Aylight raises €4.5 million to solve AI data centres' laser bottleneck from Zürich

Aylight, an ETH Zürich spinout, has raised €4.5 million in a pre-seed round co-led by Elaia (Paris) and Swisscom Ventures (Bern), with participation from Verve Ventures and Plug and Play. The company was founded in 2025 by Bahareh Marzban (CEO) and Dmitry Kazakov (co-founder) and is developing chip-scale multiwavelength laser technology designed to address one of the less-visible constraints on AI data centre scaling: the optical interconnect bottleneck.

The laser problem in AI infrastructure

Moving data between chips inside an AI data centre — between GPUs, between training and inference nodes, between storage and compute — requires interconnects. Today most of those interconnects use electrical signals. Electrical interconnects are adequate for smaller operations, but as AI training clusters grow from rack-scale to building-scale, the bandwidth demand, latency constraints, and power draw of electrical wiring become binding. Optical interconnects — light rather than electricity — offer higher bandwidth and lower energy per transmitted bit, but they require lasers, which have historically been expensive, power-hungry, and difficult to integrate at chip scale.

The current approach to on-chip optical interconnects typically requires an array of separate lasers — one per wavelength channel — which multiplies the component count, cost, and integration complexity. Aylight's technical bet is a frequency-modulated comb laser that generates multiple wavelengths simultaneously from a single chip-scale device, reducing the component count to one. The commercial signal that this matters is Nvidia's reported commitment of approximately $4 billion to addressing the laser shortage in its own interconnect supply chain.

Marzban's framing of the company's founding logic is direct: "We started from a problem rather than a technology: the laser had become one of the constraints on scaling AI infrastructure."

The ETH Zürich advantage

Aylight's provenance matters. ETH Zürich's photonics and integrated-circuit research groups are among the strongest in Europe, producing spinouts that carry genuine IP alongside their founding teams. The pre-seed capital goes toward two specific deliverables: developing the first semiconductor-foundry prototypes — demonstrating that the device can be manufactured in an industrial fab rather than a university cleanroom — and expanding the R&D team.

The co-lead structure is strategically coherent. Elaia is a Paris-based deep-tech fund with an established track record in European hard-tech companies; it understands the gap between academic proof of concept and manufacturable product. Swisscom Ventures, the VC arm of Switzerland's leading telco, adds both capital and a commercial anchor in an industry that will be one of the first customers for optical interconnect infrastructure. Verve Ventures and Plug and Play extend the network.

The milestone the pre-seed is designed to prove is not commercialisation — it is whether Aylight's comb laser performs at foundry-grade tolerances, the engineering challenge that separates a working university prototype from a chip that can be manufactured reliably. If that proof lands, the addressable market is every AI data centre being designed or expanded in the next decade.