Pat Gelsinger left Intel’s CEO chair in late 2024. He didn’t take a break. He scheduled 100 meetings in 180 days. The goal was simple: narrow the field. He needed to know what came next.
By March, the answer was clear. He joined Playground Capital as a general partner. The firm targets deep tech. Specifically, the hardware powering the next generation of computing.
Gelsinger has a specific obsession right now: light.
He wants to use advanced lithography to restart Moore’s Law. Gordon Moore predicted transistor density would double every two years. That held for decades. Now it’s stalling. Shrinking transistors to atomic limits is expensive. It’s hitting physical barriers.
The solution, according to Gelsinger, is etching chips with nanometer-scale light beams.
The Light-Based Solution to Hardware Bottlenecks
Current lithography relies on 13.5-nanometer ultraviolet light. That’s the standard set by ASML, the Dutch monopoly on chipmaking machines. It works. But it’s at the edge of what’s practical.
Gelsinger is backing xLight, a portfolio company in Playground. xLight is developing new lithography techniques using free-electron lasers. The goal? Shrink light wavelengths down to 2, 3, 4, or 5 nanometers. Smaller light means smaller features. More transistors. Better chips.
This isn’t about defeating ASML. It’s about upgrading the engine.
“We want to hook our light source up and make AS ML machines better.”
If this works, it unlocks performance that traditional design tweaks can’t match. Other innovations like new materials or superconductors are bubbling up. They all hit the same wall. You can’t print the circuit if you can’t etch it with high-resolution light.
How Pat Gelsinger Vets Deep Tech Startups
AI disrupted software. Now it’s pulling venture capital toward hardware.
Most VCs don’t know how to bet on deep tech. They used to. Now they’re guessing. Gelsinger doesn’t guess. He brings a team of engineers, PhDs, and professors to the table. Their process is rigorous.
They look for the hard problem. Can the founder explain it? Is it solvable? Are there two or three other companies chasing the same idea? You rarely find a unicorn in isolation. You find clusters. Pick the best one.
“I want to do things that matter,” Gelsinger says. “If they succeed, make a difference.”
Private equity offers bigger checks. But it’s financial engineering. Gelsinger wants to build. He wants to push science. He’s looking for the edge of discovery.
AI’s Energy Crisis and Hardware Shift
Training AI models used to be the money maker. It still is, for now. But inference is swinging. Companies will train once. They’ll use the model millions of times. That’s where the volume—and the opportunity—is.
GPUs handled training well. They’re okay for inference. But they aren’t optimal. Nvidia knows it. That’s why deals like the one with Groq matter.
The next wave of AI chips won’t look like today’s GPUs. They’ll be heterogeneous. Mixed vendors. Harmonized systems.
Memory is another bottleneck. High-bandwidth memory is the current king. But it’s flawed. By 2030, stacked memory architectures will dominate. Companies like d-Matrix, Fractile, Cerebras, and Alva Energy are breaking the current molds.
Energy is the real constraint. The US expanded energy capacity by single digits in the last ten years. That’s unacceptable.
In the AI age, energy is economic capacity.
“We were so consumed with climate,” Gelsinger notes, “we forgot about capacity.”
New gas turbines take eight years to supply. Nuclear takes a decade. Solar relies on China. Gelsinger’s Alva Energy focuses on upgrading existing nuclear plants. Faster path. Harvest value now. Reignite the build later.
The winners won’t just have smart chips. They’ll have power.
Will US Regulation Stifle AI Leadership?
The US administration is stuck. It wants to lead AI. But it also wants to control it. Chip export controls. Interventions in model releases. Leverage.
Gelsinger sees no contradiction. He wants Western values to win. Not Chinese.
The tech moves too fast for slow regulation. A major foundational model drops every four weeks. What do you regulate? Security? Quality? Integrity?
“Who decides?” he asks.
He wants transparency. He wants to know what proprietary models are trained on. He wants rigorous benchmarking. He wants security requirements baked in from the start.
If the industry can’t self-regulate, the government has to step in. That’s the only way to ensure models aren’t just first—but safe.
The Open Question
Gelsinger thinks light is the key. Not just for Moore’s Law. But for efficiency. For power. For the next decade of computing.
He’s betting on a team that can print with 2-nanometer light. He’s betting on nuclear upgrades. He’s betting that heterogeneity will win over monolithic GPU designs.
Will it work?
Nobody knows for sure. Science is still proving things out. The physics is hard. The money is real. The clock is ticking.
Gelsinger is just getting started. The question is whether the industry can keep up with the pace he’s setting. Or if the light, like the law, will finally go dark.
