Pat Gelsinger’s post-Intel plan is a bet that the chip industry can squeeze more progress out of physics, if someone can build a better light source.
Gelsinger, who left Intel as chief executive in late 2024, told WIRED he held 100 meetings in 100 days while deciding what to do next. In March, Playground Capital announced that he had joined the deep-tech venture firm as a general partner. He has also taken a board seat at xLight, a Playground portfolio company working on new lithography technology that has received US government investment.
The pitch is familiar in outline and brutal in execution: make smaller chip features by improving the light used to print them. Lithography is the process that transfers chip patterns onto silicon. ASML’s leading extreme ultraviolet systems use 13.5-nanometer wavelength light, according to the company. Gelsinger told WIRED that xLight wants to use free-electron lasers to move to shorter wavelengths, potentially in the 5-, 4-, 3- or 2-nanometer range.
That is the kind of claim that deserves the raised eyebrow it will get from anyone who has watched semiconductor road maps turn into invoice-generating rituals. Gelsinger framed xLight’s work as complementary to ASML rather than a replacement for it, saying the company’s first goal is to connect its light source to ASML machines and improve them.
The Moore’s law problem has become a lithography problem
Gordon Moore’s old observation, that transistor counts would roughly double every two years, drove decades of chipmaking expectations. The industry kept up for a long time by shrinking transistors and packing more of them onto a die. That approach now runs into atomic-scale physics and unpleasant economics.
Gelsinger told WIRED that he sees lithography as the central bottleneck. Other semiconductor ideas, including new materials and device structures, still need patterning. His argument is that better lithography could reopen some of the scaling headroom the industry has been losing.
He also cast artificial intelligence as the demand shock pushing venture firms back toward hard technology. Gelsinger claimed the semiconductor industry had been aiming for $1 trillion in annual revenue by 2030, and now may reach that level next year. He said that shift has made deep-tech startups more attractive, while arguing that many venture firms no longer have the technical diligence muscles needed to judge them.
Playground’s process, as Gelsinger described it to WIRED, leans on technical staff including engineers, PhDs and professors. He said the firm probes whether founders can explain the hard scientific problem in detail and compares teams chasing similar ideas.
AI chips, memory and power are part of the same bet
Gelsinger also told WIRED that AI inference will create room for chips that do not resemble today’s GPUs. He said model developers are trying to abstract their software away from specific hardware, which could make mixed-vendor AI systems easier to build.
On memory, he called high-bandwidth memory the best available option today while describing it as problematic. He pointed to stacked memory architectures and named d-Matrix, Fractile and Cerebras as companies pushing alternatives.
Energy remains a constraint in his account. Gelsinger said US energy capacity has expanded only in the low single digits over the past decade and argued that AI competitiveness will depend on power availability. He cited Alva Energy, one of his companies, as working on upgrades to existing nuclear plants, while also calling for more nuclear construction.
On AI governance, Gelsinger told WIRED he wants US leadership in foundational models and said model releases need clearer testing, benchmarking and process visibility. He said either industry must perform that review or the government will have to step in. That is not exactly a hands-off view, which is probably correct for a world where the chips are expensive, the power plants are political, and the models keep arriving faster than the paperwork.
This story draws on original reporting from WIRED.
