Xscape Photonics raised $81 million to replace copper wiring with light inside AI data centers

Xscape Photonics, a startup with offices in Fort Lee, New Jersey, and Santa Clara, California, raised $37 million in new funding on March 11, 2026, bringing its Series A total to $81 million, with participation from Nvidia and other investors, to address the data-center interconnect bottleneck by replacing copper wiring with optical technology. The company also launched FalconX, an external laser module compliant with the OIF ELSFP-02.0 standard, which fits into existing racks and can emit up to eight wavelengths from a single laser, aiming to solve the power and bandwidth limitations of copper in AI clusters. The funding round reflects growing investor interest in optical interconnects, as Nvidia also invested $2 billion in Coherent, Lumentum, and Marvell in March 2026.

Xscape Photonics' March funding round is still worth your attention, but not because the company suddenly made copper obsolete. The real story is that Nvidia and other AI infrastructure investors are now paying for the data-center wiring problem they can no longer ignore. Everyone watches the GPU. The constraint is increasingly the connection around it. As AI clusters stretch across racks and rooms and eventually whole campuses, the job is no longer only to buy faster accelerators. You have to move the data between them without wasting the power budget on copper. Xscape Photonics is one of the startups trying to make that shift practical. The company, which lists offices in Fort Lee, New Jersey, and Santa Clara, California, announced on March 11, 2026, that it had raised $37 million in new funding, bringing its Series A total to $81 million. Addition led the tranche, with existing investors including IAG Capital Partners and Nvidia continuing to participate, according to the company's Business Wire announcement. That matters for one blunt reason. Copper is running out of room. The physics is not new, but AI has made it urgent. Xscape says copper interconnects struggle beyond roughly 800 gigabits per second over more than a few meters, as resistance, crosstalk and power consumption all get worse at higher speeds. In a cluster running thousands of H100-class accelerators, the problem isn't only the chip at the center of the tray. It is the fabric that lets those chips act like one machine. FalconX is built to fit into racks hyperscalers already use Xscape launched FalconX alongside the funding news. The company describes it as an External Laser Small Form-factor Pluggable module that can emit up to eight O-band wavelengths, or colors of light, from a single off-the-shelf DFB pump laser. Its product page says the module produces more than one watt of optical power into an output fiber and complies with the OIF ELSFP-02.0 standard. That last detail is easy to skim past. Don't. A product that requires a hyperscaler to rebuild its rack architecture has a much harder path into production, however impressive the lab result looks. FalconX fits the rack you already have. That is what makes it interesting as a pluggable external laser source for AI fabrics - scale-up, scale-out, scale-across - not a science project waiting for the rest of the data center to catch up. The technology underneath it is Xscape's ChromX silicon photonics platform, built around its CombX approach for generating multiple wavelengths from one laser source. More wavelengths per fiber means more bandwidth through the same physical path. The company has also laid out a roadmap beyond eight wavelengths, including 16, 32 and eventually 128. That is the bet. Xscape has one manufacturing proof point that makes the company harder to dismiss. In August 2025, Tower Semiconductor and Xscape announced a prototype and validation kit for what they called the first on-chip, optically pumped, multi-wavelength laser source on Tower's PH18 silicon photonics platform. Tower described PH18 as a mature, high-volume platform. That's the phrase you want to see if you care about whether this technology can leave the conference booth and show up in real systems. Nvidia is funding the plumbing around its own chips Xscape is not alone. Ayar Labs closed a $500 million Series E on March 3, 2026, at a $3.75 billion valuation, with investors including Nvidia and AMD, according to its announcement. Ayar's approach is different: co-packaged optics placed close to the compute package itself. Xscape's external laser module is more modular. Both can be useful, because a modern AI cluster does not have one interconnect problem. It has several. Nvidia's behavior tells you where the market is going. In March, Nvidia announced $2 billion investments in Coherent and Lumentum. Later the same month it said it had invested $2 billion in Marvell as part of a partnership that includes silicon photonics work. In May, Nvidia and Corning announced a long-term optical connectivity partnership, with CNBC reporting the deal could include up to $3.2 billion of investment in Corning. You don't spend that kind of money around optical components unless the bottleneck is moving closer to the wire. SemiEngineering was blunter. In April it argued that all high-bandwidth AI data-center interconnects will become optical within five years. Grand View Research puts the broader optical interconnect market at $16.1 billion in 2024 and projects it will reach $34.5 billion by 2030. Forecasts can be wrong, of course. Purchase orders are harder to wave away. The academic roots help, too. Xscape was spun out of Columbia-linked photonics work, and its co-founders include Michal Lipson and Alexander Gaeta, both well-known names in the field. Columbia Engineering announced Lipson's election to the National Academy of Engineering in 2025, and Optica named Gaeta the 2026 Arthur L. Schawlow Prize in Laser Science recipient. In a market where the hard part is physics, credentials like that are not decoration. They are part of the product risk. Frankly, the funding number is the least interesting part of this story. Xscape's Series A total is $81 million, while MarketScreener, citing S&P Capital IQ, lists the company's total funding at about $95 million. Useful numbers, but the bigger signal is where the capital is gathering: around lasers, fibers and optical engines. The AI industry spent the last few years talking as if compute was the whole story. Now the companies selling the compute are paying to fix what surrounds it. Xscape may or may not be one of the winners. Startups in deep hardware have to survive qualification cycles and manufacturing realities - and customers who are brutal about reliability. But the direction is no longer subtle. If you want to understand the next phase of AI infrastructure, watch the light source, not only the chip. Also read: Private equity giants bet $170 million on the platform that sells their own funds to advisors https://startupfortune.com/private-equity-giants-bet-170-million-on-the-platform-that-sells-their-own-funds-to-advisors/ • How to Structure Founder Salary and Equity When You Hire a Fractional CFO https://startupfortune.com/how-to-structure-founder-salary-and-equity-when-you-hire-a-fractional-cfo/ • Simile raises $200 million at a $2 billion valuation to replace focus groups with AI-simulated humans https://startupfortune.com/simile-raises-200-million-at-a-2-billion-valuation-to-replace-focus-groups-with-ai-simulated-humans/