Optical interconnect startup CScale has exited stealth mode after closing a $145 million Series C round that included investment from Nvidia and Intel Capital. Atreides Management, Valor Equity Partners, and Premji Invest co-led the round. Sutter Hill Ventures and Maverick Silicon also participated.

The financing brings CScale's total raised capital to $188 million. The company plans to use the funds to accelerate development and commercialization of its optical interconnect offering.

Sanjai Kohli, who serves as the company's CTO, founded CScale in 2023. The startup is headquartered in Palo Alto and is developing an optical interconnect designed to support AI scale-up networking.

Failure containment at the core of the design

CScale has disclosed few technical details about its interconnect. In a statement, the company said it is architecting the product so that optical failures are contained when they occur and therefore do not interrupt compute.

CEO Martin Lund framed the problem in terms of system scale. "As AI scale-up domains extend across dozens of racks, optical interconnect becomes essential," Lund said. "At that scale, reliable, predictable communication is fundamental to system economics. Easier part replacement improves serviceability, not continuity. We're designing the interconnect for continuity. Lasers will fail. Compute shouldn't."

The distinction between serviceability and continuity appears central to CScale's engineering approach. Rather than optimizing for rapid replacement of failed optical components, the company says it is building the interconnect so that compute workloads survive component failures.

Investors cite reliability as the scaling bottleneck

Gavin Baker, managing partner and chief investment officer at Atreides Management, positioned CScale's value proposition within the broader trajectory of AI infrastructure buildouts.

"AI infrastructure is no longer just a compute problem. It is a systems problem," Baker said. "As scale-up systems move toward gigawatt-class deployments, interconnect bandwidth means nothing without system reliability. Every optical failure is a compute failure. At this scale, failures are not hypothetical—they are inevitable. CScale is designing reliable scale-up optics for next-generation AI factories, unlocking the full performance of optics without trade-offs."

Baker's argument ties directly to the economics of large AI training clusters. If a single optical link failure interrupts compute across an interconnected domain, the effective cost of that failure scales with the size of the affected workload. Containing failures at the optical layer would allow operators to keep GPUs productive even as individual components degrade.

The participation of Nvidia and Intel Capital signals strategic interest from both a leading AI accelerator vendor and a established silicon player. Nvidia's GPUs dominate the AI training market, and the company sells its own networking products for cluster interconnect. Intel, through its silicon photonics work, has a long-standing position in optical component manufacturing.

Competitive context

CScale enters a market where optical interconnects have moved from datacenter switching into direct GPU attach. As AI training clusters grow from single racks to multi-rack scale-up domains, copper links face reach and power limitations, and vendors across the industry are racing to qualify optical solutions for these shorter-reach, high-bandwidth links.

The company has not yet disclosed product specifications, timelines for availability, or named customers. With $188 million in total funding, CScale will need to move from architecture claims to qualified hardware as incumbents and well-funded startups target the same reliability problem in parallel.

What distinguishes the company's public positioning so far is its explicit framing: the design goal is continuity of compute in the presence of inevitable laser failures, not merely faster servicing of failed parts. Whether that architecture translates into deployable products at scale remains to be demonstrated.