Pat Gelsinger departed Intel as CEO in late 2024. He held 100 meetings in 100 days. In March 2026, he announced he had taken a position as general partner at Playground Capital, a venture firm specializing in deep tech. His thesis: the best way to break the impasse in semiconductor manufacturing is through advances in lithography - etching chips using nanometer-scale beams of light.
This is not a retirement project. It is a bet on the physical limits of the most important manufacturing process in the world. Moore's Law - the prediction by Intel co-founder Gordon Moore that the number of transistors on a chip would double roughly every two years - held true for decades. It is now hitting physics. Further shrinking atomic-scale transistors has become prohibitively difficult and expensive. The industry needs a new path, and Gelsinger thinks it is light.
The Lithography Wall
Here is the problem. EUV lithography at 13.5 nanometers is already in production at TSMC, Samsung, and Intel. It is how the most advanced chips in the world are made. But the next step - high-NA EUV, which uses a larger numerical aperture to print even finer features - is staggeringly expensive. Each ASML high-NA machine costs upward of $350 million. TSMC has reportedly hesitated on ordering them, preferring to extend the life of current EUV with multi-patterning techniques instead.
Gelsinger's argument is that the industry cannot multi-pattern its way to the next node indefinitely. The cost and complexity of splitting a single chip layer into multiple exposure steps grows nonlinearly. At some point, the economics break. The only way forward is shorter wavelengths - pushing beyond EUV into what the industry calls "beyond EUV" or x-ray lithography. That requires new light sources, new optics, new resists. It is a deep tech problem, not an engineering optimization.
The Geopolitical Lens
The lithography race is inseparable from the chip sovereignty question. ASML, based in the Netherlands, is the only company in the world that makes EUV lithography machines. The US has used export controls to prevent ASML from selling its most advanced machines to China. TSMC, in Taiwan, is the largest buyer. The entire advanced semiconductor supply chain runs through two countries and one company.
Gelsinger's move into deep tech venture capital is, implicitly, a bet that this bottleneck cannot hold. If the US wants chip sovereignty - the ability to manufacture advanced semiconductors domestically - it needs domestic lithography capability. No US company currently makes EUV machines. The technology is so complex, and ASML's lead so large, that catching up requires not incremental improvement but a generational leap. Gelsinger is looking for the startups that might make that leap.
The same week Gelsinger's plans were reported, Wistron opened a manufacturing plant in Fort Worth, Texas to produce NVIDIA AI systems. The onshoring trend is real, but it is assembly, not lithography. You can build a data center in Texas. You still cannot print the chips that go in it without a machine from Veldhoven.
Why This Matters for AI
The connection to AI compute is direct. Every advance in AI model capability depends on advances in chip density. More transistors per square millimeter means more compute per chip, which means larger models, faster inference, and lower cost per token. If Moore's Law slows - if the cost per transistor stops declining - the economics of AI training and inference shift. The capability curve flattens. The efficiency race that Anthropic's Opus 5 represents becomes a zero-sum game: you can only get cheaper by getting smaller, and you can only get smaller if lithography keeps advancing.
Gelsinger is betting that it can. Not through the same EUV technology that ASML has perfected, but through something new. Light at shorter wavelengths. New physics. The kind of breakthrough that does not come from optimizing an existing process but from changing the process entirely.
Moore's Law is not a law of physics. It is a bet that has paid off for sixty years. Gelsinger is placing the next one.