The $125M Bet to Break the Copper Barrier in Hyperscale AI Datacenters

OXFORD, England – October 14, 2026 – The engines of the modern economy are increasingly housed within massive, windowless warehouses humming with the processing power of artificial intelligence. Yet, as these hyperscale datacenters grow, they are colliding with the immovable laws of physics. The copper cables and electrical switches that bind millions of processors together are running too hot, consuming too much power, and transmitting data too slowly.

Today, Salience Labs announced a definitive step toward dismantling that bottleneck, securing an oversubscribed $125 million Series B funding round to commercialize its all-optical circuit switching technology. Led by Bullhound Capital, the financing injects crucial capital into a transatlantic effort to rewrite the networking layer of AI infrastructure.

"The investor base and customer relationships Salience Labs has developed really speak to the capability of the leadership," said Bullhound Capital Principal Callum Stewart, who joins the company's board of directors. "Bullhound Capital is proud to back this transformational company that is bringing to reality the next generation of high-performance data centers."

However, the most telling detail of the round isn't the dollar amount. It is the rare, synchronized participation of two of the semiconductor industry’s fiercest competitors: AMD Ventures and Intel Capital. Alongside the British Business Bank, SBI Group, and Hudson River Trading, the joint backing by competing chip giants signals a profound structural shift. The industry is quietly conceding that the future of AI computing cannot rely on proprietary electrical fabrics alone.

Fierce Rivals Find Common Ground in Light

For decades, microprocessor manufacturers have competed fiercely on transistor density and clock speeds. But in the era of large language models, individual chip performance is only a fraction of the equation. Modern AI infrastructure is designed at rack scale, where thousands of graphics processing units (GPUs) must act as a single, unified supercomputer.

“AI infrastructure is increasingly being designed at rack scale, where networking has a direct impact on system performance, efficiency and economics,” said Patrick Rundell, Director at AMD Ventures. “Salience Labs is developing a differentiated, integrated optical switching architecture aimed at addressing these challenges, and AMD Ventures is pleased to support the team as it scales its technology and manufacturing capabilities.”

When data travels between these processors using traditional electrical switches, it must constantly be converted from optical signals—which travel through fiber cables between racks—back into electrical signals for switching, and then back to optical. This optical-electrical-optical (O/E/O) conversion introduces latency, drains massive amounts of power, and generates crippling heat.

To eliminate this, Salience Labs has engineered a fully integrated silicon photonics architecture that keeps the data entirely in the optical domain. Light is switched and amplified directly on the chip.

“Photonics is quickly becoming essential to scaling AI infrastructure, and optical circuit switching can unlock significant gains in performance and efficiency,” noted Srini Ananth, Managing Director at Intel Capital. He added that the startup "stands out through its integrated silicon photonics architecture and proprietary optical amplification, addressing key challenges to deploying optical switching at scale.”

The dual investment aligns seamlessly with broader industry movements, such as the emerging Ultra Accelerator Link (UALink) Consortium. While hardware manufacturers battle for dominance in AI accelerators, there is a shared, urgent need for an open, high-speed physical layer to connect them. Optical circuit switching provides the fundamental plumbing required to make those emerging communication standards viable.

The Physics Forcing the Optical Shift

The sheer scale of the problem is difficult to overstate. Industry forecasts, including projections from the Dell'Oro Group, estimate that cumulative spending on datacenter switches for AI back-end networks will approach $1 trillion between 2026 and 2030. More than half of this market will be driven by "scale-up" networks—the tightly coupled clusters of GPUs within adjacent racks that require ultra-low latency to train massive AI models.

Traditional electrical switches, dominated by incumbents providing Ethernet and InfiniBand solutions, are mature and widely deployed. But as bandwidth demands double with each new generation of AI hardware, pushing electrical signals over copper traces becomes exponentially more difficult. The required re-timers and complex signaling schemes consume power that would be better spent on actual computation.

Optical circuit switching (OCS) circumvents this entirely. By establishing dedicated, high-bandwidth optical paths between computing nodes, OCS acts like a digital rail yard, seamlessly routing data without the overhead of inspecting and converting every packet.

“OCS will play a critical role in AI datacenter networking,” said Vaysh Kewada, CEO and co-founder of Salience Labs. “As customers roll out optical links in their datacenters, high-performance optical switches are needed to deliver the token economics that workloads demand.”

The company recently proved the viability of its approach, executing a live demonstration of an industry-first 64-port all-optical silicon photonics switch at the ECOC 2026 summit in Spain. With the new capital, the engineering team is accelerating the development of 128- and 256-port variants, expanding the technology's capacity to handle both dense scale-up clusters and broader scale-out architectures that span entire datacenter floors.

Scaling Deeptech: From Oxford to Volume Production

Transitioning a breakthrough from a university laboratory to the rigid, high-volume manufacturing lines of the global semiconductor supply chain is notoriously difficult. Founded in 2021 as a spin-out leveraging over a decade of research from the University of Oxford and the University of Münster, Salience Labs is navigating this transition by forging aggressive manufacturing partnerships.

The startup is collaborating closely with Tower Semiconductor to ensure its designs can be produced at scale. This partnership is highly strategic; Tower Semiconductor recently initiated a $4 billion expansion of its 300mm silicon photonics and advanced packaging capabilities in Japan. By aligning with a foundry aggressively scaling its optical output, the Oxford-born enterprise is positioning itself to meet the anticipated surge in hyperscale demand.

Simultaneously, the firm is working with Keysight Technologies to build the industry's first dedicated testing environment for optical circuit switches—a necessary step for validating components before they are deployed into multi-billion-dollar datacenter builds.

To bridge the gap between European deeptech research and American hyperscale deployment, the company is also using its Series B funding to expand its footprint in the United States. A second West Coast office is being established in Silicon Valley, placing the company's engineers directly in the backyard of the world's largest AI infrastructure architects and its key investors.

The Trillion-Dollar Infrastructure Race

The race to commercialize optical networking is highly competitive. Hyperscalers like Google have already deployed proprietary free-space optical switches, such as their Apollo system, utilizing microscopic mirrors to bounce light between fibers. Meanwhile, other startups are focusing on in-package optical I/O, attempting to move data optically directly from the processor package itself.

Salience Labs occupies a distinct and critical middle ground. By integrating switching and amplification onto a single, manufacturable silicon photonics chip, the company offers a more compact, faster-reconfiguring alternative to bulky free-space mirrors, while providing the essential network routing needed to connect optically enabled processors across a facility.

As artificial intelligence transitions from a software novelty to the foundational layer of the global digital economy, the physical infrastructure that supports it must undergo a radical transformation. The $125 million bet placed by Bullhound, AMD, and Intel is more than a venture capital transaction. It is a clear signal that the future of global computing power will no longer be constrained by copper wire, but will instead be dictated by the speed of light.