Guide

Who buys mature node chips: the customer map behind 28 to 110nm capacity

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

A fab does not sell wafers to the public. It sells to a small number of industrial buyers with long qualification cycles, and knowing who they are explains the whole business.

The question this page answers, and the one it refuses to answer

Mature-node manufacturing is usually defended with an argument about volume: most chips in the world are made on older processes, therefore a mature-node fab has a market. The argument is broadly right and almost always delivered with a percentage attached, and the percentage is almost always unsourced. Figures for the mature-node share of global capacity, of unit volume, or of foundry revenue circulate widely, differ from each other by wide margins, depend entirely on where the analyst drew the line between mature and advanced, and are frequently quoted years after the report they came from. This page does not print them. What it prints instead is the customer map: which industries buy chips built at 28 to 110nm, what they buy them for, how they purchase, and what all of that implies for a plant that intends to start selling around mid-2028.

The strategic case for choosing mature nodes at all is made separately, in our guide on why mature nodes. The product categories themselves, what a power management chip or a display driver actually does, are handled on the mature node products page. This page is about the buyers.

Automotive: the anchor customer a shortage created

The automotive industry is the single most important mature-node customer, and the reason is recent history rather than engineering preference. The chip shortage that ran from roughly 2020 into 2023 and idled vehicle assembly lines worldwide was overwhelmingly a shortage of mature-node parts DURABLE. Carmakers were not short of leading-edge processors. They were short of microcontrollers, power management devices, sensor interfaces and discrete components, most of them built on processes that were a decade or more old, and the resulting production losses were large enough to change procurement doctrine across the industry.

What a car consumes is a long list of unglamorous parts: body control modules, engine and transmission controllers, braking and stability systems, lighting, infotainment support silicon, battery management, motor drives, on-board chargers, tyre pressure sensors, and in electric vehicles a substantially higher count of power devices than an internal combustion equivalent. Almost all of this sits comfortably in the mature range. The trend that matters is that semiconductor content per vehicle keeps rising as electrification and driver assistance spread, which means automotive demand for mature-node parts grows even when the number of vehicles sold does not. Our automotive chips page works through the technical side.

Automotive buyers are also the most demanding customers a fab can have. Qualification regimes for automotive silicon require documented reliability testing over wide temperature ranges, production part approval processes, statistical process control evidence, and change-notification discipline that constrains what a fab may alter once a part is qualified. This is a feature for an incumbent and a barrier for a newcomer, and it is the single largest reason a new fab's automotive revenue arrives years after its first working wafer.

Industrial, energy and metering: the buyer with the longest memory

Industrial customers are the quietest large segment and in some ways the best fit for a first fab. Factory automation controllers, motor drives, programmable logic controllers, sensors and their interface circuits, building systems, grid equipment, protection relays and electricity meters all run on mature processes, and the products they go into have service lives measured in decades.

Long product lives create a specific commercial characteristic: industrial buyers hate discontinuation. A supplier that can promise a decade or more of continued availability at a given node has an advantage that no process improvement can offset, because the alternative for the customer is a redesign and a requalification of an entire product family. This is why mature nodes persist commercially long after they stop being interesting technically, and it is why capacity at 90nm and 110nm remains saleable in 2026 rather than obsolete. The persistence is a demand fact, not a sentimental one.

India has a domestic version of this demand that deserves stating carefully. The country buys enormous volumes of electronic goods and the components inside them, and the trade record on those imports is set out elsewhere in this record. A domestic mature-node fab does not automatically capture that demand, because the buyers are the equipment manufacturers who design the parts in, not the end users. But the demand exists in the same country as the plant, which is a structural advantage no amount of policy can create artificially.

Appliances, consumer power and the charger economy

Every device with a power cord or a battery contains power conversion silicon, and power conversion is mature-node territory almost by definition, because the physics of handling voltage favours older, more robust process geometries. Air conditioners, refrigerators, washing machines, fans, lighting drivers, phone chargers, laptop adapters, power banks and every wall wart in a drawer draw on this category.

Consumer power silicon is high volume and price-sensitive, which makes it a difficult segment to earn margin in and an easy one to fill capacity with. New fabs frequently start here for exactly that reason. The parts are less demanding to qualify than automotive, the volumes are large enough to run a line at useful utilisation while a team learns its process, and the customers are willing to second-source aggressively because supply security matters more to them than supplier prestige.

Display drivers and the panel supply chain

Display driver integrated circuits are their own demand pool, and they are worth separating because they are a stated part of the Dholera product set. Every panel, from a car instrument cluster to a television, needs driver silicon between the processor and the pixels, and these parts are made in very high volume on mature processes. The confusion this segment generates in India is that display drivers are chips, not panels, and the two are separate industries with separate economics. Our display driver page handles the distinction, and the approved Indian project in this category is a display-driver plant rather than a panel fab.

The buyer here is the panel maker or the module integrator, concentrated in East Asia, which means a display-driver line in India would in the first instance be selling into an export supply chain rather than a domestic one. That is not a weakness, but it does mean the customer relationships are with a small number of large, sophisticated purchasers who already have qualified suppliers.

The buyer nobody expected: AI infrastructure needs old nodes too

The most interesting recent development in mature-node demand is that the artificial intelligence build-out has become a mature-node customer. This is counterintuitive, because AI accelerators are built at the leading edge and India is not building leading-edge capacity. The parts around the accelerator, however, are not.

TrendForce reported on 30 June 2026 that AI-related applications consuming mature capacity include power management integrated circuits, power discrete devices, interposers, photonic integrated circuits and optical communication transimpedance amplifiers, alongside emerging uses such as silicon bridges REPORTED. The same release stated that strong demand for power ICs on 55nm and larger nodes had tightened high-voltage process capacity at Taiwanese foundries, that average utilisation across the top ten foundries' 8-inch fabs recovered to 88 percent in 2026 with 90 percent expected in the second half of the year, and that 8-inch prices rose by between 5 and 15 percent from the first to the second quarter of 2026, with 12-inch mature-node prices showing signs of increasing by 5 to 10 percent between the second and third quarters REPORTED. These are a single analyst house's figures with a date attached, which is how this record prefers to carry market data, and they should be read as one firm's assessment rather than as settled fact. One clarification matters for the Dholera comparison: 8-inch means 200mm wafers, an older and separate tool base, while Dholera is a 300mm line, so the 12-inch mature-node figures are the closer read-across.

The mechanism is straightforward once stated. A rack of AI servers needs an enormous amount of power delivered at low voltage and high current, and every stage of that delivery chain is built from power semiconductors and management chips at nodes far older than the processors they feed. Optical interconnect between racks needs photonic parts and analog amplifiers that are also not leading-edge. The result is that the most advanced computing build-out in history is pulling on the same capacity pool that makes washing machine controllers.

There is a second-order effect worth watching. TrendForce reported on 22 June 2026 that TSMC had reportedly cut 28nm output by more than a quarter since early 2026 as it shifted resources toward advanced nodes REPORTED. If the largest foundry reduces mature capacity while demand for it firms, the space that opens is precisely the space a new entrant can occupy. That is an argument, not a certainty, and it depends on pricing and on whether Chinese mature-node expansion absorbs the gap first.

How a mature-node customer actually buys

Understanding the purchase process explains why customer announcements lag fab construction by years. The sequence is roughly fixed across segments.

Durations here are the ordinary industry pattern rather than sourced figures for any specific project, and they vary widely by segment and by how much of an existing design can be reused. The important structural point is that the customer commits engineering resources long before revenue appears, which is why buyers prefer processes that already have a proven track record and why a technology partner with an established process design kit is worth so much to a new fab.

Two further habits shape this market. Mature-node customers second-source deliberately, keeping more than one qualified supplier for critical parts, which is good news for an entrant because it means displacing an incumbent entirely is not required. And they buy on availability and continuity as much as on price, because the cost of a line stoppage dwarfs the cost of a chip.

What the demand map implies for the Dholera ramp

Set the map against the plant. The approval record for the Tata Semiconductor Manufacturing project names power management chips, display drivers, microcontrollers and logic within a 28 to 110nm band, at up to 50,000 wafer starts per month on 300mm wafers DURABLE. Every one of those product families sits inside a segment described above, which means the product set was chosen to match the demand pool rather than to chase a headline.

The opening node is genuinely disputed on the record. Bloomberg reported on 17 July 2026 that the plant will open mostly at 90nm, TrendForce carried the same reading on 20 July 2026, and the company's stated range remains 28 to 110nm, with the Tata Sons chairman's letter in the FY25 annual report saying the fab would start at 28nm REPORTED. This record does not resolve that contradiction because nothing public resolves it, and our node investigation lays out both sides. For the customer question the practical effect is that opening at 90nm points the plant firmly at power, industrial and appliance parts before it points at anything else, since 28nm serves a different and somewhat more competitive product mix.

No customer has been announced. As of 25 August 2026 this record has identified no public agreement, letter of intent or capacity reservation naming a buyer of Dholera output, and the absence should be read as normal rather than ominous given that commercial production is guided to mid-2028 TARGET. The customer answer handles that question directly.

What to watch

• The first named customer, and its segment. An automotive tier-one supplier would signal a longer, more valuable qualification path than a consumer power customer, and would also imply a later revenue start.
• Process design kit availability. Customers cannot design in without one. A published kit at a named node is the real starting gun for the design-in cycle.
• Mature-node pricing through 2027. The reported firming during 2026 is helpful to a new entrant. A reversal would compress margins exactly as the plant ramps.
• Competing capacity additions. Mature-node expansion elsewhere, particularly in China, is the main structural risk to the demand argument, and it is a capacity question rather than a technology one.
• Any Indian design-in. A domestic equipment maker designing a part specifically for this fab would be the strongest possible evidence that the demand pool and the plant have actually met.