Pat Gelsinger’s Light-Based Bet to Save Moore’s Law

Pat Gelsinger isn’t done with chips. After getting pushed out of Intel’s top job, you’d think he’d retire to a vineyard or write a memoir. Instead, he’s backing a startup that wants to keep Moore’s Law alive with something unexpected: light.

The pitch is straightforward. For decades, we’ve made transistors smaller and packed more of them onto a chip. That’s been the engine of computing. But we’re hitting physical walls—heat, power, quantum tunneling, you name it. Meanwhile, AI models are getting hungrier by the day. They want more compute, more memory bandwidth, more everything. The traditional playbook is running out of steam.

Gelsinger’s answer is photonics. Instead of shuttling electrons across a chip, you use light. Photons don’t generate as much heat, they don’t have resistance, and they can carry way more data per second. It’s not a new idea—optical interconnects have been around in data centers for years—but the promise here is to bring photonics down to the chip level, where it’s always been too expensive and too hard to manufacture at scale.

I’ve seen this movie before. Companies like Ayar Labs and Lightmatter have been pushing chip-scale photonics for a while, with some serious venture backing. The difference now is that Gelsinger is attaching his name and reputation to it. That matters. He’s been around long enough to know where the bodies are buried, and he’s not going to stake his post-Intel legacy on something he thinks is pure hype.

The technical challenge is brutal, though. Photonics on a chip means integrating lasers, waveguides, and modulators into a silicon process that’s already mind-bogglingly complex. Yield rates are going to be low, costs are going to be high, and the ecosystem—software, packaging, testing—isn’t ready. This is not a two-year fix. It’s a ten-year slog, if it works at all.

But the potential payoff is enormous. If you can move data between compute cores using light instead of copper wires, you can build chips that are less power-hungry and faster at the same time. For AI training, which is bottlenecked by memory bandwidth more than raw compute, that could be a game-changer. You could train models that are currently impractical, or do it with a fraction of the energy.

Gelsinger is also making a bet on the broader ecosystem. He’s not just funding a lab experiment; he’s talking about manufacturing and deployment. That’s the hard part. Making one photonic chip in a cleanroom is one thing. Making a million with high yield is another. Intel’s own attempts at integrating photonics into its products have been cautious, and they’ve had their own struggles. Gelsinger knows the drill better than most, which is why he’s positioning this as a jumpstart, not a revolution.

There’s also a strategic angle. The US is desperate to maintain its edge in semiconductors, especially against China. Photonics could be a way to leapfrog some of the limitations of traditional silicon. If the US can own the photonic chip supply chain, that’s a huge geopolitical win. Gelsinger has always been a patriot, and this move fits that narrative.

I’m not going to pretend I know if this will work. The physics is sound, but the engineering is brutal. The history of photonics is littered with promises that didn’t pan out. But I like that Gelsinger is taking a swing. We need more of that in an industry that’s become risk-averse, especially when it comes to anything that isn’t an incremental shrink.

If this works, Moore’s Law doesn’t die—it just changes medium. If it doesn’t, we’ll still have AI, but it’ll be slower, more expensive, and more power-hungry than it could be. Either way, watching Gelsinger try is going to be interesting. He’s not done with chips, and neither is the industry. Light might just be the thing that keeps us going for another decade or two.