Switzerland’s silicon photonics revival: how AI is giving the technology a second life

Before the rise of generative artificial intelligence, few investors paid attention to the optical components used in data centres. The spotlight was on processors, memory and software. Two years later, the situation is changing. As AI infrastructure grows in scale, another technology is returning to the forefront: silicon photonics. The €4.5 million funding round completed by Swiss startup Aylight illustrates this shift. While it does not mark the emergence of a new technology, it reveals a deeper transformation: AI could finally provide silicon photonics with a market capable of reaching an entirely new scale.

An established technology whose moment may finally have arrived

The earliest research into silicon photonics dates back to the 1980s, while major industrial programmes began taking shape in the early 2000s. The ambition was to use manufacturing processes inherited from microelectronics to produce components capable of manipulating light directly on silicon chips.

The idea is to combine two worlds. Electronics remains the most efficient technology for computation. Light, meanwhile, can transmit information with very little attenuation and far greater capacity when large volumes of data need to be moved.

Until now, the technology has primarily found applications in telecommunications networks and the optical links connecting servers within large data centres. Companies such as Intel have been investing in the field for more than a decade, while specialists including Coherent and Lumentum already supply many of the optical components used by network operators and hyperscalers.

Artificial intelligence has not invented silicon photonics, but it is profoundly changing the economic conditions for its adoption.

AI changes the nature of the problem

A GPU has value only if it can exchange data rapidly with the other processors in a cluster. The most advanced models are trained on tens of thousands of GPUs operating simultaneously. Future generations are expected to rely on several hundred thousand accelerators distributed across multiple buildings or even several campuses.

Data moves continuously within these architectures: models are synchronised, memory is exchanged, processors communicate and storage systems are supplied with information. Internal traffic is growing faster than computing power itself. One of the primary challenges is no longer simply to calculate faster, but to move more information without compromising performance or causing energy consumption to soar. This is precisely where silicon photonics becomes relevant again.

Replacing some electrons with photons

In a conventional electronic chip, information travels as electrical signals. For certain transmission functions, photonics uses pulses of light that propagate through waveguides etched directly onto a silicon chip. Computation itself remains the responsibility of electronic components.

This distinction is essential. Photonics does not replace electronics. It complements it when data exchanges become too extensive to be handled efficiently by electrical interconnects.

This architecture relies on several categories of components: waveguides that direct light, modulators that convert electrical signals into optical ones, photodetectors that perform the reverse operation, and multiplexing devices capable of transmitting several wavelengths simultaneously through the same fibre.

At the heart of this system lies an often-overlooked component: the laser.

Why lasers are becoming strategically important

Every optical link begins with a light source. Without a laser, there can be no optical transmission. Yet these components must meet several conflicting requirements: they must produce extremely stable light, consume little energy, be compact enough to integrate into networking equipment and remain compatible with large-scale industrial manufacturing.

This is precisely the part of the value chain targeted by Aylight. Emerging from research conducted at ETH Zurich, the startup is developing multi-wavelength lasers capable of generating several optical channels from a single chip. The approach aims to replace multiple individual lasers with an integrated architecture, reducing the number of components, power consumption and assembly complexity.

The objective is not to transform data-centre architecture, but to improve a component whose performance directly determines that of optical networks.

The fact that the technology has been designed for production in existing photonics foundries is also a decisive factor. In the semiconductor industry, the most promising innovations are rarely those requiring entirely new production lines, but those that can be integrated into existing industrial infrastructure.

A new layer of AI infrastructure

The rise of photonics reflects a broader shift across the semiconductor ecosystem. Every layer of the infrastructure is now becoming strategic.

Memory is advancing through the HBM chips produced by SK hynix, Samsung Electronics and Micron Technology. Networking is dominated by Broadcom, Cisco and Marvell Technology. Advanced packaging is attracting considerable investment at TSMC, while cooling systems are themselves becoming a source of differentiation.

Photonics forms part of this increasing fragmentation of the value chain. The shift is also encouraging the emergence of a new generation of specialised startups. Some are developing photonic engines, while others are working on optical interconnects, design software or packaging technologies. Scintil Photonics, EFFECT Photonics, Ayar Labs and Celestial AI illustrate this growing specialisation.

Industrialisation remains far from complete

The current interest in silicon photonics should not obscure the challenges that remain. The first concerns manufacturing itself. Silicon is an excellent material for guiding light, but it is not naturally an effective laser source. Many companies still need to combine different materials or integrate components manufactured separately, increasing industrial complexity.

The second challenge is economic. Electrical interconnects remain extremely competitive across many distances and use cases. Photonics will not replace them everywhere. Adoption will progress where bandwidth, distance or energy consumption justifies the initial additional cost.

Finally, the growth of this industry will also depend on manufacturing capacity. As with conventional semiconductors, value will be determined not only by component design but also by the ability to manufacture products at scale. Foundries capable of producing photonic chips in high volumes could become as strategically important as the manufacturers themselves.

Does Europe have an opportunity?

Unlike GPUs, a market largely dominated by the United States and Asia, silicon photonics remains a relatively open field. Europe possesses internationally recognised academic research, several centres of expertise in integrated optics and a particularly active startup ecosystem in Switzerland, France, the Netherlands and Belgium.

This scientific lead does not, however, guarantee industrial leadership. The challenge is no longer merely to develop the best components, but to secure the manufacturing capacity, industrial partners and commercial opportunities required to scale them.

Aylight was founded in 2025 by Bahareh Marzban and Dmitry Kazakov following their research at ETH Zurich. The company has raised €4.5 million in a pre-seed round co-led by Elaia and Swisscom Ventures, with participation from Verve Ventures and Plug and Play. The capital will finance its first foundry-manufactured prototypes, the expansion of its R&D teams and the accelerated industrialisation of its on-chip multi-wavelength laser technology.