Substrate says it can use accelerator-made X-rays to print cutting-edge chip circuits and build US foundries, challenging ASML and TSMC. Experts are divided.
Substrate is a San Francisco startup that claims it can print the circuits of the most advanced chips using X-rays produced by a particle accelerator. It doesn’t want to sell machines: it wants to build its own foundries in the United States and challenge the sector’s two monopolies at once—ASML in equipment and TSMC in manufacturing. A year after the announcement, the public evidence remains limited to a few images and the figures provided by the company. Experts are split between those who see it as a serious bet and those who don’t believe it will ever make it onto a factory floor.
Who Substrate Is
Substrate was founded in 2022 in San Francisco by brothers James and Oliver Proud and emerged publicly on 28 October 2025. That day it announced a lithography machine it says can compete with ASML’s most advanced systems.
The CEO is James Proud, who at 19 received Peter Thiel’s fellowship for young entrepreneurs. His previous venture, Hello, operated in health technology: it raised $40 million and then shut down. The team, according to the company, includes engineers who came from TSMC, Applied Materials, AMD, Google, and Qualcomm.
With the announcement, Substrate declared it had raised 1 billion. Investors include Founders Fund (Thiel’s fund), as well as General Catalyst, Allen & Co, Long Journey Ventures, Valor Equity Partners, and In-Q-Tel, the CIA-backed non-profit vehicle. Proud says he has not received direct public funding.
Since then the company has said little. In 2026 it published only one update, a technical one, on 7 April. As of 10 November, its news page lists no announcements about new funding, facilities, or customers.
The technology: X-rays from an accelerator
Substrate changes the light source: instead of ASML’s tin plasma, it uses a particle accelerator.
ASML makes light with a plasma, Substrate with an accelerator
Highlighted: what Substrate changes; the diagram follows the company's public descriptions. Tap a step to read the detail.
Sources: Substrate (website and 7 April 2026 update), SemiAnalysis, Tom's Hardware, ASML data sheets. Substrate has said little about the last two boxes.
How it works, according to the company
Radio-frequency cavities accelerate pulses of electrons to nearly the speed of light. Alternating-polarity magnets force them to oscillate and emit X-rays: a beam the company describes as “billions of times brighter than the sun.” A set of polished optics then transports and shapes it all the way to the wafer.
The promise lies in the wavelength. Shorter than EUV’s 13.5 nm, it would make it possible to pattern in a single exposure what today requires multiple steps or High-NA optics. A single accelerator, in principle, can also supply multiple machines and spread its cost among them.
What is not known
The company has not disclosed the wavelength, the type of accelerator, the resist, or throughput. SemiAnalysis speaks only of wavelengths shorter than EUV. Tom’s Hardware infers soft X-rays, between 1 and 10 nm, and a setup reminiscent of a free-electron laser.
There is a clue on the mask. At the announcement the company did not mention it, and some hypothesized maskless direct writing, too slow for production. In the April 2026 update, Substrate instead describes a single mask and a single exposure.
An old idea, with new means
X-ray lithography is not new: IBM worked on it in the 1980s and 1990s with a compact synchrotron. According to Chipstrat, in 1998 IBM itself pointed to the remaining manufacturing problems: mask quality, defects, and stability. The industry chose EUV.
What is new are brighter sources and, above all, computing. Substrate designs and tunes its machines with physics simulations and machine learning. In April 2026 it reported using Google DeepMind’s AlphaEvolve to cut the computing cost of its computational lithography by 97%. With that software, it claims to have printed, in a single exposure, a metal layer with a 24 nm pitch and two-dimensional shapes.
Finally, the company says it has completed the first in-house-built lithography machine for 300 mm wafers, of production-grade quality. It does not indicate when, nor how many wafers per hour it can run.
The stated figures, alongside ASML’s
On paper, Substrate surpasses standard EUV and approaches High-NA; it lags on alignment and says nothing about throughput.
| Substrate (claimed) | ASML EUV (in production) | |
|---|---|---|
| Critical dimension | 12 nm | 13 nm with NA 0.33; 8 nm with High-NA |
| Tip-to-tip spacing | 13 nm | About 25 nm with NA 0.33; 13 nm with High-NA (imec) |
| Single-exposure metal pitch | 24 nm (April 2026) | About 26–28 nm with NA 0.33; 20 nm with High-NA (imec) |
| Overlay between layers | Within 1.6 nm | About 0.9 nm between different machines |
| Wafer uniformity | 0.25 nm | 0.7 nm (NXE:3800E) |
| Throughput | Not disclosed | 220 wafers per hour (NXE:3800E) |
| Tool price | Not disclosed; about $40 million according to a SemiAnalysis estimate | About 400 million for High-NA |
| Leading-edge wafer cost in 2030 | Target: $10,000 | About $100,000, according to projections cited by Substrate |
| Maturity | Laboratory demonstrations | High-volume manufacturing, for years |
Substrate's claimed numbers, next to ASML's
Six metrics, all 'lower is better'. Hover a row to read the note. Substrate's values are company claims, not independent measurements.
Sources: Substrate; SemiAnalysis (reporting company-supplied data); Tom's Hardware; imec (single-exposure High-NA, September 2025); ASML data sheets. Throughput: not disclosed by Substrate, 220 wafers/hour for the NXE:3800E.
Substrate’s values come from the company’s website and from SemiAnalysis, which reports data provided by the company itself: there do not appear to be measurements published by independent laboratories. ASML’s figures come from SemiAnalysis, Tom’s Hardware, and the manufacturer’s datasheets.
The weakest figure is overlay: according to SemiAnalysis, 1.0–1.2 nm would be required. The strongest is uniformity, but 0.25 nm exceeds normal measurement capabilities, the same analysis notes.
On resolution, the company claims parity with High-NA. The comparison holds on tip-to-tip spacing: 13 nm, as in imec’s High-NA tests. On metal pitch, 24 nm versus 20 nm, a gap remains.
The remaining open issues
A successful image is the beginning, not the end: between the laboratory and the factory there are at least six problems left.
- Productivity. There is no public figure for wafers processed per hour. The company only says its machine can withstand the accelerations required by factory throughput.
- Alignment. The stated figure, 1.6 nm, is far from ASML’s 0.9. It is unclear, SemiAnalysis notes, whether it applies across different machines: it is the hardest measurement, and the one that matters in production.
- Photon noise. At shorter wavelengths each photon carries more energy: at the same dose there are fewer photons, and random defects increase. SemiAnalysis lists it among the main technical risks. X-rays can also damage parts of the transistor.
- Resist and masks. EUV resists are not suitable for X-rays, Tom’s Hardware observes. Masks were the technique’s weak point in IBM’s day.
- Optics. Grazing-incidence mirrors are needed—something that no one today manufactures at scale.
- The rest of the fab. Lithography is one step among thousands. The tool has to be integrated with hundreds of others, in processes that today are tuned for EUV.
Then there is time. ASML took twelve years between the first demonstration machine in 2006 and mass production in 2018. Christophe Fouquet, ASML’s chief executive officer, recalls that about twenty years passed between the first EUV image and an industrial system.
For Dan Hutcheson of TechInsights, the next viable window to enter production processes is 2030. To seize it, he wrote in November 2025, the first machines would already have had to reach fabs by then.
The industrial model: foundry, not supplier
Substrate does not want to sell its machines: it wants to use them in its own fabs and sell processed wafers, as TSMC does. MIT Technology Review confirmed this in June 2026: no sales to TSMC, Intel, or others.
The price
The commercial promise is price, expressed in two ways. The first is an end-of-decade target: finished wafers at around 100,000 that Proud forecasts for the rest of the industry. The second, cited by SemiAnalysis, is more cautious: leading-edge wafers at half the price of the alternatives.
Austin Lyons of Chipstrat sums up the proposition in a formula: “2 nm wafers at 28 nm prices.” He also points to where the savings would come from: a single accelerator can power many machines, spreading the source cost.
Independent calculations scale back the promise: lithography is only one of the cost items in a wafer. SemiAnalysis estimates the machine at around 225 million for an EUV tool. Its model, however, yields wafer savings of about 25% under the most favorable assumptions.
Capital
Between the funds raised and the target there is a gap of two orders of magnitude. An advanced fab costs more than 100 million.
Proud acknowledges that many billions may be needed, perhaps tens or hundreds. He is counting on a mix of strategic investments, debt, and public support, TechSpot reports.
Customers
For Lyons, the hard part is not the money or the talent: it is finding customers. Chip designers would have to trust a newly born foundry, whereas TSMC has decades of process experience behind it.
Then there is the double bet: developing a new lithography system and a network of fabs at the same time. Most experts consider this vertical integration unsustainable in such a costly industry, SiliconANGLE reports. Proud counters that in-house machines and cheaper processes will cut costs.
Timelines
There are two stated deadlines, 2028 and 2030. At the announcement, Proud spoke of three years for the first processors; The Wall Street Journal reported large-scale production by 2028 (TrendForce). In June 2026, MIT Technology Review reports 2030 as the milestone for chips at scale.
SemiAnalysis already considered 2028 aggressive for the first designs. In the best-case scenario, it saw volumes arriving at the end of the decade: two years to mature the process, one for the designs, and one to ramp.
Why it matters to Washington
Substrate promises to bring back to the United States two links in the supply chain that are now abroad: the machines, Dutch, and leading-edge production, Taiwanese. Proud calls them the two choke points for chips and considers dependence on those who control them dangerous, TechSpot reports.
Ties to the State
The company says it has not received direct public funding, but it has grown close to the state. Among its investors is In-Q-Tel, the CIA-linked vehicle. Some of the demonstrations took place in national laboratories, Reuters writes.
Stephen Streiffer, director of Oak Ridge National Laboratory, has called the project “of national importance.” Proud says that Commerce Secretary Howard Lutnick has followed him since the start of the administration. He adds, however, that the venture must stand on its own commercially.
The company’s thesis is also a claim of ownership. Research on EUV began in U.S. national labs in the 1990s, Substrate recalls, then leadership shifted overseas. The company also predicts that China will be self-sufficient in chipmaking machines by the end of the decade.
The State as shareholder
The backdrop is a government that is taking equity stakes in the chip industry. In 2025 Washington took a stake in Intel. In December of the same year, the Department of Commerce signed a letter of intent with xLight: up to $150 million, in exchange for shares.
The final award arrived in June 2026. xLight, led by former Intel chief Pat Gelsinger, however develops only the source: a free-electron laser that can power up to 16 EUV machines. Substrate, by contrast, wants to replace those machines, and has not announced any public funding.
What would change
If it worked, the United States would have leading-edge chips made at home, without going through either ASML or TSMC. Today it depends on both, and for EUV machines the export rules are applied by the Dutch government.
Challenges to ASML’s monopoly are now at least three. Japan’s Canon is proposing nanoimprint; China is working on an EUV prototype, according to Reuters; the United States has Substrate’s X-rays. Added to these is Norway’s Lace Lithography, which uses helium atoms and aims to sell machines in 2029–2030, MIT Technology Review reports.
Skeptics and supporters
The prevailing judgment is cautious: the demonstration is taken seriously, the industrial plan much less so.
| Position | Who | Argument |
|---|---|---|
| Those who believe | Stephen Streiffer (Oak Ridge), Paul Kwan (General Catalyst), Austin Lyons (Chipstrat) | The United States can return to a leading role in a market it lost; the industry had resigned itself to the TSMC–ASML duopoly; the strategic logic is sound, but execution will decide |
| Those in the middle | SemiAnalysis (Jeff Koch) | The machine is real, but the claims are extraordinary and the evidence scarce; a long road to high volumes, no imminent threat to ASML |
| Those who doubt | Dan Hutcheson (TechInsights), Chris Miller (“Chip War”), Marc Hijink | No product to sell, only one piece of a process among thousands; changing lithography has always required very long timelines; secrecy as a red flag |
| ASML | Jos Benschop, Christophe Fouquet | Impossible to assess without knowing the processes; producing an image is only the starting point |
| The accusation | Fox Chapel Research (Substack, anonymous) | Fraud hypothesis: images resemble electron-beam writing. An opinion without independent verification |
Those who believe
The most convinced supporters are investors and circles close to the government. Streiffer sees in it the possibility for the United States to return to a leading role in a market it has lost. Paul Kwan of General Catalyst argues that the industry had resigned itself to the duopoly of TSMC and ASML.
Lyons of Chipstrat finds the strategic logic solid. He warns, however, that a good strategy does not guarantee success: execution decides.
Those in the middle
SemiAnalysis considers the machine real, based on external contacts and third-party accounts that are consistent with one another. It nonetheless calls the company’s claims extraordinary and the available evidence scant.
The group discloses a relationship: it has worked with Substrate since 2022 under a non-disclosure agreement. It specifies that the technical analysis is the work of authors without access to that information. As of June 2026, its analyst Jeff Koch still sees a long road to high volumes, and no imminent threat to ASML.
Those who doubt
The doubts come from those who know the industry’s timelines. Hutcheson of TechInsights sees a startup that still has no product to sell and has mastered only one piece of a process among thousands. Chris Miller, author of “Chip War,” notes that in the past changing lithography techniques has taken a very long time.
Marc Hijink, the author of a book on ASML, considers it likely “unfeasible and impossible” to master a new lithography technology and high-volume fabs at the same time. He reads the secrecy as an alarm bell in a sector that runs on open collaboration.
ASML’s response
ASML is not showing alarm. Jos Benschop, an executive at the group, says he cannot assess Substrate because the company has not explained its processes. For Fouquet, being able to produce an image is only the starting point: you then have to do it at volume, fast, and at low cost.
The fraud allegation
The harshest criticism comes from Fox Chapel Research, an unsigned publication on Substack, which two days after the announcement suggested fraud. It argues that the images resemble direct electron-beam writing and that the spaces shown are too small. It is an opinion, as the author himself states: it contains no independent technical checks and no response from the company.
The April 2026 update, with its mask and its single exposure, points in the opposite direction. One fact remains, however: Reuters has not been able to independently verify the company’s claims. Until there are third-party verifiable data, the debate will remain suspended between trust and suspicion.
What to watch
There are two stated deadlines: the first chips in 2028 and scaled production, with $10,000 wafers, by 2030.
Substrate's decisive tests all still lie ahead
On the left what the company says it has done, on the right what it promises. Tap a stage.
Sources: Substrate, Reuters, SemiAnalysis, MIT Technology Review. Past stages rest on company-supplied data; the two that matter to the industry, 2028 and 2030, remain promises.
The milestones already achieved rest on data provided by the company; the two that matter to the industry—2028 and 2030—remain promises. Before then, there are five signals to track.
- Verifiable data. Throughput in wafers per hour, alignment between different machines, and defects, measured or confirmed by third parties.
- Capital. A new multi-billion capital raise, an industrial partner, or public support like the one for xLight.
- A site. The announcement of a factory, with location, permits, and a construction site. As of today, the company’s news page says nothing about it.
- A customer. A chip designer willing to state that it will use the foundry.
- The 2030 window. If the machines are not ready, the next generation of processes will launch with EUV High-NA.
Meanwhile, the target is moving. ASML is studying machines with a numerical aperture of 0.75, which Benschop expects on the market in seven to eight years, MIT Technology Review reports.
Why it matters
Lithography is the bottleneck for the entire semiconductor industry, and today it goes through only two doors: ASML’s machines and TSMC’s fabs. Export controls, the supply chains for artificial intelligence, and part of Taiwan’s security rest on that duopoly. Substrate promises to open a third, entirely American door, with the implicit blessing of Washington and money from investors close to the state. If the bet succeeds, chip prices and the balance of power between the United States, Europe, and Asia would change; if it fails—as happened with X-rays thirty years ago—the lesson would remain that in this field physics bends more easily than industrial scale. Either way, it is a thermometer of how the United States intends to regain the chip supply chain: no longer through subsidies to foreign champions, but through homegrown champions built from scratch.
Sources
Main sources, consulted on 10 November 2026. The figures for Substrate come from company statements: there do not appear to be measurements published by independent laboratories.
Company
- Substrate, Our purpose
- Substrate, technical update of 7 April 2026
- Substrate, news page
Press and news agencies
- Reuters, via Yahoo Finance, Substrate’s announcement, 28 October 2025
- SiliconANGLE, the $100 million raise, 28 October 2025
- TechSpot, company profile, October 2025
- TrendForce, roundup of the international press, 29 October 2025
- TechCrunch, interview with Christophe Fouquet, 5 May 2026
- MIT Technology Review, report on ASML and its challengers, 23 June 2026
Analysis
- SemiAnalysis, How to Kill 2 Monopolies with 1 Tool, 29 October 2025
- Tom’s Hardware, technical analysis of the announcement, October 2025
- TechInsights, The Chip Insider: Substrate X-Ray Lithography, 5 November 2025
- Chipstrat, Substrate, 15 April 2026
- Fox Chapel Research, opinion piece, 30 October 2025
Context
- The Register, the letter of intent for xLight, 3 December 2025
- xLight, via BusinessWire, final award of CHIPS funds, 2 June 2026
- Reuters, via The Jakarta Post, the Chinese EUV prototype, 18 December 2025
- imec, via Semiconductor Digest, single-exposure High-NA results, 22 September 2025
- Canon, nanoimprint press release, 13 October 2023
- ASML, datasheet for the TWINSCAN NXE:3800E