Pat Gelsinger's post-Intel bet: a startup trying to push lithography past the limits of today's chipmaking

Pat Gelsinger's Post-Intel Bet: Whoever Controls the Photon Controls AI's Future

Every advanced AI chip on earth is printed with light at 13.5 nanometers. That wavelength comes from lithography machines developed by ASML, the Dutch company whose technology sits underneath the entire frontier of semiconductor manufacturing 1. Pat Gelsinger thinks the next leap in AI compute depends not on a better GPU design but on a fundamentally shorter wavelength of light.

Since departing Intel as CEO in late 2024, Gelsinger has resurfaced as a general partner at Playground Capital, a venture capital firm specializing in deep tech 1. He arrived at the role after 100 meetings in 100 days to narrow his options, choosing venture over private equity because he wanted to work at the edge of science rather than manage financial returns 1. His first major portfolio play: xLight, a portfolio company developing novel lithography techniques that has also drawn investment from the US government 1.

The physics, simplified. ASML's machines etch chip patterns using 13.5-nanometer light. Gelsinger's bet is that free-electron lasers could push that wavelength down to 5, 4, 3, even 2 nanometers 1. Shorter light means finer etching. Finer etching means denser transistors. And denser transistors are exactly what the semiconductor industry needs, because the doubling cadence that drove decades of progress has slowed as manufacturers press against atomic-scale limits 1.

"If we solve light, that is the hardest problem," Gelsinger told WIRED 1.

That framing matters for anyone building or investing in AI infrastructure. The leading lithography technology comes from one company. Every chipmaker pushing the frontier depends on the 13.5-nanometer light that ASML's machines produce. For an industry racing to feed AI demand, that is a striking concentration of dependency in a single node of the supply chain. New materials, new transistor designs, and new packaging approaches all help. But as Gelsinger points out, every path runs through lithography 1.

Gelsinger is careful to position xLight as a partner rather than a rival. "xLight is not versus ASML, it's with ASML," he said, explaining that the first step is to attach xLight's light source to existing machines and improve them 1. The cooperative framing is strategically sensible. But the stated ambition, pushing toward 2-nanometer light, aims well beyond what current ASML machines were designed to deliver.

The urgency is quantified. Gelsinger predicts the semiconductor industry will reach $1 trillion in revenue next year, pulling forward a target that was previously set for 2030 1. That acceleration is driven by AI demand, and it places pressure on every link in the manufacturing chain.

Beyond lithography, Gelsinger sees the hardware landscape shifting in two directions. First, a swing from model training toward inference 1. Second, new inference and memory architectures that could deliver 10 to 100 times the efficiency of today's GPUs. In his framing, the payoff is concrete: "one gigawatt produces 10 gigawatts worth of tokens" 1. His stated goal at Playground is characteristically scaled: "My job at Playground is to make AI 10,000 times better, not 10 times" 1.

Gelsinger contends that few venture investors are better equipped than he is to identify which technologies will break through, noting that most firms moving into deep tech have, in his words, "forgotten how to do deep tech" 1. That conviction fits someone who has built a career around betting on where semiconductor physics goes next. Whether he is right about free-electron lasers depends on a question the WIRED interview does not address: how soon can xLight move from concept to a light source rugged enough for the relentless throughput of a commercial fabrication plant?

The $1 trillion semiconductor market Gelsinger sees arriving next year will be built on today's 13.5-nanometer light. The question is whether the next trillion runs on something shorter.