Briefing

What the ASML MoU actually signals

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

MoUs are the confetti of industrial policy. This one deserves a closer look, because lithography is the one input a fab cannot improvise.

The base rate on MoUs

Most industrial MoUs are optional futures: cheap to sign, cheap to abandon. A record like ours tags them REPORTED and moves on. The reason this one clears the noise floor is the counterparty. ASML does not need announcement publicity, its tools are supply-constrained, and its engagement calendar is a scarce resource. Toolmakers of that class spend serious diligence only where money, land and timeline already exist.

What it does not prove

An MoU is not an order book. It does not prove tool allocations, delivery slots or the mid-2028 production guidance. Anyone presenting it as a guarantee is selling something, which on this subject is usually literal.

Three follow-ups that would harden it

• A confirmed purchase order or tool delivery schedule from either company.
• Cleanroom readiness signals from the site consistent with tool installation windows.
• Hiring patterns for lithography process engineers attached to the Dholera ramp.

Any of the three moves the fab's equipment layer toward DURABLE on the ledger. Until then the record stays exactly as sourced: a world-tier toolmaker, formally at the table, nothing more claimed.

What ASML actually sells, and why the tool list matters more than the headline

ASML is not a general-purpose equipment vendor in the way that phrase is usually understood. It builds photolithography scanners, the machines that project a circuit pattern from a reticle onto a resist-coated wafer, and it sells alongside them the light sources, the optical metrology used to measure overlay and focus, and the computational lithography software that corrects for the fact that the printed shape is never quite the drawn shape. A fab buys deposition, etch, implant, clean and planarisation tools from several suppliers who compete on price and cycle time. Lithography, especially at the higher-resolution end, has no comparable field of competitors, which is why a lithography relationship reads differently from a general procurement note.

Two structural facts follow. The first is financial: lithography is typically the single largest line in a fab's equipment budget, commonly on the order of a fifth to a quarter of total wafer fab equipment spend at mature nodes, and equipment is usually the majority of a fab's total capital cost. The Rs 91,000 crore headline for the Dholera project DURABLE is a whole-project figure rather than an equipment budget, and on the usual convention for announcements of this kind it would cover land preparation, shell, cleanroom, utilities, abatement and tools together. No breakdown has been published, so how much of it sits in the scanner fleet is not something anyone outside the project can state, and no valued equipment purchase order has been published by either party. The second fact is operational: every patterned layer must queue for a scanner, so litho capacity, not etch or deposition capacity, usually sets the throughput ceiling of a wafer line. Buying too few scanners caps the fab below its nameplate. Buying too many strands capital in machines that idle. This is why the fleet plan tends to be settled early, and why a memorandum at this stage is chronologically unremarkable rather than premature.

Reading the node list backwards: what 110nm to 28nm demands of a scanner fleet

The nodes named in the announcement, 28, 40, 55, 90 and 110nm DURABLE, map onto three or four distinct exposure technologies rather than one. The coarser layers at 110 and 90nm can still be printed with i-line steppers at 365nm, the oldest and cheapest platforms in the industry, available from more than one vendor and frequently sold refurbished. Tighter layers at 90 and 55nm sit in krypton fluoride territory at 248nm. The 40nm generation leans on dry argon fluoride at 193nm. Only at 28nm does immersion normally become necessary, where a water film between the final lens element and the wafer raises the effective numerical aperture and pushes single-exposure resolution far enough that the critical layers of a 28nm-class process can be printed in one pass instead of by double patterning.

The most consequential thing about that list is what is absent from it. Nothing in the 110nm to 28nm range requires extreme ultraviolet. EUV is the part of ASML's catalogue that is genuinely capacity-rationed, priced several times above an immersion scanner, and entangled in export licensing. A fab planned entirely within the deep ultraviolet range sidesteps the industry's hardest supply constraint. It also sits outside the Netherlands-and-United-States control regime that has restricted sales of advanced immersion systems to China specifically; India is not the subject of those measures, though Dutch export licensing still applies to particular system classes as a matter of ordinary process. In supply terms, this is a comparatively easy fleet to assemble. That is a point in the project's favour and it is rarely made, because the coverage tends to treat any ASML mention as a proxy for leading-edge capability, which the node list plainly does not claim.

The reported shift in opening node also changes the phasing rather than the shopping list. If the fab opens mostly at 90nm with 55nm alongside and 28nm introduced later REPORTED, the immersion tools, which are the expensive end, are needed considerably later than the i-line and KrF tools. Deferring the costliest scanners until the process is proven is conventional capital discipline for a first-time fab operator, and it is consistent with a technology transfer that starts at mature nodes.

An order-of-magnitude scanner count

No tool count has been published, so the only honest way to size the fleet is to build the arithmetic openly and label it as illustrative rather than reported. Take the stated ceiling of up to 50,000 wafer starts per month TARGET. A mature logic or power management process in this node range typically involves somewhere in the region of twenty to twenty-five lithography exposures per wafer, fewer at 110nm, more at 28nm. At twenty-two exposures, full loading implies roughly 1.1 million wafer passes through litho per month. A modern scanner in this class runs at a couple of hundred wafers per hour when it is running, and no fab achieves continuous operation once preventive maintenance, reticle changes, lot gaps and qualification time are counted, so eighty per cent effective utilisation is a generous planning figure. On those assumptions the arithmetic lands at roughly ten scanners as a bare floor, and a realistic fleet would be meaningfully larger, because tools are not interchangeable across exposure technologies and a mixed product portfolio fragments capacity further. Change any one of the assumptions and the number moves, which is the point: this is a sanity check on scale, not a reported figure.

The number matters less than what it implies. A fleet of that scale is not a single transaction. It is a multi-year procurement sequence with different tool classes arriving at different points, and it carries an installed-base relationship behind it: field service engineers resident at the site, a spares inventory held locally, applications support during process transfer, and software licences that renew. The after-sale relationship is usually where the enduring commercial content sits, and it is the part that no memorandum describes.

Scarcity, lead times and the reason to sign early

Lithography scanners are built to order in small annual volumes from a supply chain with severe single-source concentration, most visibly in precision optics. Historically, lead times for advanced systems have been measured in quarters and, during the capacity crunch of the early 2020s, were widely reported to have stretched towards years, with order books extending well beyond the current build slate. The mature-node end has had its own scarcity: the global build-out of 28nm-and-above capacity absorbed new i-line and KrF systems and much of the secondary market in used tools, so the refurbished route that once offered a cheap and fast alternative has been thinner than it was.

Against that background, a memorandum signed years before tool move-in is not evidence of vapour. It is what a rational buyer does to establish a named relationship, begin technical alignment on process and tool configuration, and position itself in the queue before the specifications are frozen. There is a further consideration specific to a technology transfer. Process recipes are tuned to particular tool models, illumination settings and metrology feedback loops. A recipient fab that assembles a scanner fleet closely matching the donor's fleet inherits a working process; one that substitutes different platforms inherits a re-qualification project. Given that the PSMC technology transfer agreement completed on 26 September 2024 DURABLE, aligning the litho fleet with the donor's is the low-risk choice, and an early supplier relationship is how that alignment gets made.

The ladder from memorandum to qualified tool

The distance between an announcement and a wafer is a sequence of discrete, unequally visible steps. A memorandum of understanding records intent and typically carries no consideration, no volumes, no delivery dates and no remedy for non-performance. Below it sits a definitive supply agreement with commercial terms. Below that, a purchase order, usually with a down payment, at which point the supplier books it as an order. Then a manufacturing slot, then shipment, at which point the supplier generally recognises revenue. Then the physical work: crating, transport, uncrating in a certified environment, mechanical install and hook-up to utilities, then tool acceptance against contractual specifications for overlay, focus and throughput. Only afterwards comes process qualification, where the fab demonstrates it can print its own layers on that tool, followed by yield learning and finally customer product qualification, which is what turns a functioning fab into a shipping supplier.

The physical steps are gated by the building, which is why apparently unglamorous facts carry weight. A scanner requires a floor meeting stringent vibration criteria, temperature control to fractions of a degree, tightly held humidity, airborne molecular contamination control, ultrapure water, bulk and specialty gases, and a subfab able to take the load. The reported engagement of geotechnical firms to redesign the main fab foundations after soil testing found the ground unsuitable REPORTED, and the reported onboarding of a cleanroom designer and installer REPORTED, are therefore litho-relevant news, not merely construction news. The specialty gas hub reported to have begun construction in October 2025 REPORTED belongs to the same dependency chain. No tool qualifies in a building that is not finished, and none of these items has a published completion date.

What industrial memoranda are worth as evidence

The honest position on conversion rates is that no reliable universal base rate exists. Memoranda are cheap to sign, are usually not disclosed in full, and are rarely followed up publicly when they lapse, so the denominator is unobservable. The gap between headline memorandum values and realised investment at Indian state investment summits is commonly reported to be wide, but published estimates vary enormously with methodology and time window, and quoting a single figure would imply a precision the evidence does not support. What can be said without inventing numbers is that the survivorship bias runs strongly in one direction: successful conversions are announced twice, failures are announced never.

That said, memoranda are not all equally weightless, and the discriminating factors are observable. Ones that convert tend to sit downstream of committed capital, an existing working relationship inside the underlying project, and a defined technical scope. On that test, this instance sits better than most. The fiscal support agreement was signed on 5 March 2025, with the central government committing half of the fiscal support DURABLE; the technology transfer agreement with PSMC was completed in September 2024 DURABLE; and the product scope is defined on the company's own foundry material as power management ICs and microcontrollers DURABLE. Memoranda that fail tend to lack all three. What remains missing here is exactly what a memorandum by definition omits: value, volume and schedule. Whether the buying entity has any prior commercial history with this particular equipment supplier has not been disclosed by either side, and should not be assumed either way.

What the dual-newsroom publication changes, and what it does not

The announcement of 16 May 2026 appeared on both ASML's press release channel and Tata Electronics' newsroom DURABLE. That symmetry matters more than it might appear. A great deal of Dholera coverage rests on a single source, frequently a trade outlet or a real-estate site reporting a claim no principal has confirmed, and the correct discount for that is heavy. Here, two corporate communications functions each cleared the text, one of them belonging to a listed company operating under securities disclosure discipline in two jurisdictions, which does not attach its name to commercial relationships that do not exist. It is reasonable to treat the existence of the document, the identity of the counterparties and the named node range as established, and to treat the quoted reference to a timely ramp as a statement of intent that both sides were willing to own in public.

What it does not establish is anything about money or time. A memorandum creates no enforceable obligation to buy or supply. Equipment suppliers of this kind do not, as a rule, issue press releases for individual purchase orders; bookings are reported in aggregate on a quarterly basis with no customer attribution. Tata Electronics, as a privately held subsidiary rather than a separately listed company, carries no independent stock-exchange disclosure obligation either. The consequence is structural and worth stating plainly: neither party is ever likely to publish the tool count, the order value or the delivery schedule. Anyone waiting for confirmation in that form will wait indefinitely, and the absence of it should not be read as a negative signal any more than the memorandum should be read as a purchase.

The checkpoints that would actually move this claim

Because the direct disclosures will not come, the useful indicators are indirect and mostly physical. Import and customs records for scanner shipments are the hardest single tell, since a machine of this class crosses a border as an identifiable, heavily insured consignment and cannot be moved quietly. Recruitment is the second: lithography process integration and scanner field-service roles posted for the Dholera site would indicate that install is being staffed rather than contemplated, and the training cohorts reported to have been sent to Taiwan across equipment, yield, process technology and quality engineering REPORTED suggest where that staffing is being grown. Supplier-side commentary is the third: PSMC has been reported to expect a substantial multi-year revenue stream from consulting, technology transfer and training across the project lifecycle REPORTED, and material revisions to that expectation in its own reporting would carry information that no Indian announcement will.

On the site itself, the sequence to watch is cleanroom certification, utility hook-up including the specialty gas supply, and then tool move-in, in that order, because none of them can be credibly claimed out of sequence. Claims that foundations are complete, that construction has passed a given percentage, or that cleanroom installation is underway currently circulate only on secondary property and search-optimised sites with no primary source behind them, and should be treated as unsupported until a principal or a government release says otherwise. Against a commercial production target of mid-2028 TARGET, with no first silicon reported, the memorandum is best read as one early and reasonably well-evidenced link in a chain whose later links remain entirely unobserved.