The Light That Etches Chips Is Now the Bottleneck. A Former Intel CEO Is Betting a New Laser Can Change It.

Pat Gelsinger left Intel in late 2024 after a foundry push that reshaped the company's capital spending but not its market share. He ran roughly a hundred meetings over the next hundred days to decide what came next. By March 2025, he had joined Playground Capital as a general partner, focused on deep-tech semiconductor bets, and taken a board seat at xLight, a startup building a free-electron laser (FEL) as a successor to the extreme ultraviolet (EUV) tools that define the current generation of advanced chips.

"The constraint that decides whether the chip industry can keep scaling is no longer the transistor. It's the light," Gelsinger said in a WIRED interview and at the RAISE Summit in Paris in early July 2026.

The chip industry has kept Moore's Law alive for sixty years by handing each new constraint to a different set of engineers. Smaller process nodes gave way to finFET transistor geometries, then to gate-all-around designs, with strained silicon, high-k metal gates, and 3D stacking each arriving as the previous layer stopped yielding easy gains. In 2026, the relay is at a new handoff. The next generation of transistor architectures is largely on public industry roadmaps. What determines whether those architectures ship at scale is whether ASML's current EUV tools, which use 13.5-nanometer light (about fifty times shorter than visible light) to print features at 8 to 20 nanometers, can be pushed further, or whether a new light source can come online to take over.

Gelsinger says pushing EUV further has become "prohibitively difficult and expensive." The industry's consensus path, ASML's high-NA EUV, is a refinement of the same architecture. "High-NA" refers to a higher numerical aperture in the projection optics, which lets the tool print finer features at the same wavelength; the trade-off is a larger, more expensive optical system. Bigger mirrors, more power, the same fundamental 13.5-nm wavelength. It works. It also costs more per wafer than any previous generation.

A free-electron laser does not depend on a fixed wavelength. It accelerates electrons in a linear accelerator and forces them to oscillate through a magnetic structure called an undulator. Each oscillation emits photons, and the photons add up coherently into a laser beam whose wavelength can be tuned. For lithography, the pitch is simple: shorter wavelength means smaller features can be etched. That is the bet xLight is trying to build out at the Albany Nanotech Complex in New York, with about $150 million in CHIPS and Science Act incentives finalized by the US Department of Commerce in June 2026. The first prototype is a piece of test equipment, not a manufacturing tool. It has to answer whether an FEL can produce a beam stable enough, at the right wavelength, to make a wafer stepper actually run.

He is a general partner at Playground Capital and a board member at one portfolio company. The thesis he is buying into is one he has spent the last decade reading, because he has spent the last decade at the largest chip buyer in the world. "The only thing that mattered in the last five years at Intel was whether ASML could ship enough EUV tools," Gelsinger said in the WIRED interview. "The only thing that will matter in the next ten is whether someone can build the next EUV."

AI demand has pulled capital into chips, and chips have pulled capital into the tools that make chips. The $150 million of public money going to xLight is the visible part of that movement; the larger pattern is a roster of deep-tech funds building positions in lithography, deposition, and metrology, the three categories where a single equipment generation can change the entire industry.

The thesis faces three concrete tests in the next eighteen months: whether xLight's prototype produces a working shorter-wavelength beam at Albany by mid-2027, whether ASML's high-NA EUV roadmap delivers a cost-per-wafer improvement steep enough to make a successor optional, and whether another FEL or alternative source is visibly ahead of xLight in a national lab or a competing startup. Gelsinger's own caveat is the cleanest one: he does not claim to have cracked the problem. He has placed a venture-scale bet on someone who is trying.