Photonics // AI Hardware // Founder Profile
Zaijun Chen
He is teaching computers to think in light. The physicist behind Opticore is betting the next AI chip runs on photons, not electrons.
The boss of light. Berkeley, California.
He is teaching computers to think in light. The physicist behind Opticore is betting the next AI chip runs on photons, not electrons.
Walk into most AI labs and you will find the same fight: too much heat, too little power, and a wall nobody can climb over. The wall has a name - the memory wall. Data crawls between memory and processor, burning electricity the whole way. Zaijun Chen looked at that wall and decided to stop pushing electrons through it. He decided to send light instead.
That is the whole bet behind Opticore, the company he co-founded in 2023 and now leads as CEO. The chips Opticore is building are called optical processing units, or OPUs. Instead of moving data as electric current through transistors, they convert it into optical signals and move it through waveguides - tiny channels that carry light across the silicon. The math happens by something the company calls photoelectric multiplication.
The claim is bold and specific: up to 100 times more energy efficient than leading GPUs, and 25 times denser. The architecture, Chen says, "isn't constrained by Moore's Law." It is a sentence built to provoke an industry that has spent sixty years worshipping that law.
What makes the claim worth a second look is where it came from. Opticore is not a slide deck wrapped around a buzzword. It grew out of a chain of peer-reviewed physics - a 2019 theory paper, a 2023 demonstration in Nature Photonics, a 2025 result in Science Advances. Chen did not start with a market. He started with an experiment that worked.
Before any of this, there were frequency combs. Chen earned his Ph.D. in physics at the Max Planck Institute of Quantum Optics and Ludwig Maximilian University of Munich, finishing in 2019 with the highest distinction. His work sat in the lab of Theodor Hansch - the man who shared the 2005 Nobel Prize in Physics for the optical frequency comb, a ruler made of light precise enough to measure the ticking of atoms.
It is a strange apprenticeship for a chip founder. Comb spectroscopy is about precision, not computation. But it taught Chen the one thing his company would later depend on: how to control light with absurd exactness, and how to read signals out of it without losing them to noise. The skill turned out to be portable.
In 2021 he crossed the Atlantic and joined MIT as a postdoctoral fellow in Dirk Englund's Quantum Photonics Group. There the question changed. No longer "how precisely can we measure light?" but "how much computing can we hide inside it?" Chen built photonic AI accelerators and neuromorphic devices. In 2022, he was first author on a Science paper with a title that reads like a manifesto - "Delocalized photonic deep learning on the internet's edge."
The idea was deceptively simple. Put a smart optical sensor where the data is born, let it detect and process light without first converting it to electricity, and you cut energy, latency, and data traffic by orders of magnitude. The sensor sees and thinks in the same breath. By 2023, the lab work had matured into a demonstration in Nature Photonics - the result that would become Opticore's technical spine.
The same year, Chen took a job that most founders would not dare hold at the same time: director of the Laboratory of Intelligent and Quantum Photonics at USC. He runs an academic group studying optical computing architectures and neuromorphic devices, while running a venture-backed startup chasing the commercial version of the same dream. Two lives, one obsession.
"We envision a future where data centers won't need their own power plants." - Zaijun Chen, CEO of Opticore
Here is the engineering sleight of hand that makes physicists lean forward. Opticore uses temporal multiplexing - it spreads computation across time, encoding neural data as pulses of light marching one after another. That lets the chip pull off something electronics struggles with: O(N-squared) performance from only O(N) physical devices. In plain terms, you get the work of a much larger chip out of a much smaller one.
The payoff is aimed straight at the memory wall. By turning high-bandwidth memory data into optical signals and doing the moving and the math on-chip in light, the design sidesteps the energy tax of shuttling electrons back and forth. Chen quotes a target of 100 TOPS per watt, with dynamic weights that allow real-time training as well as inference.
And crucially, it is meant to be buildable. Opticore says its chips are compatible with standard foundry processes on mature semiconductor nodes - no exotic fab, no science-fiction materials. That is the difference between a paper and a product. The company reports it completed tapeout demonstrations in 2025 and is targeting its first scaled system in 2026.
The vision Chen keeps returning to is almost utopian for an industry drowning in power bills: offload the heaviest AI workloads to photonic processors, and let the data center stop behaving like a small city. Whether the physics scales to that promise is the open question. The early results say it is worth asking.
His doctoral mentor, Theodor Hansch, won a Nobel Prize for inventing a ruler made of light. Chen now uses light to do arithmetic.
Opticore stands on a 2019 theory paper, a 2023 Nature Photonics demo, and a 2025 Science Advances result. Few startups can show the receipts.
He directs an academic lab at USC and runs a chip startup at the same time - science by day, scaling by night.
O(N-squared) computing from O(N) devices, by stretching the work across time with pulses of light.
The chips are designed for standard foundry processes on mature nodes - the unglamorous detail that makes it shippable.
The "memory wall" - the energy lost shuttling data around. Opticore's whole reason to exist is to walk around it.
Investors do not usually fund physics experiments. They fund margins. So the more telling part of Opticore's story is who showed up with checks. The company has raised $14.5 million across two seed rounds - an initial $7 million that closed around the end of 2024, then a $7.5 million extension in September 2025. Both rounds were co-led by Jetha Global and Origin Ventures, with Sagax Capital, Neotribe Ventures, Thunderbolt Ventures and Bioeconomy.XYZ joining the cap table.
The pattern matters. When the same lead investors come back to double down inside of a year, it usually means the milestones between the rounds were real. Between those two raises, Opticore says it completed chip tapeout demonstrations - the moment a design stops being a simulation and becomes silicon that can be tested. For a hardware company, that is the line between hope and hardware.
It also explains the timeline Chen keeps pointing at. Tapeout in 2025, a first scaled system in 2026. The roadmap is short and falsifiable, which is its own kind of confidence. Photonics has a long history of dazzling demos that never escaped the lab. Chen's wager is that compatibility with mature foundry nodes is the escape hatch nobody else used.
The competition is not gentle. Lightmatter, Lightelligence and a handful of others are chasing the same prize - optical compute for AI - and the incumbents selling GPUs are not standing still. What Opticore offers as its edge is the marriage of a clean scientific pedigree with a manufacturing story that does not require reinventing the fab. Whether that is enough to dent a market this large is the bet every name on the cap table is making.
Chen did not build Opticore alone, and the names around him are part of the credibility. His co-founders are Mengjie Yu and Ryan Hamerly. Hamerly is the through-line back to the science: the theoretical basis for the architecture traces to a 2019 paper in Physical Review X that he helped author. The advisor on the masthead is Dirk Englund, the MIT professor whose Quantum Photonics Group hosted Chen as a postdoc. The startup is, in a real sense, that lab's ideas wearing a business suit.
That continuity - from Englund's lab to Hamerly's theory to Chen's demonstration to a product roadmap - is rare. Most deep-tech startups have to reverse-engineer a scientific justification after the fact. Opticore had the science first and went looking for the company second. It is the difference between a story told forward and a story told backward.
It also shapes how Chen leads. He is a researcher who learned to run a profit-and-loss statement, not a salesperson who hired some physicists. The vocabulary he uses in interviews stays close to the lab - waveguides, temporal multiplexing, photoelectric multiplication - even when the audience is investors. He is selling a future, but he is selling it in the language of evidence.
Profile compiled from public sources: Opticore, USC, MIT, SPIE, Science Advances, and press coverage of Opticore's funding. Figures such as 100x efficiency and 25x density are company-stated claims.