He spent 25 years shrinking laboratories into products. His next act: putting a multi-wavelength laser on a single chip - and convincing AI to run on light.
Inside an AI data center, the headlines belong to the chips. Matt Crowley is looking at the wiring between them. His argument is unfashionably blunt: a GPU can only run as fast as the fabric that feeds it, and the fabric is running out of copper. “Optics has reach measured in km and the highest speed copper has reach measured in cm,” he says. The fix Scintil is selling is a chip called LEAF Light - a single piece of silicon that fires a comb of precisely spaced laser wavelengths down a fiber, the optical equivalent of widening a one-lane road into a sixteen-lane highway without pouring more concrete.
Crowley took over as chief executive of SCINTIL Photonics in November 2024, moving from Qualcomm into a company most people outside the photonics world had never heard of. Scintil is fabless, headquartered in Grenoble with roots in Toronto and a freshly opened US arm, and it builds its parts in a standard silicon foundry using a heterogeneous integration recipe it calls SHIP. The pitch to hyperscalers is efficiency: roughly 6.4 terabits per second of edge bandwidth per millimeter, at about one-sixth the power of the pluggable optics in racks today.
What is unusual is that Crowley did not arrive to clean house. He joined the founder. Sylvie Menezo, who started Scintil in 2018, stepped sideways into the CTO and managing-director seats and kept driving the technology and customer partnerships. Crowley came to scale what she built. “I’m honored to join Sylvie in leading the company into its next chapter,” he said on day one - a sentence most incoming CEOs do not bother to write.
“The next gains in AI infrastructure will not come from how fast a single processor runs.”
Crowley grew up around shopkeepers and operators - his father and both grandfathers ran their own small businesses. The lesson stuck before he could name it. He went to Princeton and chose an unusual blend: physics married to the philosophy of science, equal parts how the world works and how we decide we know it. It is the kind of double major that does not obviously lead anywhere, and then leads everywhere.
In 1999 he landed at Boston University, not in a lab but in the venture-and-company-creation group. As director of the Office of Technology Development he ran an internal fund, co-invested alongside larger firms, and spent his days turning professors’ inventions into companies. He was learning the translation problem that would define his career: a brilliant result on a bench is not a business until someone can make it twice.
In 2007 he stopped advising founders and became one. Sand 9, a BU spinout, chased piezoelectric MEMS for mobile timing - tiny resonators to keep phones in sync. He raised more than $50 million and signed partnerships with Intel, Ericsson and Analog Devices. Then, around 2014, he got a hunch that voice was about to become the way people talk to machines, and that the microphones of the day were not ready. He cold-reached a researcher named Bobby Littrell, whose piezoelectric-acoustic MEMS work looked like the missing piece. That email became Vesper.
Vesper was where the translation problem nearly broke him. In early 2015 the manufacturing process controls turned out to be inadequate, and the company lost roughly eighteen months recovering. Crowley does not dress it up; he calls it a daunting technical hurdle cleared by teamwork and stubbornness. The payoff was real: Vesper’s rugged microphones shipped into consumer devices by the tens of millions, and Qualcomm bought the company. Crowley moved inside Qualcomm as senior director of business development, the founder now an operator within a giant. He had taken a research result all the way to volume and an exit - twice.
So when Scintil came looking for a CEO who could carry a breakthrough from demo to deployment, the resume was almost too on-the-nose. “Having recently led Vesper Technologies to high-volume shipments and through its successful acquisition, I understand the immense potential of breakthrough technologies in rapidly evolving markets,” he said. Pascal Langlois, Scintil’s co-founder and chairman, framed the hire in one phrase: a proven ability to scale up high-tech companies.
Crowley’s whole thesis lives in the gap between how far a signal can travel and how much power it burns getting there. Copper is cheap and excellent - until you need speed and distance at the same time. LEAF Light’s claim is that a single DWDM chip can deliver dense bandwidth at a fraction of the energy of today’s pluggable optics. The bars below visualize the headline figures Scintil puts forward.
The network is no longer supporting infrastructure. It has become part of the performance engine.
If it cannot be repeated at volume, it is not yet infrastructure.
There’s no way you can do that with copper. It can’t scale, it can’t go long distances, it can’t reach high data rates per channel.
When thousands of accelerators must behave like a single superprocessor, the quality of the fabric determines total compute capacity.
Father and both grandfathers ran their own shops. He learned the economics of making something twice before he learned the physics of making it once.
Vesper started because he read the voice-interface wave early and emailed the one researcher whose MEMS work could ride it. Conviction first, company second.
A 2015 manufacturing failure nearly sank Vesper. He talks about it openly - proof that he treats “repeatable at volume” as a hard-won belief, not a slogan.
The ambition is plain and large: make integrated photonics the default fabric of AI data centers, replacing power-hungry copper and pluggables with single-chip laser engines made at foundry volume. “This investment marks a pivotal moment for Scintil as we move to full-scale deployment,” he said when the Series B closed. The harder test is the one he set for himself years ago in a microphone fab - whether the thing can be built, reliably, a million times over.