Sign in Subscribe
Western Hemisphere

iPronics Raises US$125 Million to Speed Up AI Data Centers by Moving Data Between Chips With Light

The Editor

The Editor

— 4 min read
Share
iPronics CEO Christian Dupont CTO Daniel Pérez-López
iPronics CEO Christian Dupont CTO Daniel Pérez-López

iPronics, a silicon photonics company based in Valencia, Spain, raised US$125 million in a Series B round announced on September 2, 2026, bringing total funding to US$177 million. Maverick Silicon and Light Street Capital co-led the round, with participation from Nvidia, 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. Manish Muthal of Maverick Silicon and Geoffrey Tate, an advisor to Light Street Capital, are joining the board.

The company was founded in 2019 as a spin-off from the Universitat Politècnica de València and is led by chief executive Christian Dupont, with co-founder Daniel Pérez-López as chief technology officer. Its product, iPronics ONE, is a rack-mounted optical circuit switch built on programmable silicon photonics rather than the mechanical mirror arrays used in conventional optical switching. It reconfigures the physical connections inside an AI cluster in under a millisecond, which lets operators rearrange which GPUs talk to which as training and inference workloads change, instead of leaving processors waiting on a fixed network topology. iPronics says the approach shrinks existing optical circuit switch hardware by roughly 20 times and draws less power. The company has been shipping for about a year to large AI infrastructure providers it has not named, and will use the money to scale manufacturing and expand commercially, including from a new office in Santa Clara, California.

Market Context

Optical circuit switching went from a research curiosity to a purchasing line item in under two years. Google built its own optical switches internally and has reported roughly 40% power savings against electrical switching, and the rest of the industry is now trying to buy what Google made. Analysts tracking the segment describe revenue expanding from about US$100 million to US$400 million within a handful of quarters, one customer deal reportedly worth close to US$1 billion, and a total addressable market above US$3 billion. Nvidia's disclosed backlog for AI compute has been reported at US$1 trillion, and every dollar of that hardware is worth less if the network keeps the processors waiting.

iPronics is competing on a crowded field with different technical bets. Lumentum, Coherent, n-Eye and Salience Labs are also in optical circuit switching, while Nvidia, Broadcom, Ayar Labs, Intel, Marvell and Lightmatter are pushing co-packaged optics, which moves the optical interface onto the chip package rather than switching light between racks. Industry forecasts have all high-bandwidth data center interconnects going optical within roughly five years. Nvidia investing in iPronics while building competing capability is normal for the company, which routinely funds the ecosystem around its own roadmap, but it is a signal about the category rather than an endorsement of exclusivity.

The Number

"Data centers can fully utilize their GPUs without compromising performance." — Christian Dupont, CEO, iPronics

Sub-millisecond reconfiguration and a 20-fold reduction in equipment size are the two claims the entire commercial case rests on.

Regional Relevance

For the United States, this is a supply-chain story wearing a funding announcement. American hyperscalers are the customers, American capital co-led the round, Nvidia is on the cap table and the company is opening in Santa Clara. But the technology and the manufacturing base are European. That is the pattern the US CHIPS program was designed to avoid, and photonics is one of the few semiconductor-adjacent areas where Europe holds genuine research depth. The practical read for US buyers is that the bottleneck in AI infrastructure has moved from compute to interconnect, and that whoever solves the interconnect gets pricing power over the most capital-intensive buildout in the industry's history.

For Spain and the European Union, iPronics is the outcome the bloc's deeptech policy has been chasing for a decade. A university spin-off from Valencia, backed early by the European Innovation Council Fund, reaching a US$125 million round with US strategic investors is a rare complete sequence. Roughly €107.6 million at current rates makes it one of the larger Spanish deeptech rounds on record, in a country better known for exporting founders than for keeping hardware companies at home.

The question Europe usually loses on comes next. The capital is now majority American, the commercial center of gravity is shifting to California, and the customers are US hyperscalers. Whether the manufacturing and the intellectual property stay in Valencia, or follow the money and the demand across the Atlantic, is the part worth watching over the next two years.

The Other Side

Is optical circuit switching the winning architecture, or an interim one? Co-packaged optics attacks the same problem by eliminating the electrical-to-optical conversion at the chip rather than switching light between racks. If Nvidia and Broadcom deliver CPO at volume, the value may migrate onto the accelerator package and leave rack-level switching as a smaller market than the current forecasts assume.

How much of the performance claim is verifiable? iPronics has not named a customer, published independent benchmarks or disclosed deployment scale beyond "shipping for about a year." Sub-millisecond reconfiguration and a 20-fold size reduction are strong claims in a field where hyperscalers test for years before committing. Buyers will want third-party numbers before the round's implied expectations are met.

What does Nvidia's investment actually buy? Nvidia funds companies across its ecosystem, including some it later competes with directly. Its participation validates the problem and gives iPronics access, but it carries no commitment to purchase and no protection if Nvidia decides to build the same function into its own networking stack. Concentrated dependence on a handful of hyperscale buyers, several of whom are also potential competitors, is the structural risk in this business.

Sources & Transparency

Read more