Photonium Thinks Optical Design Is Stuck in the 1990s, and They Might Be Right

The Macro: Optics Is Everywhere, and the Tools Are Ancient

Every AR headset, every LiDAR sensor on a self-driving car, every quantum computing experiment, every biotech microscope relies on precision optical systems. The demand for complex optics has exploded over the last decade. The tools for designing those systems have not kept pace.

If you are building an optical system today, you are probably using Zemax OpticStudio or Code V. Both are powerful. Both are also expensive, slow, and fundamentally manual. An optical engineer loads a design, tweaks parameters, runs a simulation, studies the output, tweaks more parameters, runs another simulation. This loop can take weeks for a complex system. Months, if you are dealing with something at the edge of what is physically possible, like a waveguide for an AR display or a sensor array for a quantum computing rig.

The deeper problem is talent. There are not enough optical engineers. The skillset is niche. It sits at the intersection of physics, mathematics, and manufacturing knowledge that takes years to develop. Companies building products that require custom optics often find themselves waiting months just to get time with someone qualified to design what they need.

This is the kind of bottleneck that AI should be good at solving. Optical design is fundamentally an optimization problem with well-defined physics constraints. The math is known. The simulation tools exist. What is missing is a layer of intelligence that can explore the design space faster than a human can, and do it without requiring a PhD in optical engineering to operate.

The incumbents are not standing still. Zemax has added some automation features. Synopsys keeps updating Code V. But these are bolted-on improvements to tools that were architected before machine learning was a serious engineering discipline. The opportunity is to start from scratch with AI as the foundation, not an afterthought.

The Micro: Quantum Physicists Building Design Software

Photonium is a three-person team building AI-driven optical design software and consulting services. The founders are Jennifer Song and Adam Mhatre, and their backgrounds are precisely what you would want for this problem. Song comes from quantum computing research at Harvard, Stanford, and QuEra. Mhatre brings computational physics from Stanford. These are people who have personally experienced the pain of designing optical systems with inadequate tools.

They came through Y Combinator’s Spring 2025 batch. The product is a web-based optical engineering application that handles the full design stack: optimization, verification, sourcing, and prototyping. They are also running a consulting arm, which is a smart move for an early-stage company in a domain where trust matters and the sales cycle is long.

The target markets read like a wish list of industries that are growing fast and need better optics: AR/VR, quantum computing, biotech, metrology, and LiDAR. Each of these sectors is spending heavily on custom optical systems and struggling to find the engineering talent to design them.

What I find compelling about their approach is the scope. They are not just building a simulation tool that runs faster. They are trying to cover the entire workflow from initial design through to finding manufacturers and getting prototypes made. That is ambitious for a team of three, but if the AI layer actually works, it collapses what used to be a multi-vendor, multi-month process into something a single platform can handle.

The consulting business gives them two things: revenue and domain expertise. Every consulting engagement is a training data opportunity. Every client problem they solve by hand today is a workflow they can automate tomorrow. That flywheel is real, and it is how several successful B2B companies have bootstrapped their way into product-led growth.

The Verdict

I think Photonium is sitting on a genuine market gap. The optical design space is dominated by legacy tools, constrained by a talent shortage, and experiencing surging demand from multiple high-growth industries at the same time. That combination does not come along often.

The risk is execution breadth. Covering design, verification, sourcing, and prototyping is a lot of surface area for a three-person team. Zemax alone has hundreds of engineers maintaining a product that only handles part of that workflow. Photonium will need to be disciplined about which pieces of the stack they automate first and which they leave to the consulting arm.

Thirty days from now, I would want to know how many paying consulting clients they have and whether those engagements are generating reusable training data for the AI product. Sixty days, whether the web-based design tool is handling real client workloads or still in demo mode. Ninety days, the question is whether any of those consulting clients have converted to software-only customers. The path from “we solve your optics problems” to “our software solves your optics problems” is the entire company in miniature. If that conversion happens, Photonium has something. If it does not, they have a consulting firm with good branding.