Lightspring's $3.7M Bet to Make AI Chips Cheaper

·
Listen to this article~4 min

Lightspring raises over €3.5M ($3.7M) to industrialize photonic chip packaging with 3D lithography, targeting sub-0.5 dB coupling losses and slashing costs for AI infrastructure.

Photonics is the backbone of AI infrastructure. But there's a catch: optical packaging—the step that connects light to chips—is slow, expensive, and stuck in the lab. That's why Lightspring, a Besançon-based DeepTech startup, just raised over €3.5 million ($3.7 million) to change that. ### The Problem: Optical Packaging Is a Bottleneck Here's the thing: photonic integrated circuits (PICs) are manufactured by mature foundries, but assembly and optical packaging alone account for roughly 50% of a PIC's cost. Compare that to the foundry step, which is only about 20%. And fibre-to-chip coupling? Still done part by part with micrometre-scale active alignment. The result: 2–3 dB of typical optical loss, low throughput, and no standard. As AI infrastructure pushes optics closer to compute—think co-packaged optics—the hardest manufacturing problem moves into the package itself. "That step never industrialised," Lightspring noted in its press release. ### Lightspring's Solution: 3D Lithography on the Chip Lightspring's approach is elegantly simple: use 3D lithography (two-photon polymerisation, or 2PP) to fabricate optical coupling structures—microlenses, 3D waveguides, beam-shaping elements—directly on the chip. It's a single process step that handles fibre-to-chip, laser-to-chip, and chiplet-to-chiplet coupling, with target coupling losses below 0.5 dB per interface. No precision pre-alignment. No active alignment. And it's materials-agnostic and equipment-agnostic, with design cycles under two weeks (versus several months for bespoke optical packaging). > "The whole industry is working on making photonic chips faster, yet light still loses most of its energy the moment it enters or leaves the device. We decided to attack that specific problem: to build the optical interface the same way the rest of the chip is built, in one reproducible process step. This funding will turn a lab result into a qualified industrial building block." — Grégoire Bonnat, co-founder and CEO of Lightspring ### The Funding and the Team The €3.1 million ($3.3 million) Seed round was led by Atlantic and OVNI Capital, with participation from Concept Ventures, Ixcore, Plug and Play Ventures, and industry angels. Add a €350,000 ($370,000) grant from i-Lab (France's national DeepTech innovation competition) and other public funding, and Lightspring has secured over €3.5 million ($3.7 million) in dilutive and non-dilutive financing. Founded in 2026 by Grégoire Bonnat, Adrià Grabulosa, and Daniel Brunner, Lightspring is building the first 3D-fabricated photonic integration platform. The technology originates from research at the FEMTO-ST Institute in Besançon, a CNRS-UMLP joint research unit with over 700 staff. ### Why This Matters for AI Infrastructure Every generation of AI infrastructure pushes optics closer to compute. Co-packaged optics moves the optical interface off the faceplate and onto the substrate, next to the switch or accelerator. That shift moves the hardest manufacturing problem into the package itself. Lightspring's core IP—the first Process Design Kit (PDK) dedicated to 3D optical coupling, paired with software-driven design tools—aims to make that step scalable. Quentin Calleja, Principal at Atlantic, put it this way: "Photonics is scaling into AI infrastructure fast, but optical packaging remains a slow and expensive step. Lightspring has developed a technology that finally makes it scalable. We back teams with the potential to become the leaders of categories that don't exist yet." In short, Lightspring is turning a lab result into a qualified industrial building block. And if it works, it could make AI chips faster, cheaper, and more efficient.