- CScale emerged from stealth with a $145 million Series C, taking total funding to $188 million; Nvidia and Intel Capital joined as strategic investors
- CEO Martin Lund says the real scale-up problem is no longer just bandwidth: at gigawatt scale, optical failures become inevitable and the interconnect has to contain them
- Lund brings a deep networking pedigree from Cisco, Metaswitch and Broadcom
The interesting thing about CScale isn’t simply that another well-funded startup wants to replace copper with optics inside AI infrastructure. It is the problem the company has chosen to design around: lasers will fail, and at gigawatt scale the compute cannot be allowed to fail with them.
The Palo Alto startup emerged from stealth last week with a $145 million Series C, bringing its total funding to $188 million. Nvidia and Intel Capital joined as CScale’s first strategic investors, alongside Atreides Management, Valor Equity Partners, Premji Invest, Sutter Hill Ventures and Maverick Silicon.
CEO Martin Lund told Fierce Network that CScale is taking a deliberately integrated approach to optical interconnect. “The fundamental difference is that we’re fully integrated,” Lund said.
That matters because CScale is designing for failure rather than pretending failure can be eliminated. Even when laser interconnects fail — “which they will,” Lund said — the company’s architecture is intended to contain the failure and keep the compute running. Lund would not disclose much about how CScale achieves that yet, saying the company has proof-of-concept systems with customers under NDA.
CScale describes itself as a scale-up company: it is targeting the tightly coupled domain in which thousands of accelerators increasingly have to operate as one machine across racks. Lund said the goal is not only higher bandwidth and very low latency, but also less human intervention inside the data center. “You don’t really want people wandering around the data center,” he said.
The copper wall
CScale is one of a growing group of companies attacking a basic physical constraint in AI infrastructure: copper stops being practical as bandwidth rises and the distance between tightly coupled compute elements grows.
At today’s AI data rates, electrical links have sharply limited reach. That is pushing the industry toward optical interconnects that can move far more data over longer distances while consuming less power per bit. The shift is increasingly important as AI systems expand from individual servers and racks into enormous clusters.
The competitive field is already crowded. Ayar Labs, Lightmatter, Celestial AI, Lumilens and others are pursuing different versions of the optical-interconnect opportunity. Lund said Ayar Labs and Lightmatter are among the companies approaching the problem in ways most comparable with CScale.
What distinguishes CScale’s pitch is reliability at system scale. Its argument is that an optical link that looks highly reliable in isolation can become an operational problem when an AI factory contains enormous numbers of links. A rare component failure stops being rare at fleet scale.
Lund has been here before
CScale’s CEO has spent much of his career inside the companies now shaping AI networking. Before joining CScale, Lund led Cisco’s Common Hardware Group, overseeing silicon, hardware systems and optics and helping drive the company’s Silicon One strategy. Earlier, he was CEO of Metaswitch, which was acquired by Microsoft, and spent roughly 12 years at Broadcom, where CScale says he built its switching business to $1 billion.
That background makes his move noteworthy. In a LinkedIn post announcing his decision to join the startup, Lund said he had spent his career pursuing bottlenecks that were “real but underrecognized.” His latest target is the interconnect itself: keeping networking from stalling compute as AI infrastructure moves toward gigawatt scale.
CScale was founded in 2023 by CTO Sanjai Kohli, the co-founder of GPS-chip pioneer SiRF. The company is headquartered in Palo Alto and also has teams in Athens, Bangalore, Taipei and Shanghai.
The company is not alone in attracting large amounts of capital into optics, and the market is moving fast enough that consolidation looks plausible. The more immediate question, however, is whether CScale can turn its reliability-first architecture into a manufacturable product that AI infrastructure builders will deploy at scale.
For now, Lund’s formulation is the simplest description of the bet: “Lasers will fail. Compute shouldn’t.”
Learn more about AI infrastructure
Corning to build 3 new optical plants in the U.S. with a little ($500M) help from Nvidia
Verizon signs $2 billion-plus deal with Corning to supply fiber for FTTP and AI
Nvidia GTC: Dell upgrades AI Factory to enable ROI
AI Security, Infrastructure Threats, and the Battle to Protect Carrier Networks