As AI data centers grow, connecting their chips without letting the network become a bottleneck is an increasingly important infrastructure challenge. Palo Alto startup CScale has spent three years developing optical-interconnect technology for AI systems. On Wednesday, it exited stealth mode and announced $145 million in Series C funding, bringing its total raised to $188 million, with Nvidia and Intel joining as strategic backers.
The funding round, first detailed by Yahoo Finance, was co-led by Atreides Management, Valor Equity Partners and Premji Invest. Nvidia and Intel Capital marked their entry as CScale's first strategic investors, a notable vote of confidence from two of the chipmakers whose hardware CScale's technology is ultimately designed to connect. The company, headquartered in Palo Alto, now employs roughly 85 people globally, according to the same report. As Electronics Weekly noted, Sutter Hill Ventures has backed CScale since its inception and joined this latest round alongside Maverick Silicon.
Betting On Light Over Copper
CScale is developing optical interconnects for AI scale-up systems, where fast communication among accelerators is needed to coordinate large workloads. The company’s goal is to support high-bandwidth, low-latency connections so thousands of chips can operate as a larger system, as GamesBeat described in its coverage of the stealth exit. The available reporting describes an optical-interconnect product, not a plan to replace every copper connection in data centers. Reuters reports CScale plans to ship its chips by 2028 and formerly operated under the name CSpeed.
GamesBeat reported that CScale is designing its interconnect to contain optical failures without stopping compute. CEO Martin Lund has presented continuity as a central design concern, noting that “lasers will fail, but compute should not.” That is a design objective, not evidence that the technology has already demonstrated this performance.
Why Scale-Up Networking Matters Now
Lund has argued that the need for this kind of interconnect will grow as AI systems expand. According to DataCenterDynamics, he said optical interconnect becomes essential as AI scale-up domains extend across dozens of racks. Atreides Management managing partner and chief investment officer Gavin Baker made a related point: bandwidth alone is not enough if a system cannot maintain reliability as deployments grow.
Baker has framed AI infrastructure as a systems challenge as well as a compute challenge, arguing that optical failures can disrupt compute if they are not contained. He said CScale is designing scale-up optics for next-generation AI factories. Nvidia has also argued that silicon-photonics switches could improve power efficiency and resiliency compared with traditional optics and help enable gigawatt-scale AI factories, according to a post on Nvidia's blog.
Data-Center Power and Optical Design
The scale-up debate also has an electricity dimension. According to the U.S. Department of Energy, data centers consumed about 4.4% of total U.S. electricity in 2023, and the department projected they would consume approximately 6.7 to 12% by 2028.
The Founders Behind CScale
CScale was founded in 2023 by CTO Sanjai Kohli, who previously co-founded SIRF, with Martin Lund serving as CEO. The company says its new funding will go toward accelerating development and commercialization of its optical interconnect technology for AI scale-up systems, per the Yahoo Finance report.
A Crowded, Fast-Growing Field
CScale's debut lands amid a surge of capital pouring into optical and photonic technology for data centers. Reuters notes that public companies and startups alike are pouring billions of dollars into replacing copper connections with optical ones, a shift driven by the physical limits of copper at the bandwidth and distances modern AI clusters require. Tom's Hardware reports the global data-center optical-interconnect market is projected to grow from $13.7 billion in 2024 to $144.4 billion by 2030, a 48.1% compound annual growth rate, with silicon photonics expected to capture nearly two-thirds of that revenue.
The same outlet reports the AI industry has invested more than $15 billion over the past year in co-packaged optics, photonic chips, transceiver modules and fiber. Rival startup Ayar Labs closed a $500 million Series E round in March, led by Neuberger Berman, bringing its total funding to $870 million and its valuation to $3.75 billion, according to the company's own announcement. Ayar Labs has said co-packaged optics can replace bandwidth-limited copper interconnects and let thousands of GPUs operate as a unified system.
Consolidation is also reshaping the sector. DataCenterDynamics reported that Marvell confirmed plans to acquire optical-interconnect startup Celestial AI, which had closed a $250 million Series C1 round at a $2.5 billion valuation after raising more than $580 million in total. The proposed deal includes $1 billion in cash and Marvell shares worth $2.25 billion, with a total value that could reach $5.5 billion if milestones are met. Elsewhere in the space, The Next Web reported that Lumilens raised $700 million in August, Celero raised $275 million, and Eliyan closed a $145 million Series C in July — underscoring just how much money is chasing the same bet that optical links, not copper, will carry the next generation of AI factories.
