CScale, a Palo Alto startup developing optical connections for artificial intelligence data centres, emerged from stealth on 30 September 2026 with $145 million in Series C funding. The round brings the company's total funding to $188 million.
Atreides Management led the round, with Valor Equity Partners and Premji Invest, the investment arm of Indian technology entrepreneur Azim Premji, joining as co-leads. Sutter Hill Ventures, which has backed CScale since its founding, and existing investor Maverick Silicon also participated. Nvidia and Intel Capital joined as first-time strategic investors.
The participation of Nvidia, the dominant supplier of AI accelerators, and Intel's venture arm gives CScale two of the most important names in computing infrastructure on its cap table, and signals that both companies see optical interconnect as a critical piece of future AI systems.
The problem: moving data between thousands of chips
Modern AI models are trained and run on clusters of thousands of accelerators. For those chips to work together efficiently, they must exchange enormous amounts of data at very high speeds and with very low delay. The connections that let accelerators behave like a single, much larger computer are known as scale-up interconnect.
Today, many of those connections inside a rack use copper cables. But copper struggles to carry data over longer distances at the speeds AI requires, and as AI systems grow from single racks to dozens of racks, the industry is turning to optical links, which transmit data as light through fibre.
Optics bring their own problem: reliability. Optical components, especially lasers, fail more often than copper connections. In a tightly coupled AI system, a single failed link can interrupt the work of many chips.
"As AI scale-up domains extend across dozens of racks, optical interconnect becomes essential," said Martin Lund, CEO of CScale. "At that scale, reliable, predictable communication is fundamental to system economics. Easier part replacement improves serviceability, not continuity. Lasers will fail. Every optical failure is a compute failure."
CScale's approach: contain the failure
CScale is building what it calls an integrated light engine, an optical component designed to provide dependable connectivity for AI infrastructure at gigawatt scale. Its core idea is that optical failures should be contained so that computation keeps running, rather than bringing down part of a training job or inference workload.
The company says it combines photonics, electronics and software in its design. It has released few technical details, but its pitch is that reliability, rather than raw bandwidth alone, will determine the economics of large AI systems. Training runs on the largest models can last weeks and cost hundreds of millions of dollars in computing time, so even brief interruptions can be extremely expensive.
"CScale is designing reliable scale-up optics for next-generation AI factories, unlocking the full performance of optics without trade-offs," Lund said.
An experienced leadership team
CScale was founded in 2023 by Sanjai Kohli, who serves as chief technology officer. Kohli previously co-founded SiRF Technology, a pioneer in GPS chips, and Inovi. Lund previously led Broadcom's switching business and Cisco's Common Hardware Group, giving him deep experience in the networking silicon that underpins data centres.
