does not transfer by hiring. The chipmaking business is a device business. Its customers are electronics companies, and its advantage is yield on a specific process. The two industries sit next to each other in the chain and almost never merge, which is why a country can build fabs and still import every wafer they consume. The mechanics of how a wafer is grown and finished are covered on our fab desk page on wafers; this guide is about who sells them and what that means for India.

Who makes wafers, and why the list stays short

Polished 300mm wafer supply is one of the most concentrated markets in industrial manufacturing. The names that recur are Shin-Etsu Handotai and SUMCO of Japan, GlobalWafers of Taiwan, Siltronic of Germany and SK Siltron of South Korea, with a further tier of producers serving smaller diameters and specialty grades. This record states that structure qualitatively and prints no share figures, because none has been verified here to the standard this desk applies to numbers.

The list stays short for reasons that are worth spelling out, because they are the same reasons a new entrant cannot appear quickly. Crystal growth is unforgiving: pulling a single crystal ingot more than a metre long with almost no dislocations, at consistent diameter and doping, is a slow thermal process where small deviations produce scrap. Finishing is worse. Flatness and surface quality on a 300mm disc are specified at scales where the whole plant becomes a metrology problem, and the yield learning behind that is accumulated rather than bought.

Then comes qualification. A fab does not treat wafers as a commodity even though they look like one. Substrate properties feed directly into device parameters, so a fab qualifies a wafer supplier line by line, and typically dual-sources deliberately rather than switching opportunistically. Supply is contracted long, often years ahead, which is another reason a new fab in a new geography negotiates from a weak position: it is asking for allocation from suppliers whose capacity was planned around customers with a track record.

What a fab is actually buying: grades, not discs

The word wafer hides a product range. A fab specifies what it needs and pays accordingly, and the difference between grades is the difference between a usable process and a scrapped lot.

The commercial character of this layer is different from equipment in one decisive way. Tools are capital, bought once and depreciated. Materials are consumed every day the line runs, so the money a materials firm earns follows how busy a fab is, not how much it cost to build. Those companies therefore watch the ramp rather than the groundbreaking, and their arrival in a region lags the fab instead of leading it.

How wafers are actually bought

Wafer supply is contracted, not shopped. A fab and a wafer maker negotiate multi-year agreements covering volume bands, specification, delivery cadence and, in tight markets, capacity reservation, because the supplier is being asked to plan crystal-pulling and finishing capacity years in advance against a customer's ramp. Spot purchasing exists at the edges and is where a buyer pays for its own lack of planning.

Two habits follow from that structure. Fabs dual-source almost everything, qualifying at least two suppliers per grade so that a single plant outage or a single quality excursion does not stop production, and they hold specification control tightly, since a substrate change can move device parameters without anything else in the process moving at all. Both habits favour incumbents, and both make a first-time buyer in a new geography an awkward customer: no operating history, no volume record, and a ramp curve that exists only on paper.

There is also a detail about new fabs that rarely gets stated. A plant in start-up consumes a disproportionate share of test and monitor wafers rather than prime wafers, because every tool being installed, calibrated and qualified is run against blanks before it is trusted with product. The first wafer orders a new fab places therefore say more about tool commissioning than about production intent, which is one reason this desk treats early wafer procurement news carefully rather than as a proxy for output.

Why the materials layer concentrates where it does

Purity is the first reason. The contamination limits a fab writes into a chemical specification sit far below anything the wider chemical industry works to, and meeting them is a property of the whole plant rather than of a final filtration step. Reactor metallurgy, the grade of the incoming solvent, the liner of the drum, the transfer line and the habits of the person handling it all show up on the certificate of analysis, which is why a supplier cannot clean up an existing product line and offer it to a fab.

Qualification is the second. Bringing in a different resist or a different slurry is treated as a process change, so it is proven on split lots, watched for defect signatures, checked electrically and then put through reliability testing before anyone releases it to production. That cycle runs for months and consumes engineering time the fab would rather spend on yield, and the party carrying the risk of the experiment is the buyer rather than the seller. The asymmetry is what keeps incumbents in place and newcomers outside.

The third reason is arithmetic. Any single formulation serves a market too small to justify the depth of knowledge behind it, and most of that knowledge was learned by taking apart failures over decades rather than written down anywhere a competitor could read it. High friction and thin revenue per product line keep entrants out more reliably than a trade barrier would, which is why the map of this layer has barely moved in thirty years. Our Fujifilm company file works through the same mechanics from the supplier's side.

What India imports today, and the one materials move on the record

India today imports its wafers and effectively all of its front-end process materials. That statement should be read alongside the approved project list rather than as an accusation: of the twelve manufacturing units approved under the India Semiconductor Mission per its status published through the Press Information Bureau and current to July 2026, one is a silicon fab, two are compound-semiconductor fabs and nine are packaging and test units DURABLE. No wafer plant and no front-end materials plant sits in that portfolio.

One materials move is on the record and it is worth reading with the verb intact. Fujifilm India signed a memorandum of understanding with the Gujarat State Electronics Mission on 30 June 2026 to explore a semiconductor materials production base at Dholera, per fujifilm.com and The Hindu BusinessLine REPORTED. The verb carries the whole story. A global materials company running diligence on a site is real information, because that work costs money and engineering time nobody spends on a project they doubt, and it is also a position that can be abandoned without penalty. The tier stays REPORTED and the word exploratory stays attached to it, and this entry moves only when a site, a company, an investment figure or a construction start appears.

The gases layer is the exception that proves the pattern, because gas cannot be shipped economically at fab volumes and therefore localises first. INOX Air Products began construction in October 2025 of an electronic specialty gas hub at Dholera at about Rs 500 crore, including a 200 tonnes-per-day air separation unit, reported by DeshGujarat on 6 October 2025 and by Indian Chemical News the same month REPORTED. The full picture of that layer sits in our specialty gases guide.

The back end is where the materials question gets more interesting for India, because the demand base is broader than one fab. Assembly and packaging plants consume organic substrates, leadframes, bonding wire, die-attach films, moulding compounds and solder materials continuously, and nine of the twelve approved manufacturing units are packaging and test plants. Those materials are less exotic than front-end chemistry, their qualification cycles are shorter, and several of them are made by industries India already has in adjacent forms. If any part of the materials chain localises this decade, this record expects it to be that part rather than photoresist.

What would make a domestic wafer or materials plant rational

Volume, and then proximity value. A wafer plant is justified by a customer base measured in millions of wafers a year, not by one fab consuming up to 50,000 a month at full ramp DURABLE, and the arithmetic behind wafer size and cost per die sits on our 300mm economics page. A materials plant is justified earlier, because formulated chemistries carry costs that grow with distance. Several of them age, since their properties drift with light and heat and the usable window is short enough to make stock rotation a discipline rather than an afterthought. Many are classified as dangerous goods, which narrows the freight options and complicates every border crossing. And a plant consuming just in time, with no wish to hold a large hazardous inventory on site, is badly served by a customs process whose timing it cannot predict.

One adjacent business deserves more attention than it gets, because it is the plausible first step rather than the ambitious one. Monitor and test wafers can be reclaimed: stripped of films, repolished and returned to service several times before they are scrapped, and reclaim houses exist precisely because a fab in ramp consumes these wafers in quantity and would rather not buy them new. Reclaim is a real manufacturing operation with real cleanliness requirements, it sits close to the customer by nature, and it is an order of magnitude less demanding than growing crystals. A country that wanted a foothold in wafer processing without attempting the wafer industry would start there.

There is also a service argument. Chemistry problems get diagnosed jointly: when defect numbers move, the fab expects the supplier's applications engineers at the microscope alongside its own process team, working the same lots. Distance degrades that arrangement, time zones and visa queues degrade it further, and proximity turns a vendor into something closer to a joint development team. The honest sequence, then, is gases first, formulated chemistries and slurries next if the ramp is real, photomask capability somewhere after that, and wafers last if at all. Anyone promising the reverse order is selling something.

What this desk watches next

The first signal is conversion: whether the Fujifilm exploration becomes a site, a company, an investment figure and a construction start, or quietly expires. The second is any second materials name doing the same diligence, since one exploratory memorandum is an outlier and two is a pattern. The third is whether a substrate, leadframe or moulding-compound supplier commits to an Indian plant, since that is the first materials category where the approved base of packaging units, rather than one fab, carries the demand.

The fourth signal is the least glamorous and the most informative: import composition. When electronic-grade chemicals and wafer imports start rising against a schedule, the fab is buying for a ramp, and that will be visible in trade data before any company announces anything. Against the mid-2028 commercial production guidance given by the electronics and IT minister Ashwini Vaishnaw on 17 July 2026 TARGET, that is the series this desk expects to move first, and the one it will report against the approved capacity of up to 50,000 wafers per month rather than against anybody's ambition.