xLight Is Betting Free-Electron Lasers Can Break ASML’s Grip on Chip Manufacturing

The most important machine in semiconductor manufacturing isn’t built in the United States, Taiwan, or South Korea. It comes out of Veldhoven, Netherlands, from a company called ASML, and every advanced chip in every AI server and data center traces back to it. Their extreme ultraviolet lithography machines run about $380 million apiece, weigh 180 tons, and pull parts from more than 5,000 suppliers. They are the only tools on Earth that can print the sub-5nm circuit patterns modern logic chips are made of. Nobody else has come close in over a decade.

xLight, now chaired by former Intel CEO Pat Gelsinger, wants to end that. The company is negotiating a $350 million round expected to be led by Boardman Bay Capital Management and Bain Capital, sitting on top of a $150 million equity investment from the U.S. Commerce Department’s CHIPS R&D Office. That stake is the first direct equity the Trump administration has taken under the CHIPS and Science Act. The technology underneath all of it is a free-electron laser, born in particle physics research and never once commercialized at chip-fabrication scale. Make it work and you rewrite the economics of advanced lithography. Fail and you’ve funded one of the most expensive physics experiments in the history of the industry.

ASML’s moat is deep because of what its machines actually do. EUV lithography generates light at a wavelength of 13.5 nanometers, short enough to resolve features under 10nm on silicon. The way ASML gets that light is laser-produced plasma: a high-power CO₂ laser fires at a stream of tin droplets running about 50,000 droplets per second, blasting each one into a plasma that spits out EUV photons. Those photons get gathered by multilayer mirrors, which are their own engineering miracle since no material transmits EUV at all, then steered through a reticle onto the wafer.

That plasma source is the hardest piece to build and the thirstiest. High-NA EUV machines, the current generation, eat on the order of a megawatt of facility power. Conversion efficiency from laser input to usable photons at the wafer sits at a few percent at best, so most of that energy leaves as heat somebody has to manage. The light source’s power dictates how fast the machine can expose wafers, and wafers per hour is the whole game. More power, more chips per hour, lower cost per die. That single constraint is what xLight is aiming at.

A free-electron laser makes light a completely different way. In an ordinary laser, photons come from stimulated emission as bound electrons jump between fixed atomic energy levels, which is why every laser medium has its own characteristic wavelength. A FEL doesn’t carry that baggage. It accelerates a beam of electrons to relativistic speed through a linear accelerator, then runs the beam through an alternating magnetic field structure called an undulator. The field whips the electrons into a sinusoidal path, and as they curve they shed synchrotron radiation. The wavelength of that radiation is tunable, set by the electron beam energy and the undulator period rather than locked to any atomic transition.

What comes out is a coherent, high-brightness source you can tune across a wide range, EUV included. The physics has been demonstrated for years at the Jefferson Lab FEL in Virginia, at FLASH in Hamburg, at the European XFEL. None of those are production tools. They’re enormous research installations. xLight has to compress that physics into something that fits inside a fab, runs continuously at production power levels, and stays up reliably enough to clear the bar a commercial foundry sets, which for ASML’s machines means better than 90% availability.

The advantages are worth the trouble if they hold. FELs can in principle hit far higher electrical-to-photon conversion efficiency than a plasma source, which would cut both operating cost and the thermal headache. The light comes out more coherent and more directional than plasma emission, which could simplify the optics that deliver it. And tunability means a single accelerator platform could feed multiple lithography wavelengths, including wavelengths shorter than 13.5nm that matter for nodes beyond anything current EUV can resolve.

The prototype is being built in Albany, New York, with U.S. national labs, in the same corridor that hosts SUNY Poly’s College of Nanoscale Science and Engineering and has long been a home for pre-competitive semiconductor R&D. First functional light source is targeted for 2028.

The $150 million from the CHIPS R&D Office is structured as equity, which means the U.S. government becomes an actual shareholder in a private startup rather than a grant-giver or a lender. It started as a non-binding letter of intent in December 2025 and was signed as a final award on June 2, 2026, the first equity investment the CHIPS R&D Office has made under this administration. Commerce Secretary Howard Lutnick put it in flag-waving terms: “For far too long, America ceded the frontier of advanced lithography to others. Under President Trump, those days are over.”

Close the $350 million private round and total funding lands around $550 million. Some reports describe $4.2 billion in non-binding project financing lined up with lenders for xLight’s first seven facilities, but the more authoritative sources don’t confirm that number, so treat it as aspirational until somebody verifies it independently. What isn’t in doubt is the scale of capital this takes and how far off revenue is. A 2028 prototype is a proof-of-concept, not a product. Going from a working light source to a machine that exposes wafers at production throughput, slots into existing fab infrastructure, and survives the reliability standards at TSMC or Samsung is more years and more money on top.

Gelsinger himself complicates the picture. The Intel board pushed him out as CEO in late 2024 when the manufacturing turnaround, the 14A node and the Intel Foundry Services buildout, didn’t gain traction fast enough. He’s now executive chairman of a company chasing the same mission he just lost his old job over: restore U.S. leadership in advanced chip manufacturing by building a domestic alternative to foreign-controlled lithography. His venture firm, Playground Global, also put money into xLight in 2025, so his financial interests track the company’s success directly. Call that alignment or call it a conflict depending on where you sit.

The strangest part of xLight’s position is its relationship with ASML. The whole pitch is that xLight offers an alternative to ASML’s light source, loosening the industry’s dependence on one Dutch supplier and maybe opening the door to rival tool makers. Yet ASML CEO Christophe Fouquet has reportedly talked about collaborating with xLight on technology demonstrations while calling the road to commercialization “a long journey.”

That tension resolves once you remember what ASML actually sells. Its business is building lithography machines, not light sources. The LPP source in today’s EUV machines came out of a partnership with Cymer, which ASML bought in 2013, and Trumpf. If xLight can show a FEL source that beats plasma on power and efficiency, ASML could fold it into a future machine generation instead of fighting it. ASML stays the systems integrator and the dominant tool supplier, xLight becomes a component supplier, and both sides do fine. That’s a long way from “break ASML’s monopoly,” which makes a better headline than a better business plan. The other outcome, where xLight’s source lets entirely new tool makers stand up against ASML, is far harder, because it demands a new optical column, new reticle infrastructure, and a new wafer stage, every one of them decades of ASML engineering.

xLight isn’t the only one swinging at this. Substrate, another startup, is chasing X-ray lithography as a separate route to sub-EUV resolution. X-ray uses even shorter wavelengths than EUV, which could buy finer features, but it drags along its own problems in mask fabrication, alignment, and throughput. Several well-funded startups attacking one wall from different angles is a decent sign the opportunity is real and that smart money thinks the barriers, high as they are, can be cleared.

The customer list is short and absurdly valuable. TSMC, Samsung, and Intel Foundry are the three fabs that will need next-generation lithography to print chips at 2nm and below. IBM Research, embedded in the Albany ecosystem, is another possible collaborator. xLight is reportedly courting TSMC, Intel, and Micron as strategic investors, which would validate the technology and build a commercial pipeline at the same time, though those talks aren’t confirmed.

ASML’s EUV backlog runs long, its machines are spoken for years out, and the geopolitical pressure to escape Dutch export-controlled equipment keeps climbing. A government willing to buy equity in xLight has decided the strategic value of a domestic EUV alternative is worth the technical risk, even knowing a 2028 prototype won’t be production-ready until deep into the 2030s, by which point ASML will have shipped multiple generations of High-NA and possibly Hyper-NA machines.

A closed $350 million round buys the runway to answer one question: can a free-electron laser cross from national-lab curiosity to fab-floor production tool? My doubt sits entirely on the engineering, not the science. The physics already works in Hamburg and Virginia. Sound physics has stranded plenty of companies that couldn’t make it run every shift for a decade straight, and that decade is where this bet lives or dies.