iPronics: Spanish Photonics Startup Raises $125 Million From Nvidia and Others

Set Trending Topics as a preferred source on Google.

While most of the money in AI still flows toward GPUs, a growing share is going into a quieter question: how those GPUs are wired together in the first place. iPronics, a startup based in Valencia, Spain, has just announced a $125 million Series B round with participation from Nvidia. The round is co-led by Maverick Silicon and Light Street Capital, with new and existing backers including Triatomic Capital, Bosch Ventures, Catalight Capital, the European Innovation Council Fund, The Tate Family Trust, Fine Structure Ventures, Amadeus Capital Partners, Build Collective and Criteria Venture Tech. Total funding to date now stands at $177 million.

For a European deeptech company, that ranks among the larger rounds of the year, and it lands in a stretch where other photonics firms on the continent are drawing capital as well: German startup Q.ANT recently raised 62 million euros for photonic AI chips.

What iPronics Builds

The product is called iPronics ONE, an optical circuit switch (OCS): a rack-mounted device that reroutes data paths in a data center purely in the optical domain, with no detour through electrical signals. The technology became widely known through Google, which uses optical switches in its TPU clusters and solved a good part of the cabling problem in its AI supercomputers that way. That success set off a wave of startups aiming to make the same capability available to outside customers.

The central advantage of an OCS is resilience. When a single chip fails in a cluster of hundreds of thousands, traffic can be steered onto a different path in a fraction of a second, keeping the training run alive. “The failures are happening on a scale of minutes now,” founder and CTO Daniel Pérez-López told Reuters. “The first motivation is resiliency: a robust and reliable network that can basically rewire itself.”

The first generation of the product, the ONE-32, is a 32-port switch in a compact 1U form factor. iPronics quotes a latency below 30 nanoseconds and a reconfiguration time under 300 microseconds. Because the design cuts the number of transceivers required, switching power consumption is said to drop by as much as 50 percent. The company has been shipping the unit for about a year, according to its own account to major AI infrastructure providers whose names it declines to give.

The software layer is where iPronics positions its real differentiator. ONE ships with integrated control, telemetry and APIs, so the optical layer can be addressed directly from cluster management. That lets an AI cluster reshape its connectivity topology in real time, depending on whether a training run or inference traffic is on the wire.

The fresh capital will go toward shrinking the hardware by roughly a factor of 20. The logic is straightforward: the more switches fit into a server rack, the larger the clusters they can stitch together. The rest funds scaling operations and commercial rollout. “With this investment, we can significantly accelerate our commercialization, delivering the scale our customers need to drive broad adoption,” says CEO Christian Dupont. His co-founder Pérez-López points to the reference customers: “Together with our tier-one partners we are successfully demonstrating that we can build and deploy this technology at scale for AI infrastructure.”

Who Is Behind the Company

iPronics was founded as a spin-off from the Universitat Politècnica de València, one of Europe’s strongest centers for integrated photonics. Co-founder and CTO Daniel Pérez-López comes straight out of that research lineage: programmable photonic circuits, meaning chips whose light paths can be reconfigured in software instead of requiring a purpose-built chip for every application. The company’s first product, SmartLight, was exactly such a programmable photonic processor and went to more than 20 customers in research and industry.

Making the leap from research instrument to data center product is the job of CEO Christian Dupont, who has been at the helm for about two years. He brings more than 30 years in the semiconductor industry, ran wireless as general manager for the U.S. and Europe at Texas Instruments, and previously led several optical MEMS startups. He holds an engineering degree from EPFL in Lausanne.

More telling than the list of investors is who is joining the company around this round. Manish Muthal, senior managing director at Maverick Silicon, takes a board seat, as does Geoffrey Tate, an advisor to Light Street Capital. Tate is an institution in the semiconductor world: senior VP of microprocessors at AMD, then founding CEO of Rambus, which he ran for over a decade and took public, and most recently co-founder and CEO of AI inference specialist Flex Logix, acquired by Analog Devices. He also sits on the board of Ayar Labs, one of the best-known players in optical chip-to-chip interconnect.

Young Sohn of Catalight Capital, which runs a dedicated AI infrastructure fund, joins as an advisor. Sohn served as corporate president and chief strategy officer at Samsung and as CEO of Inphi, and sits on the boards of Arm and Cadence Design Systems. “Next-generation AI infrastructure will require a more dynamic, optical-first network fabric,” he says of his involvement.

On the operational side, iPronics has opened an office in Santa Clara, California, is hiring there for product strategy, customer deployments and AI infrastructure partnerships, and has set up a dedicated packaging and assembly line with contract manufacturer Fabrinet for its next OCS generation.

Why This Matters for the Industry

The data centers going up right now hit the limits of the network and the power supply well before they hit the limits of raw compute. An AI training run spread across tens of thousands of accelerators is fundamentally a communication problem: the chips spend a substantial share of their time waiting on each other. Every percentage point of additional GPU utilization translates into hard money at billion-dollar cluster scale, and every watt saved in the network stays available for the compute chips.

Underneath that sits a materials shift now sweeping the industry. High-speed links inside a rack have run over copper, an approach reaching its physical limits at the bandwidths and distances of today’s clusters. The move to optics in the scale-up domain, meaning the tightly coupled accelerator fabric, creates demand for a new class of high-density switches. That is precisely the gap iPronics is aiming at. “The explosive growth of AI workloads has pushed traditional network infrastructure to its power and bandwidth limits,” says Muthal of Maverick Silicon. Light Street partner Shef Osborn puts the investment thesis more bluntly: iPronics offers the most scalable and cost-effective approach to solid-state optical switching at the interconnect layer.

That context is what makes Nvidia’s participation more than a financial bet. Nvidia is pushing optical networking itself with its own photonics switches and has a concrete interest in seeing the infrastructure around its accelerators keep pace. The company is currently investing broadly across the ecosystem of its own demand, most recently in orbital AI data centers.

For Europe, the case is interesting on different grounds. The debate over technological sovereignty tends to circle logic chips and fabrication capacity, where the gap is close to unbridgeable. Integrated photonics is one of the few fields where European institutes and startups are genuinely competitive. That a spin-off from a Spanish university is now plugging into the network architecture of the world’s largest AI clusters shows where that leverage sits. At the same time, the round follows a familiar pattern: growth capital comes largely from the U.S., and the company expands to California. The same dynamic was visible recently at British chip unicorn Olix, which raised $312 million for AI inference.

Until recently, the market for optical circuit switching was in practice a one-company affair belonging to Google. It is opening up to everyone else right now, and iPronics has bought itself a strong starting position with this round.