Company file

ASML and the Dholera fab

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

ASML signed a lithography MoU with Tata Electronics for the Dholera fab, announced 16 May 2026.

Status tier: REPORTED

What is verified

The MoU itself, as reported. No tool orders, delivery schedules or financial terms are public, and this page will not invent them. The right label is REPORTED: real, sourced, announced, not yet delivered hardware.

Why an MoU from a toolmaker is still a signal

Fabs do not run without lithography, and lithography at commercial scale effectively means ASML. Equipment makers engage seriously when a project has money, land and a timeline, because their delivery slots are scarce and contested. An ASML MoU does not guarantee the mid-2028 production guidance, but it removes one of the classic failure modes of announced fabs: having a building and no tools. Read it alongside the wider supply chain forming around the fab.

What ASML actually sells, and why one Dutch company sits on the industry's choke point

ASML does not make chips. It makes the machines that print them: photolithography scanners, which project the image of a circuit pattern held on a quartz photomask down onto a light-sensitive film coated on a silicon wafer, then step and scan that exposure field by field across the wafer, on the order of a hundred fields per 300mm wafer, before moving to the next wafer. The company is best understood as a systems integrator on top of a very narrow supply chain of its own, combining precision optics, mercury lamps or excimer or plasma light sources, and stages that position a 300mm wafer to within a few nanometres while accelerating at several times gravity.

Lithography became the choke point for two structural reasons. The first is physical. Every other major process step, deposition, etch, implant, planarisation, cleaning, is in some sense forgiving: you can trade time for quality, or run a slightly different chemistry. Lithography sets the smallest feature the process can resolve and the accuracy with which each new layer lands on the one beneath it, and once a pattern has been etched into the film neither can be recovered later in the flow. The second is commercial. Deposition and etch have several credible vendors competing on each tool class. Lithography does not. Nikon and Canon still ship exposure tools, largely into older and less demanding segments, but at the leading edge the field is one company deep, and in the mature-node segment ASML holds a very large share of the installed base of workhorse deep-ultraviolet scanners.

Scarcity of that kind converts into allocation power. A lithography vendor decides not only who gets a machine but who gets a delivery slot, which field service engineers are assigned, how quickly a spare part travels, and how much applications support a first-time customer receives. That is why an equipment maker engaging publicly with a new fab operator carries more information than a generic industrial partnership. It also explains why lithography is among the most heavily export-controlled equipment categories in the world: the most advanced immersion systems and everything in the extreme-ultraviolet class sit inside Dutch and allied licensing regimes, while mature-node tools generally do not.

Dry DUV, immersion DUV and EUV: three different machines, three different worlds

Optical resolution follows the Rayleigh relationship: the smallest printable half-pitch scales as a process factor multiplied by the exposure wavelength and divided by the numerical aperture of the lens, while depth of focus, the vertical window in which the image stays sharp, falls with the square of numerical aperture. Every generation of lithography has pushed one of those terms and paid for the damage done to the others.

Dry lithography is the mature workhorse. Mercury-lamp i-line tools at 365 nanometres, strictly near-ultraviolet rather than deep, handle coarse layers. Krypton fluoride excimer lasers at 248 nanometres, the first true deep-ultraviolet class, handle the middle ground. Argon fluoride excimer lasers at 193 nanometres, with air or nitrogen between the final lens element and the wafer, take critical layers into the 90 to 65 nanometre region with heavy help from resolution enhancement: phase-shift masks, off-axis illumination, and optical proximity correction, which deliberately distorts the shape drawn on the mask so the printed shape comes out right.

Immersion is the next step, and it is a plumbing trick rather than a new light source. A thin film of ultrapure water is held between the last lens element and the wafer. Water has a refractive index of about 1.44 at 193 nanometres, which shortens the effective wavelength in the imaging medium to roughly 134 nanometres and, more importantly, allows numerical apertures above one, impossible in air. Immersion scanners are mechanically far more demanding: the water film must move with the wafer at speed without leaving bubbles, watermarks or thermal gradients, and defect control in the fluid becomes a yield issue. In exchange, a single immersion exposure comfortably resolves the critical layers of a 28 nanometre logic process, which is why 28 nanometre is generally regarded as the last node reachable without either multiple patterning or extreme ultraviolet.

Extreme ultraviolet is not an incremental step at all. At 13.5 nanometres everything absorbs, so there is no transmissive optic: the entire imaging path is reflective, built from molybdenum-silicon multilayer mirrors, and the beam path runs in vacuum. The light comes from firing a high-power laser at microscopic tin droplets tens of thousands of times a second to create a plasma. The result is a machine that arrives as a convoy of crates, costs on the order of hundreds of millions of dollars, and draws roughly an order of magnitude more electrical power than a DUV scanner.

Why a 28 to 110 nanometre fab does not need EUV, and what it does need instead

The Dholera fab's official node range is 28, 40, 55, 90 and 110 nanometres DURABLE, with products described on Tata's own foundry material as power management integrated circuits and microcontrollers DURABLE. Every one of those nodes was in high-volume manufacture worldwide before EUV reached production. Buying an EUV scanner for such a fab would be like buying a wind tunnel to design a bicycle.

What the fab does need is a mixed exposure fleet, because a node name describes a design rule set, not a single machine. A 90 nanometre flow might involve twenty-five to thirty separate lithography layers, only a handful of them critical. Gate and contact layers go to the most capable tool available; implant and pad layers can run on much older i-line equipment at a fraction of the cost per exposure. Fab economics reward buying the cheapest tool that holds the specification for each layer and reserving the expensive scanners for the layers that need them.

A rough sense of scale, by arithmetic rather than any published figure: at up to 50,000 wafer starts a month DURABLE and twenty-five to thirty lithography passes per wafer, the line has to deliver well over a million wafer passes through lithography a month. A modern scanner running north of two hundred wafers an hour at high utilisation absorbs a meaningful share of that, older tools far less, so the fleet at full build-out plausibly runs to well over a dozen systems. None of that has been disclosed for Dholera DURABLE, and the split between new and refurbished tools is unknown.

There is one place where the equipment question and the node controversy meet. Reporting in July 2026 indicated the fab will open mostly at 90 nanometres rather than 28, with a Tata spokesperson stating the plan had always been to start at 55 and 90 before introducing 28 REPORTED. In lithography terms that is not a marketing distinction. Opening at 90 and 55 means the critical-layer workhorse can be a dry argon fluoride scanner. Introducing 28 means, in practice, bringing in immersion, with its water handling, its overlay budget in single-digit nanometres, its different resist and mask strategy, and its own qualification programme. Read through the toolset, a staged node plan is a staged capital and capability plan.

What a memorandum of understanding commits, and what it does not

An MoU is a statement of intent. In the ordinary case it records that two parties wish to work together, sketches the areas of cooperation, and expressly leaves binding terms to definitive agreements that may or may not follow. It typically does not fix price, volume, delivery dates, specifications, penalties for non-performance or liability, and it can be abandoned by either side with no disclosure obligation and no financial consequence. This is not a criticism. MoUs are a normal instrument for setting up the diligence a real order requires. They are simply not orders.

The contrast on this same project is instructive. The Intel arrangement announced on 8 December 2025 was described explicitly as exploratory and non-binding REPORTED. The ASML announcement of 16 May 2026 is an MoU REPORTED with no public purchase order, valued contract or delivery schedule attached to it DURABLE. Both sit at the same tier, and the difference between them is one of subject matter rather than legal weight.

Commitment, by contrast, is well defined in this industry. A purchase order with a scheduled delivery slot. A down payment, since vendors typically take a substantial prepayment against long-lead systems. Appearance in a vendor's disclosed backlog or bookings commentary. Letters of credit and shipping documentation. Where relevant, an issued export licence. Site acceptance testing on installed hardware. Each leaves a trace in a filing, a customs record or a regulatory notice.

The 16 May 2026 sourcing, and why a dual newsroom beats a wire pickup

The engagement was announced on 16 May 2026 by both the ASML press release and the Tata Electronics newsroom REPORTED, with wire and outlet coverage following. A single-sourced report tells you what one journalist was told. A statement published by both counterparties on the same day tells you that two communications functions, and in the case of a listed European equipment maker its legal and investor-relations functions too, signed off on the same characterisation. It is the strongest evidence available short of a filing.

The releases named the node range of 28, 40, 55, 90 and 110 nanometres REPORTED and carried a quotation from Randhir Thakur, chief executive and managing director of Tata Electronics, referring to a timely ramp REPORTED. A named executive and a named node range are specific enough to be embarrassing later if wrong, which is itself quality control on the claim.

They are equally informative in what they omit: no tool count, no model designation, no delivery window, no contract value, no description of the service or applications support scope DURABLE. Listed equipment vendors rarely disclose customer-level bookings, so a missing number is not evidence of an empty announcement. It does mean any figure circulating for the value of ASML equipment bound for Dholera has no primary source behind it DURABLE. The same pattern recurs across this project: widely repeated construction claims, including completion percentages and cleanroom progress, appear only on property and search-optimised sites with nothing primary underneath DURABLE.

Lead times, move-in and qualification: the calendar between a signature and a wafer

Suppose an order follows. The sequence from there is long and largely invisible from outside. Advanced scanners are built against allocated slots, and through the last capital cycle lead times for the more capable systems ran to many quarters rather than weeks. The tool then ships as a set of crates, and the receiving fab must already have solved the unglamorous problems: a rigging path sized for the heaviest module, sub-fab space for chillers, vacuum pumps and power distribution, and utilities at specification, meaning ultrapure water, process nitrogen, house vacuum, exhaust and clean dry air.

Vibration is where civil engineering becomes a lithography problem. Exposure tools sit on isolated plinths and demand very low floor vibration and long-term settlement stability, because an overlay budget measured in nanometres cannot tolerate a slab that moves. Seen that way, the engagement of geotechnical firms to redesign the main fab foundations after soil testing found the ground unsuitable REPORTED is not a construction footnote: foundation performance sits upstream of every metrology and exposure specification the fab will later have to hold. The same logic applies to the cleanroom, with a Malaysia-based specialist onboarded to design and install it REPORTED, and to gas supply, where construction of a specialty gas and air separation hub began in October 2025 DURABLE.

After hook-up comes installation and qualification, measured in weeks per tool, ending when the machine prints to specification on test wafers rather than when it is switched on. Then comes the part no vendor can do for the operator: process integration, statistical process control, defect reduction and yield learning across the full flow. That is the knowledge the technology transfer agreement completed on 26 September 2024 DURABLE and the training cohorts sent to PSMC in Taiwan REPORTED are meant to import. Commercial production guidance is mid-2028 TARGET, and the fab has not reported first silicon as of 4 August 2026 DURABLE. Working backwards, tool move-in has to begin substantially earlier, which makes cleanroom readiness the most informative near-term milestone on the project.

Three follow-ups that would harden this from reported to durable

The MoU is real, sourced and announced, and it stays at that tier until specific, checkable things happen. Three would do most of the work.

• A valued order or a disclosed slot. A purchase order, a prepayment, an entry in vendor bookings commentary, an export licence where one is required, or import records showing exposure systems landing in India. A model list would also settle the node argument: dry argon fluoride hardware points at a 90 and 55 nanometre opening, immersion hardware at 28.
• A dated cleanroom certification and first tool move-in. The cleanroom class, the certified area, and a dated first-tool-in event. This is the milestone that converts civil progress into manufacturing progress, and the one most often skipped in announcements because it is the hardest to fake.
• A service and spares footprint inside India. Resident field service engineers, a local spares depot, applications engineering attached to the site. A scanner without local service is hostage to a visa and a flight, and that organisation is a better indicator of how seriously a vendor treats a ramp than any signing ceremony.

Related records: the Tata Electronics file, why mature nodes matter, the Tata and PSMC technology transfer.