Pat Gelsinger Bets on Light to Revive Moore’s Law

Former Intel chief executive Pat Gelsinger has joined venture firm Playground Capital and is using the role to back semiconductor startups that he believes could revive Moore’s law, with lithography and advanced light sources at the center of the push. His view matters because he argues the next leap in chip performance will come not from squeezing current transistor designs a little further, but from fundamentally changing how chips are printed and powered.

Gelsinger, who left Intel in late 2024, told WIRED he spent the months after his departure meeting dozens of people to narrow his next move. By March, he had landed at Playground as a general partner and taken a board seat at xLight, a startup developing new lithography approaches that recently received backing from the US government.

That decision reflects a broader shift across Silicon Valley and Wall Street: as artificial intelligence rewires the software market, investors are moving deeper into hardware, energy, and the physical infrastructure that makes AI possible. Gelsinger’s pitch is that the companies best positioned to benefit will be those solving the hardest engineering problems, from chip manufacturing to memory systems to power generation.

He sees lithography as the key lever. In his view, the semiconductor industry has spent years approaching the edge of what conventional transistor scaling can deliver. The answer, he argues, is not to abandon Moore’s law but to reawaken it through better light sources and more capable manufacturing tools.

Why Gelsinger Left Intel and Chose Venture Capital

Gelsinger said the period after leaving Intel was deliberately methodical. Rather than rushing into a new executive post, he explored options across government, academia, private equity, and venture capital before concluding that he wanted to stay close to frontier technology without returning to the daily demands of a public company.

His explanation was simple: he wanted work that felt consequential, involved people he enjoyed, and did not require him to manage earnings calls or political trade-offs. That left venture as the best fit.

He also made a distinction between private equity and deep-tech venture investing. Private equity, he said, can deploy larger amounts of capital, but it is usually less focused on the technical substance of early-stage innovation. At this stage in his career, he preferred to back technologies that could reshape industries rather than simply produce financial engineering.

Gelsinger said he wanted to spend the rest of his career on “things that matter,” especially projects that would change the future if they succeeded.

How AI Changed the Deep-Tech Investment Landscape

The former Intel chief says artificial intelligence has dramatically expanded the opportunity set for deep-tech startups. In his telling, the AI boom has made the market for semiconductors, power systems, advanced materials, and manufacturing tools far larger than many investors expected only a year earlier.

That shift has drawn a flood of venture money into hard-tech fields. But Gelsinger believes many newcomers are underestimating how difficult it is to identify winners when the underlying science is still emerging.

He argues that the best deep-tech investors need more than enthusiasm for AI. They need the technical fluency to evaluate physical feasibility, the discipline to test claims, and enough experience to distinguish between a clever idea and a viable platform.

In his view, AI has not just created demand for software tools; it has created a market pull for the entire stack beneath those tools. That includes fabrication, memory, interconnects, energy delivery, and the specialized systems that will support future model training and inference.

What Makes Deep-Tech Due Diligence Different?

Deep-tech investing is different because the core idea may not yet have been demonstrated at commercial scale. According to Gelsinger, that means standard startup evaluation is not enough. A venture firm needs people who understand engineering constraints, physics, materials science, and manufacturing pathways.

At Playground, he said the investment team includes engineers, PhDs, and professors, and that diligence can involve extensive interviews and background checks. The aim is not simply to assess whether founders are smart, but whether they understand the hardest problem they are trying to solve.

He described the process as an attempt to identify the strongest team in a given area, rather than just one promising team among many. In fields at the edge of science, the challenge is often not whether an idea is plausible in the abstract, but whether one company has the combination of insight, talent, and execution needed to beat rivals to market.

Why Multiple Startups Often Chase the Same Scientific Idea

Gelsinger said it is unusual to find a single isolated startup working on a truly novel deep-tech idea. More often, two or three companies will converge on the same technical direction at roughly the same time.

That creates a competitive sorting problem for investors. The question is no longer whether the market exists, but which team can solve the science faster, scale manufacturing earlier, and build a product others cannot easily copy.

For venture firms, that means diligence has to go beyond pitch decks and financial projections. It has to include a working understanding of the scientific bottlenecks that will determine which company survives.

Why Lithography Is the Center of Gelsinger’s Thesis

Gelsinger’s biggest conviction is that lithography remains the heart of semiconductor progress. He believes the industry’s next breakthrough will come from light sources capable of printing finer chip features with greater precision.

Today’s leading lithography systems, dominated by Dutch supplier ASML, use extreme ultraviolet light at a wavelength of 13.5 nanometers. That technology is already extraordinarily advanced, but Gelsinger believes further progress will depend on pushing beyond that limit.

He frames the problem as one of physics as much as engineering: as transistor structures shrink, the difficulty and cost of manufacturing rise steeply. Moving to more advanced light sources could create a new route to density gains, allowing manufacturers to build faster, more efficient chips without relying solely on smaller transistors.

For Gelsinger, that is why xLight matters. He said the startup is not trying to replace ASML but to improve the machines that already dominate chip production. The logic is additive rather than adversarial: better light, better lithography, better chips.

How xLight Fits Into the Semiconductor Ecosystem

According to Gelsinger, xLight aims to pair novel light generation with existing manufacturing infrastructure. The idea is to make the current production base more capable before moving to even smaller wavelengths.

He suggested that free-electron laser approaches could eventually enable wavelengths in the 5-nanometer range and below, opening the door to further gains in chip density and performance.

The immediate payoff, though, would be more practical: improving the output of current lithography systems and extending the useful life of the industry’s most advanced tools.

That distinction matters because semiconductor manufacturing is not a clean-slate business. The leading fabs, supply chains, and equipment makers are built around enormous capital investment. Any new technology that can slot into those systems rather than replace them entirely has a clearer path to adoption.

How AI Chips Are Changing the Market

Gelsinger also sees the current wave of AI chip startups as a sign that the market is fragmenting beyond the GPU. Training large models has been the domain of graphics processors, but inference—the repeated use of trained models—may become a larger and more diverse market over time.

He argued that many of the most important chips of the coming years will not look like today’s GPUs. Instead, he expects a broader ecosystem of specialized processors and heterogeneous systems, where different types of silicon are combined to serve different workloads.

That shift is already visible in the market. Major model developers are trying to abstract their software away from specific hardware, while chip designers are pursuing architectures tailored to memory access, latency, and energy efficiency rather than raw parallel compute.

Gelsinger said that the next step is to make AI dramatically better, not just incrementally faster. In his view, the businesses that matter will be the ones that can deliver order-of-magnitude gains in cost, throughput, or energy efficiency.

What He Sees in Memory and Heterogeneous Compute

One of the biggest bottlenecks, he said, is memory. High-bandwidth memory remains the leading option for advanced AI systems, but he considers it a temporary solution rather than the end state.

He expects stacked memory architectures to become more important before the end of the decade. Startups such as d-Matrix, Fractile, and Cerebras, he said, are pushing at the boundaries of how compute and memory can be arranged in future AI hardware.

That matters because the economics of AI are changing quickly. If the new generation of hardware can produce many times more tokens for the same energy input, the winners will be the companies that can deliver those gains first.

For investors, the implication is clear: the market is no longer just about making the fastest chip. It is about designing systems that deliver usable performance at scale.

Why Energy Capacity Has Become an AI Bottleneck

Gelsinger argues that energy, not just silicon, is now one of the limiting factors in the AI economy. He said the United States has expanded power capacity only modestly over the past decade and called that a serious strategic failure.

His broader point is that digital growth depends on physical infrastructure. Large-scale AI systems require enormous amounts of electricity, and the grid is not keeping up with the speed of demand.

He said the challenge is complicated by the long timelines of energy buildout. New gas turbines can face multiyear supply constraints, nuclear plants take years to develop, and solar power depends heavily on global supply chains concentrated in China.

That is why he sees energy investment as inseparable from AI competitiveness. Without more generation, more transmission, and more efficient power use, even the best chips will hit a ceiling.

Gelsinger said that in the AI era, energy capacity is essentially economic capacity.

What Is Alva Energy Doing?

One company in Gelsinger’s orbit, Alva Energy, is focused on nuclear upgrading. His description suggests a two-track approach: extract more value from the existing nuclear fleet while also encouraging new nuclear construction.

The logic is pragmatic. If fresh generation takes too long, then increasing the productivity of today’s assets may be the fastest way to relieve the power crunch.

Gelsinger also pointed to work on voltage regulation and power conversion as part of the same broader effort. For him, the AI stack runs from chips to power delivery to infrastructure, and each layer needs innovation if the whole system is going to scale.

How Does Gelsinger View US AI Policy?

Gelsinger does not see a contradiction between US leadership in AI and stronger government involvement in the sector. He argues that Washington can use export controls, security requirements, and model oversight to shape the field without giving up the country’s strategic advantage.

When asked whether the US and China should compete for leadership in foundational models, he was unequivocal: he wants the United States and its allies to win. In his view, these systems should reflect Western values and be used to solve major global problems.

At the same time, he acknowledged that policy is moving in real time because the technology itself is evolving so quickly. New major models are appearing at a pace that makes static rulemaking difficult.

His central concern is not whether government has a role, but how that role is structured. He believes model developers should be transparent about testing and benchmarking, and that the public should know more about what went into training proprietary systems.

Should Governments Review AI Models Before Release?

Gelsinger did not give a blanket endorsement of a formal pre-release approval regime, but he did say that models need clearer integrity checks and visible testing standards.

He wants rigorous benchmarking, greater disclosure around training data, and assurance that frontier models meet security and alignment requirements before broad deployment.

In his view, if the industry cannot credibly police itself, regulators will eventually need to step in. He framed that as a practical matter rather than an ideological one: somebody has to make sure the review happens.

That position puts him in the middle of a broader debate over how to govern powerful AI systems. On one side are calls for speed and innovation; on the other are demands for safety, accountability, and international competitiveness. Gelsinger’s stance is that those goals can coexist only if testing becomes a non-negotiable part of the release process.

What This Means for the Semiconductor Race

Gelsinger’s move into venture capital says as much about the current chip race as it does about his own career. The industry is no longer defined only by the companies that design processors. It now includes startups working on light sources, memory, materials, energy systems, and manufacturing tools that could determine whether Moore’s law continues at all.

For decades, the semiconductor industry advanced through a predictable rhythm of miniaturization. That rhythm has become harder to sustain as engineering costs rise and physical limits approach. Gelsinger believes the next era will depend on whether the industry can solve the bottlenecks around lithography and power, not just architecture.

He also believes this is exactly the kind of moment that rewards technical investing. As AI expands the market for compute and as governments treat semiconductors as strategic assets, venture firms that can evaluate science rather than hype may have an edge.

That makes his role at Playground notable. It combines his experience as a chip executive with a new mandate to place bets on the technologies that could shape the next decade of computing.

Timeline: From Intel Exit to Deep-Tech Investor

The Bigger Picture: Betting on the Physical Layer of AI

Gelsinger’s thesis reaches beyond any single startup. He is betting that the most valuable breakthroughs in AI will come from the physical layer beneath the models: chips, lithography, memory, and power.

That is a major shift from the last decade, when software captured the bulk of venture attention. Now, as AI increases demand for infrastructure and exposes the limits of existing hardware, investors are rediscovering the economics of atoms, not just code.

Whether Gelsinger is right will depend on whether companies like xLight can turn laboratory promise into industrial reality. But his argument is becoming harder to dismiss: if AI is going to keep scaling, the industry will need new ways to make chips, move electricity, and build the systems that support both.

For Gelsinger, that makes the moment unusually exciting. He sees a rare alignment of technical possibility, market demand, and geopolitical urgency. In his telling, the task ahead is not merely to improve semiconductors. It is to restart the engine that made modern computing possible in the first place.