Intel's ex-CEO bets on light-based chips to revive Moore's Law

Pat Gelsinger departed Intel as CEO in late 2024 and spent the following hundred days in deliberation, holding 100 meetings to narrow his options. He considered government roles, university positions, CEO positions, private equity, and venture, ultimately ruling out public company earnings calls, political work, and private equity's focus on financial returns over technology. In March 2025, he announced his role as general partner at Playground Capital, a venture capital firm specializing in deep tech and early-stage technologies built on novel science.

Gelsinger's primary rationale for joining Playground was the opportunity to work with xLight, a portfolio company developing advanced lithography techniques. He took a board seat at xLight, which recently received investment from the US government. Gelsinger's central thesis is that lithography represents the hardest and most critical problem in semiconductor manufacturing. The leading lithography technology, developed by Dutch company ASML, uses 13.5-nanometer wavelength light to etch chip features. Gelsinger believes the next breakthrough lies in moving to shorter wavelengths—potentially 5-, 4-, 3-, or 2-nanometer using free-electron lasers. Rather than positioning xLight as a competitor to ASML, he frames the relationship as complementary: the first goal is to integrate xLight's light source into ASML machines to improve them.

Gelsinger connected this lithography bet to the broader semiconductor landscape. He noted that the semiconductor industry aimed to reach $1 trillion（約160兆円） in revenue by 2030, but AI demand has compressed that timeline—the industry will now hit that figure next year. This acceleration has drawn venture capital toward deep tech, but Gelsinger contends that most VCs lack the technical foundation to distinguish winners from losers. Playground's investment team comprises engineers, PhDs, and professors who conduct rigorous technical diligence on founders, examining whether they can articulate core problems in depth. He emphasized that his job is to make AI "10,000 times better, not 10 times," and that this will require chips that do not resemble today's GPUs. Beyond lithography, he identified memory architecture as another critical frontier—high-bandwidth memory remains the current best option but is problematic, and by the end of the decade stacked memory architectures will likely become dominant. He cited companies like d-Matrix, Fractile, and Cerebras as breaking boundaries in memory design, with potential gains of 10 to 100 times in efficiency—meaning one gigawatt of power could produce the token output of ten gigawatts.

Energy emerged as a second major focus. Gelsinger flagged the US expansion in energy capacity in the last decade as "despicable," noting low-single-digit growth when digital AI demands require massive increases. New gas turbines have eight-year supply chains, nuclear power takes a decade to build, and solar depends on Chinese supply chains. Among his portfolio companies, Alva Energy focuses on nuclear upgrading—both harvesting more value from existing nuclear infrastructure and reigniting new nuclear builds. He stressed that winners and losers in the AI age will be determined by energy capacity. Gelsinger also highlighted opportunities in efficiency: voltage regulation, power distribution, and making chips themselves more power-efficient. On the regulatory front, he acknowledged the US administration's mixed signals—sometimes hands-off, sometimes imposing chip export controls and model release interventions—but argued these are not contradictory. He favored vigorous benchmarking and security requirements for foundational models, calling for either industry self-review or government oversight to ensure proprietary models are trained responsibly and aligned with values.

Gelsinger's pivot from Intel to venture capital marks a strategic recalibration at a inflection point in both semiconductors and AI. During his 100-day deliberation after leaving Intel in late 2024, he weighed government roles, academia, CEO positions, and private equity before settling on venture—a choice he framed as driven by personal preference for advancing frontier technology over pure financial returns. The semiconductor industry's timeline itself has accelerated: Gelsinger notes that projections moved from hitting $1 trillion（約160兆円） in revenue by 2030 to achieving that milestone as soon as next year, a shift he attributes directly to AI's surge in demand.

The lithography bet reflects a deeper conviction that incremental improvements to existing chip design will not meet AI's power and performance needs. While Nvidia's GPU dominance in AI training is undisputed, Gelsinger sees inference—the repeated use of trained models—as an emerging frontier where diverse chip architectures from multiple vendors could thrive. His focus on xLight and light-based manufacturing is not framed as competition with ASML but as complementary: the first step is integrating xLight's light source into ASML's existing machines to make them better. This reflects both humility about the scale of ASML's installed base and confidence that solving lithography unlocks everything downstream.

Gelsinger also emphasizes that most venture firms, despite rushing toward deep tech in response to AI disruption, lack the technical depth to distinguish between viable breakthroughs and dead ends. His assembled team—engineers, PhDs, and professors—is presented as a corrective. The combination of his Intel pedigree, a portfolio spanning lithography, memory architecture, and energy infrastructure (including nuclear upgrades), and his stated mission to make AI "10,000 times better, not 10 times" positions Playground as a bet on the hardware stack underneath AI, not software abstractions.