Mian Zhang

A crystal that refused to be small, and the physicist who shrank it anyway

Lithium niobate has been the diva of the optics world for decades: dazzling on stage, impossible backstage. Engineers loved what it could do to a beam of light and hated how stubbornly it resisted miniaturization. Mian Zhang's entire career is a long argument with that contradiction - and lately he has been winning it.

Today Zhang is co-founder and CEO of HyperLight, a Cambridge, Massachusetts company that builds photonic integrated circuits - chips that move information as light rather than electrons. The customers he cares about most right now are the ones building AI: hyperscale data centers where the bottleneck is no longer raw compute but how fast you can shuttle data between racks without melting the power budget. HyperLight's pitch is that thin-film lithium niobate (TFLN) moves that data faster, on less voltage, in smaller packages than anything silicon can manage.

In March 2026 the company put numbers on the claim. It introduced 400G-per-lane photonic integrated circuits on its TFLN Chiplet platform, aimed squarely at next-generation AI networking, and announced a supply-chain alliance with foundry giant UMC, its compound-semiconductor arm Wavetek, and manufacturing heavyweight Jabil to push the technology to data-center scale. For a material long dismissed as a laboratory curiosity, that is a remarkable line-up of industrial muscle.

From Ithaca to a Harvard basement

Zhang earned his PhD at Cornell University in 2015, cutting his teeth on silicon photonics - the dominant, well-behaved workhorse of integrated optics. But silicon has a ceiling. It is a brilliant material for carrying light and a mediocre one for modulating it, the act of stamping data onto a beam. So Zhang went looking for something better and landed at Harvard's Laboratory for Nanoscale Optics, in the group of Marko Loncar, the Tiantsai Lin Professor of Electrical Engineering.

There, alongside collaborators including Kevin Luke and Christian Reimer, he did the work that would become the company's origin story. The team figured out how to fabricate thin films of lithium niobate into waveguides that looked and behaved like silicon photonics - but with the crystal's superior electro-optic punch intact. The headline result, published in Nature, was a roughly hundred-fold reduction in optical loss. In plain terms: they got light to travel a full meter across a chip with almost nothing leaking out.

That is the kind of number that makes other physicists sit up. It is also the kind that usually dies a quiet, celebrated death inside a journal. Zhang's refusal to let that happen is the more interesting part of his story.

The hard part is not the physics

Plenty of labs can make a beautiful one-off device. Almost none can make ten thousand identical ones that survive a customer's reliability tests. Zhang's bet - the one that turned a Nature paper into a business - was that manufacturability would decide the winner. HyperLight spun out of Harvard in 2018 with early backing from The Engine, the tough-tech fund seeded by MIT, and has spent the years since doing the unglamorous work of industrializing a crystal.

The payoff is a production line, not just a prototype. HyperLight now runs what it describes as the industry's first high-volume, qualified 6-inch TFLN manufacturing line, and in 2026 launched a modular TFLN Chiplet platform - an ecosystem meant to let other people build on its photonics the way the chip world builds on standard process nodes. The UMC, Wavetek and Jabil partnerships extend that reach to 6-inch and 8-inch wafers and to high-volume assembly.

Along the way the milestones have come in a steady cadence: the integrated TFLN electro-optic platform debuted at ECOC 2021; an industry-first hybrid transmitter with a DFB laser sitting on TFLN followed the same year; sub-volt modulators at 110 GHz, 65 GHz and 20 GHz arrived in 2024; and 145 GHz reference modulators pushing toward 448 Gbps per lane landed in early 2026. The $37 million Series B led by Summit Partners in September 2024 bought the runway to keep that pace.

The recognition, and the reticence

Zhang remains, at heart, a scientist who happens to run a company. He is a prolific author across integrated photonics, electro-optic frequency combs and nonclassical light, and was named to Electro Optics' Photonics100 list. He shows up at the right rooms - the Optica Photonic-Enabled Cloud Computing summit, ECOC - and lets the technology do most of the talking. His public quotes tend to be measured, even modest, which is unusual for a founder selling into the loudest market in technology.

That restraint is itself a tell. The companies that win in deep tech are rarely the ones with the best slogan; they are the ones still standing when the physics gets hard. Zhang has spent a decade proving he is comfortable in exactly that territory - the long, quiet stretch between a breakthrough and a product. The AI build-out just made his timing look brilliant.

Milestones worth framing