CScale Raises $145M Series C to Build Resilient Optical Interconnect for Gigawatt-Scale AI

CScale has emerged from stealth with $145 million in Series C funding and a sharply defined target: the optical interconnect layer that will determine whether tomorrow’s enormous AI clusters can function as coherent systems rather than expensive islands of compute.

The financing brings the Palo Alto company’s total funding to $188 million. Atreides Management led the round alongside Valor Equity Partners, with Premji Invest joining as co-lead. Sutter Hill Ventures continued its backing alongside existing investor Maverick Silicon, while NVIDIA and Intel Capital joined as CScale’s first strategic investors.

The unusually large round reflects a problem moving rapidly toward the center of AI infrastructure design. As training and inference systems grow, raw accelerator performance is no longer enough. Thousands of GPUs and other processors must exchange data continuously, at extremely high bandwidth and with predictable latency. When those links slow down or fail, the entire cluster loses useful compute.

The Network Is Becoming Part of the Computer

CScale is building an integrated light engine for scale-up networks—the tightly coupled links that allow accelerators to work together as one much larger computing system. The company is designing for future gigawatt-class AI data centers, where a single scale-up domain could span thousands of accelerators across dozens of racks.

At that scale, reliability becomes as important as peak bandwidth. A fault that appears rare on one optical link becomes an everyday fleet-level event when a system contains hundreds of thousands of accelerators and an enormous number of connections. CScale says its architecture is intended to contain optical failures without interrupting compute, shifting the design goal from easy component replacement to continuous operation.

“Lasers will fail. Compute shouldn’t,” CScale CEO Martin Lund said in the announcement. The line captures the economic argument behind the company’s technology: an AI factory cannot produce a return on its accelerators when the network repeatedly leaves them waiting or takes portions of the cluster offline.

That framing also places CScale within a broader industry shift. As optical engines move deeper into AI infrastructure, networking is becoming part of the computing architecture itself. Bandwidth, latency, power consumption, serviceability and failure isolation increasingly determine how much useful performance a data center can extract from its silicon.

Experienced Operators Behind a New Optical Architecture

Lund brings three decades of experience across networking, silicon and hardware systems. He previously built Broadcom’s switching business into a billion-dollar operation, held senior roles at Microsoft and Cadence, and most recently led Cisco’s Common Hardware Group, which included the Silicon One portfolio as well as the company’s silicon, hardware systems and optics work.

CScale was founded in 2023 by CTO Sanjai Kohli. Kohli previously co-founded SiRF, whose technology helped bring GPS into mass-market devices, and later founded Inovi, which Facebook acquired in 2014. CScale now employs approximately 85 people globally, according to the company.

The founding team’s background matters because optical interconnect is not a single-component problem. A commercially viable platform must coordinate photonics, electronics and software while fitting the operational realities of large data centers. CScale’s pitch is therefore architectural: build the light engine and its surrounding system to behave predictably at a scale where isolated component specifications tell only part of the story.

Why Investors Are Betting on Reliability

The investor roster combines financial firms with two of the semiconductor industry’s most consequential strategic players. That mix suggests CScale is moving beyond early technical validation toward commercialization in an ecosystem where accelerator vendors, networking suppliers and data center operators must align closely.

Atreides Managing Partner and Chief Investment Officer Gavin Baker described AI infrastructure as a systems problem rather than simply a compute problem. Premji Invest Managing Partner Sandesh Patnam similarly emphasized CScale’s ability to connect technical choices with value across the broader system.

The opportunity is significant, but so is the execution challenge. Optical technology must deliver more bandwidth while controlling power use, heat, manufacturing complexity and failure rates. It must also integrate with fast-changing accelerator and rack architectures. Large-scale customers will judge not only laboratory performance but qualification timelines, supply-chain readiness and behavior under sustained production workloads.

CScale’s $145 million Series C gives the company substantial resources to address those hurdles. It also signals that the race to scale AI is expanding beyond processors. In the next generation of AI factories, the interconnect may be what turns thousands of powerful chips into one dependable machine.