Pat Gelsinger’s next chip bet is light

After leaving Intel in late 2024, Pat Gelsinger says he took 100 meetings in 100 days to figure out what came next. By the following March, he had landed at Playground Capital as a general partner, betting on deep-tech startups he believes could restart Moore’s law.

His core thesis is simple: the semiconductor industry is running into the physical and economic limits of shrinking transistors, so the next breakthrough has to come from lithography. Today’s leading systems from ASML use 13.5-nanometer wavelength light. Gelsinger argues that pushing to smaller wavelengths could unlock another leap in chip performance.

At Playground, that view led him to xLight, where he has taken a board seat. The startup is developing new lithography technology and has also received investment from the US government. Gelsinger told WIRED that xLight is not trying to replace ASML but improve its tools by supplying a better light source. With free-electron lasers, he said, the industry could move beyond 13.5 nanometers to 5-, 4-, 3-, and 2-nanometer wavelength light.

Recommended reading

Dimension Capital lands $800M as science-compute bets pay off

What Gelsinger is betting on

Gelsinger said he chose venture capital over private equity because he wanted to focus on “cool tech” rather than simply writing larger checks. At Playground, he says the goal is to back the best team on a given scientific problem, using a diligence process built around engineers, PhDs, professors, interviews, and background checks.

He argues that AI has dramatically expanded the opportunity for deep-tech investing. According to Gelsinger, the semiconductor industry in 2024 was aiming to reach $1 trillion by 2030, but will now hit $1 trillion next year.

He also expects the center of AI hardware to shift:

• from training toward inference
• from today’s GPUs toward new chip architectures
• from current memory approaches toward more stacked memory architectures by the end of the decade

He cited d-Matrix, Fractile, and Cerebras as examples of companies pushing memory design in new directions, and said some hardware advances could be 10 to 100X better.

Energy, export controls, and model oversight

Gelsinger tied chip progress directly to power constraints. He called the US expansion of energy capacity over the last decade “low-single-digit” and “despicable,” arguing that in the AI era, energy capacity is effectively economic capacity. He pointed to long lead times across the board: eight years for new gas turbine supply chains and a decade to build nuclear.

One Playground-backed company, Alva Energy, is focused on nuclear upgrading, which Gelsinger described as a faster route to getting more out of the current nuclear base while broader nuclear construction ramps up.

On policy, he said he does not see tension between extending US leadership and broadly sharing AI’s benefits. If the choice is between the US and China leading in foundational models, he said he wants the US to win. But he also argued that model releases need stronger review.

“Models need to have integrity of process and visibility of the testing that was done on them. I want to know what proprietary foundational models are trained on. I want vigorous benchmarking.”

His bottom line: either the industry creates that review process itself, or the government will need to step in and ensure it happens.