Pat Gelsinger has placed his post-Intel bet on light. The former Intel CEO joined Playground Capital as general partner in March 2025 and took a board seat at xLight, a lithography startup developing free-electron laser sources aimed at pushing chip printing below ASML's current 13.5-nanometer wavelength standard, down to 5, 4, 3, and 2 nanometers. xLight has also drawn investment from the US government. Gelsinger's thesis is that lithography, not processor design, is the gating factor for reviving Moore's law.
The timing is aggressive. In 2024, the semiconductor industry set a target of $1 trillion in annual revenue by 2030. Gelsinger now says the industry will hit that mark in 2026, pulling the milestone forward by four years on the back of AI demand. He left Intel in late 2024, spent roughly four months in what he calls a deductive process, held 100 meetings in 100 days, and ruled out public-company CEO roles, government posts, and private equity before landing at Playground.
The deep-tech pivot is not unique to Gelsinger. Venture firms across the board are rotating out of application software, which generative AI has commoditized, and into hardware, materials, and energy. Gelsinger's argument for why Playground has the edge is that most VCs rotating into deep tech have forgotten how to underwrite it, and his firm's investment team is stacked with engineers, PhDs, and professors who can run technical diligence rather than pattern-match on founder résumés.
“Do I want to write big checks and worry about financial returns, or do I want to do cool tech?”— Pat Gelsinger, General Partner at Playground Capital
Key facts
- 01Pat Gelsinger joined Playground Capital as general partner in March 2025, four months after leaving Intel as CEO in late 2024.
- 02He took a board seat at xLight, a lithography startup targeting wavelengths below ASML's current 13.5-nanometer standard, down to 5, 4, 3, and 2 nanometers.
- 03Gelsinger says the semiconductor industry will hit $1 trillion in 2026, four years ahead of the 2030 target set in 2024.
- 04US energy capacity grew only low-single digits over the past decade; new gas turbines carry an eight-year supply chain and nuclear takes a decade to build.
- 05A major foundation model now ships every four weeks, Gelsinger said at the RAISE Summit in Paris in early July 2026.
xLight is positioned as a supplier to ASML, not a competitor. The company's goal is to plug its light source into existing ASML machines to improve throughput and enable smaller feature sizes. That framing matters — ASML holds an effective monopoly on extreme ultraviolet lithography and any credible next step in the technology has to route through its installed base rather than around it. Getting to sub-13.5-nanometer wavelengths would open the door to new transistor densities that today's tooling cannot reach.
Beyond xLight, Gelsinger sees the inference market as the next battleground for silicon. He points to Nvidia's own recognition, via its Groq deal, that GPUs are not optimal for all inference workloads, and to memory startups d-Matrix, Fractile, and Cerebras as candidates to break the current high-bandwidth memory bottleneck with stacked architectures by the end of the decade. His estimate is that these hardware innovations can deliver 10 to 100 times better performance per watt.
The math he cites is stark: one gigawatt of power producing 10 gigawatts worth of tokens. That kind of efficiency gain is the only path to squaring AI's compute demand with a US grid that has grown only low single digits in capacity over the past decade. New gas turbines carry an eight-year supply chain, nuclear plants take a decade to build, and solar depends heavily on Chinese supply chains. Gelsinger's own portfolio includes Alva Energy, which works on upgrading existing nuclear capacity.
“My job at Playground is to make AI 10,000 times better, not 10 times.”— Pat Gelsinger, General Partner at Playground Capital
On policy, Gelsinger is direct about the geopolitical stakes. He said at the RAISE Summit in Paris in early July 2026 that if forced to choose between US or Chinese leadership in foundation models, he picks the US, and framed the current moment as a race that Western nations should win. He argues export controls and model-release interventions are not necessarily in tension with broad AI diffusion, though he acknowledges the US administration is still figuring out where to land on regulation.
The regulatory uncertainty is a real constraint. A major foundation model now ships every four weeks, Gelsinger said, which makes it structurally difficult to write rules that are still relevant by the time they take effect. He described the current posture as figuring it out in real time.
“Energy capacity is economic capacity.”— Pat Gelsinger, General Partner at Playground Capital
The skeptic's case on Gelsinger's thesis is that lithography advances are extraordinarily capital-intensive and slow. ASML spent decades and tens of billions of dollars getting to 13.5-nanometer EUV in production. A startup pushing to shorter wavelengths, even in partnership with ASML, faces a science-risk profile that most venture portfolios cannot absorb. Gelsinger's own framing — that he only needs his companies to win a decent percentage of the market, not all of it — is an implicit acknowledgment that individual bets will fail.
For the AI market, the read is that the compute-supply story is entering a new phase. The first phase was GPU scarcity. The second is power. The third, if Gelsinger is right, is the physical limit of how small a transistor can be printed — and whoever cracks sub-13.5-nanometer lithography first sets the ceiling on model scaling for the rest of the decade. That is a narrower race than the model-lab wars getting the headlines, but it may matter more.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
