OpenLight

Here is a fact about silicon that is both very useful and very inconvenient: it is a wonderful material for building transistors and a terrible material for making light. This is the central tension of the entire silicon photonics industry, and OpenLight, a roughly 48-person company in Goleta, California, has built its business on a specific answer to it.

The answer is heterogeneous integration, which is a fancy way of saying: bond a sliver of indium phosphide - a material that is good at making light - directly onto the silicon, and now you have a chip that can both compute-adjacent things and lase. OpenLight calls the result a PASIC, a photonic application-specific integrated circuit. If you know what an ASIC is, you already know the joke. It is an ASIC, but for light.

Why does anyone care? Because of a problem that gets more expensive every quarter. AI data centers are stuffed with extraordinarily fast chips that need to talk to each other, and the traditional way of moving that data - electrons through copper - runs into physics. Copper gets hot, loses signal, and eats power at exactly the distances and speeds AI wants. Light does not have this problem in the same way. So the industry is racing to replace as much of the electrical plumbing as it can with optical plumbing, and the plumbing has to be small, cheap, and manufacturable at enormous scale.

The "Open" is the strategy

Most photonics companies would build this platform and guard it. OpenLight does the opposite, and it is right there in the name. Rather than selling one finished widget, OpenLight licenses a process design kit - a PDK, plus a library of validated building blocks: integrated lasers, modulators, amplifiers, detectors - and lets customers design their own chips on it. It then helps get those chips built through a foundry partner. The closest analogy is ARM, the company that licenses chip designs to basically everyone rather than manufacturing its own. OpenLight wants to be that, for photons.

This is a genuinely interesting business decision, because it means OpenLight's success depends on an ecosystem existing rather than on any single product it ships. The PDK has been validated at Tower Semiconductor, a commercial foundry, which is the unglamorous but load-bearing part of the story: a design that works in simulation is a science project, and a design that works on a foundry's production line is a product. OpenLight spends a lot of effort making sure it is the latter.

Where it came from

The technology did not appear in 2022. It has a lineage. The core heterogeneous-integration work traces back to Aurrion, a spin-out from the University of California, Santa Barbara - which has quietly been one of the world's most important photonics research clusters for two decades. Juniper Networks acquired Aurrion in 2016, folded the technology in, and then in 2022 spun the photonics unit back out as an independent company following a collaboration with Synopsys, the electronic-design-automation giant whose tools help customers actually lay out these chips. So OpenLight is simultaneously a 2022 startup and a twenty-year research program, depending on how you count.

Running it is Dr. Adam Carter, who became CEO in early 2023 and whose resume reads like a bet placed over and over on the same idea. He ran the transceiver group at Cisco and was involved in Cisco's acquisition of Lightwire, another silicon photonics startup. He was on the senior team at Oclaro when Lumentum bought it for $1.85 billion. The through-line is optics winning the data center. OpenLight is the largest version of that bet he has made.

The physics, briefly and without the headache

It helps to understand what a PASIC is actually doing, because the value proposition falls out of it. In a conventional optical link, you have a laser in one package, a modulator that imprints data onto that laser's light in another, and a photodetector somewhere on the far end that turns the light back into an electrical signal - plus a small forest of fibers, lenses and alignment steps connecting them. Every one of those interfaces is a place to lose light, add cost, and introduce a manufacturing yield problem. Integration collapses that forest onto a single die. The laser, the modulator, the amplifier and the detector are fabricated together, aligned by lithography rather than by hand, and packaged as one component.

This is the same argument that has driven the entire semiconductor industry for fifty years, just pointed at light instead of logic. Discrete transistors gave way to integrated circuits; integrated circuits gave way to systems-on-chip. Each step traded some flexibility for enormous gains in cost, power and density, and each step looked, at the time, like it might be too hard to manufacture at scale. OpenLight is placing the same wager on photonics: that integration is not merely a nice-to-have but the thing that eventually decides who wins.

The company it keeps, and competes with

OpenLight is not alone in this belief, which is either reassuring or alarming depending on your temperament. Intel has run a silicon photonics program for years. Startups like Ayar Labs and Lightmatter are attacking the co-packaged-optics problem from the compute side. POET Technologies, Rockley Photonics and foundry programs at GlobalFoundries and imec are all circling the same opportunity. What distinguishes OpenLight is less the physics - heterogeneous integration is a known technique - and more the go-to-market: it has decided to be the neutral, licensable platform rather than a vertically integrated competitor to its own customers. In a field where everyone needs the same building blocks and nobody wants to reinvent a laser, being the shared foundation is a defensible place to stand, provided the ecosystem shows up.

The team reflects the strategy. This is a small, engineering-led, venture-backed company - roughly 48 people - staffed heavily with veterans of the optical-networking industry and rooted in the UC Santa Barbara photonics community that produced the founding technology. It is not trying to be a consumer brand. Its customers are other companies deep in the supply chain, and its product is, in a real sense, other companies' ability to ship product.

What you can actually do with it

If you are a company that makes optical transceivers, telecom gear, LiDAR sensors, or the increasingly optical guts of an AI cluster, OpenLight offers three doors. You can license the PDK and design your own PASIC in-house. You can hand OpenLight your requirements and have it design and lay out the chip for you. Or you can go straight to production of a proven design. The pitch is that in all three cases, you are starting from validated building blocks rather than inventing photonics from scratch - which, given how hard photonics is, is a meaningful head start.

And the market it is chasing is not only the data center, even if that is where the urgency and the headlines are. The same integrated-photonics platform is aimed at telecom, where long-haul and access networks have been optical for decades and are hungry for cheaper, denser components; at LiDAR and automotive, where solid-state optical sensing needs lasers and detectors small and rugged enough to sit in a car; and at quantum computing, where photons are not just the wiring but sometimes the qubits themselves. A single validated platform that can address several of these markets is worth more than a point solution for any one of them, because it spreads the enormous fixed cost of getting photonics to work across more customers. That is the quiet financial logic underneath the "open" branding.

The money agrees, at least so far. OpenLight raised a $34 million Series A in 2025, then a $50 million Series A-1 in April 2026 led by Matter Venture Partners - an oversubscribed round, the company says - bringing the total to $84 million. That is not hyperscaler money; it is a deep-tech company's fuel to turn a validated platform into deployed volume. Whether light wins the data center is not really in doubt anymore. The interesting question is who supplies the picks and shovels, and OpenLight has decided the answer is: whoever makes the platform everyone else builds on.