The race to build ever-larger AI data centers has a new pressure point, and it is not the chips themselves. On October 14, 2026, Salience Labs, a photonics company spun out of research at the University of Oxford, announced an oversubscribed $125 million Series B round to scale its all-optical network switches β the kind of hardware that decides whether thousands of GPUs in a cluster actually behave like one computer, according to the company's funding announcement.
The round was led by Bullhound Capital, with Callum Stewart, a principal at the firm, joining the Salience Labs board of directors. What makes the investor list notable is who else showed up: AMD Ventures and Intel Capital, the venture arms of the two biggest rival chipmakers, both took part β a rare moment of agreement between arch-competitors. Other new backers include SBI Group, the trading firm Hudson River Trading, and the British Business Bank. Existing investors Applied Ventures LLC, Oxford Science Enterprises, Cambridge Innovation Capital, and Silicon Catalyst Ventures all followed on, according to reporting from Photonics Spectra.
An optical circuit switching bet on the AI networking bottleneck
Salience Labs builds optical circuit switching (OCS) technology β network switches that route data as light, end to end, without converting it to electrical signals at every hop. In today's AI clusters, data leaves a GPU as light over fiber, is converted back into electrons at each switch, routed, and turned back into light again. That round trip repeats hundreds of times across a large cluster, and each conversion burns power and adds latency.
For the scale-up networks that connect thousands of GPUs across multiple racks, that conversion tax is becoming the binding constraint. The problem is one of token economics: as workloads demand more from each accelerator, the networking layer determines how much useful work a dollar of power actually buys. Salience Labs' fully integrated, all-optical switch architecture sits on a single silicon photonics chip, which the company describes as designed to scale to high-volume manufacturing β a claim the new funding is meant to turn into reality.
The timing is deliberate. Salience Labs recently demonstrated a 64-port all-optical photonic switch live at ECOC 2026, which the company described as an industry first for silicon photonics optical circuit switching, according to coverage from Sesamers. The Series B money will go toward accelerating deployment of that OCS technology, scaling volume manufacturing operations, and developing larger 128-port and 256-port versions of the switch. The company also plans to expand its U.S. presence with a second West Coast site in Silicon Valley.
From Oxford research to volume production
Salience Labs was founded in 2021 in Oxfordshire, England, and its technology draws on more than a decade of photonics research associated with the University of Oxford and the University of MΓΌnster. It is a 65-person team, and the capital is meant to take it from demonstration to production. Manufacturing scale is the hard part of this story: optical networking technologies must meet demanding commercial requirements on yields, reliability, and integration with existing infrastructure before data center operators will trust them, as industry analysts have noted.
The company is not starting from zero on manufacturing. Earlier this year it entered into a manufacturing partnership with Tower Semiconductor, and it says it will continue working with partners including Keysight Technologies to scale its operations for high volume. The round follows a $30 million raise last February led by the venture arm of Applied Materials, which means Salience Labs has now told the market β twice in a year β that its path to volume is credible enough to keep raising on.
Competition will not be gentle. Established networking vendors and a growing field of photonics startups are all chasing the same bottleneck, and Salience Labs will need to show measurable economic advantages over conventional switches, not just elegant physics. But the signal from this round is hard to misread: when the investment arms of both AMD and Intel write checks to the same 65-person Oxford photonics company, the industry's biggest players agree the next leap in AI data centers will be measured in light, not electrons.
Comments 0
No comments yet. Be the first to share your thoughts!
Leave a comment
Share your thoughts. Your email will not be published.