The Grenoble company quietly putting an entire dense-wavelength laser source onto a single chip - and feeding AI's bottomless appetite for bandwidth.
Across the AI industry, the expensive chips are mostly fine. The problem is getting them to talk to each other fast enough, and cheaply enough, without the building melting. Copper runs out of breath. Pluggable optics burn power. And every extra picojoule spent moving a bit is a picojoule not spent thinking.
SCINTIL Photonics builds the part that fixes this. It is a fabless silicon photonics company, headquartered in Grenoble and expanding into the United States, that designs laser light engines for AI data centers. Its flagship, LEAF Light, is described as the industry's first single-chip, DWDM-native light engine - a slab of silicon that carries many wavelengths of laser light at once. In September 2025 the company closed a $58M Series B with NVIDIA on the cap table. For a 36-person company, that is a loud vote of confidence in a quiet idea.
Silicon is a marvelous material for routing light and almost useless at making it. The lasers themselves want exotic III-V semiconductors - indium phosphide, gallium arsenide - that do not grow on silicon and historically had to be glued on afterward, one painstaking part at a time. That assembly step is where cost, power, and yield go to die.
So the industry mostly settled for compromise: separate laser modules, pluggable transceivers, lots of discrete components, lots of wasted watts. It works. It also does not scale gracefully into a world where a single AI cluster wants petabits of bandwidth moving between racks.
SCINTIL's read was blunt: co-packaged optics will not win on a slide. It wins when it can be made, in volume, in a foundry, without bespoke heroics on every chip. The missing piece was a process - not a prototype.
SCINTIL was founded in 2018 by Sylvie Menezo, who spun the company out of CEA-Leti, the European research institute that is to photonics roughly what a good vineyard is to wine. Her bet was a process she calls SHIP - Scintil Heterogeneous Integrated Photonics - which uses a BackSide-on-BOX bonding technique to place III-V laser material onto silicon and integrate the lasers, modulators, and detectors on one chip.
The clever part is not just that it works. It is that it is designed to ride standard silicon-photonics fabrication, the kind foundries already run. CEO Matt Crowley, who now leads the company from the U.S. side, frames it as a manufacturability story first and a physics story second - which, in semiconductors, is the only story that ends in revenue.
The headline product is LEAF Light: a single-chip, DWDM-native light engine for co-packaged optics. DWDM - dense wavelength division multiplexing - means stuffing many colors of light down one fiber at once, which is how you get absurd bandwidth without absurd fiber counts. SCINTIL puts the dense multi-wavelength source on a single die.
Scintil Heterogeneous Integrated Photonics: DFB lasers, silicon modulators, germanium photodetectors and waveguides, monolithically integrated via BackSide-on-BOX III-V bonding on standard silicon photonics.
The industry's first single-chip, DWDM-native light engine. Up to 6.4 Tbps/mm edge bandwidth density, transmission exceeding 1 Tbps per fiber, at roughly one-sixth the power of conventional pluggable optics.
An evaluation kit that lets customers test and qualify the LEAF Light engine inside their own systems before designing it in.
Sylvie Menezo founds SCINTIL Photonics in Grenoble to commercialize heterogeneous integrated photonics.
First institutional funding from Supernova Invest, Innovacom and Bpifrance to prove out the SHIP process.
Series A backing brings in Bosch Ventures, Applied Ventures and ITIC-Taiwan as the team develops integrated laser sources and chases foundry qualification.
Led by Yotta Capital Partners and NGP Capital, with NVIDIA and BNP Paribas Développement, to scale production of the single-chip DWDM light engine and expand into the U.S.
Conviction is cheap; capital is less so. SCINTIL's Series B drew lead checks from Yotta Capital Partners and NGP Capital, the strategic interest of NVIDIA, and continued backing from Supernova Invest, Bpifrance, Innovacom, Bosch Ventures and the Applied Ventures ITIC fund. The money is earmarked for production scale-up and hiring across France and the United States.
SCINTIL's stated mission is to revolutionize AI scale-up interconnects with dense multi-wavelength laser sources that are faster, denser and far more power-efficient - while cutting data center operating costs and carbon footprint. That last clause is not garnish. At hyperscale, power is the budget. Shaving the energy cost of moving data is, in practice, an environmental story and a margin story wearing the same coat.
The customers are the people building AI factories: hyperscalers, GPU cluster designers, co-packaged optics integrators and optical transceiver makers. The model is fabless B2B - SCINTIL designs the chips, foundry partners build them, and evaluation kits get the technology into customers' hands. Competitors range from photonics specialists like Ayar Labs, Lightmatter and Sicoya to incumbent component vendors; SCINTIL's wedge is a manufacturable, single-chip laser source rather than a science project.
The chip that was waiting on a wire does not care about cap tables or process technology. It cares about how fast the next byte arrives and how much power that costs. If SCINTIL is right - and a roomful of serious investors is betting it is - that byte arrives on a beam of multi-wavelength light from a single chip, at a fraction of today's energy, with fewer parts to fail.
That is the whole company in one sentence. The hard work was making the impossible part - lasers on silicon - boring enough to manufacture. The payoff, if it scales, is an AI infrastructure that spends less of its energy talking and more of it thinking. Somewhere in a data center, a GPU stops waiting.