CScale Raises $145 Million to Build Fail-Safe Optical Interconnects for AI

CScale, a startup focused on optical interconnects for AI infrastructure, has emerged from stealth with a $145 million Series C funding round, bringing its total funding to $188 million. Strategic investors include Nvidia and Intel Capital, alongside other venture capital firms. The company is tackling a critical issue in AI data centers: the inevitability of laser failures at gigawatt scale and the need for systems that can contain these failures without disrupting compute operations.

CEO Martin Lund, a veteran from Cisco, Metaswitch, and Broadcom, emphasizes that CScale's approach is uniquely integrated. Unlike other companies in the space, CScale designs its optical interconnects with the expectation that failures will occur. The company's architecture aims to isolate and manage these failures, ensuring that compute operations continue uninterrupted. Lund noted that the company has proof-of-concept systems with customers under NDA but did not disclose further details.

CScale is targeting the tightly coupled domain where thousands of accelerators operate as one machine across racks. The goal is not only higher bandwidth and low latency but also reduced human intervention inside data centers. Lund stated, "You don’t really want people wandering around the data center." The company is part of a growing field of startups addressing the limitations of copper in AI infrastructure, with competitors like Ayar Labs and Lightmatter pursuing similar optical interconnect solutions.

Lund's background in driving silicon and hardware strategies at Cisco and other major tech companies positions him uniquely to lead CScale. The company was founded in 2023 by CTO Sanjai Kohli, co-founder of GPS-chip pioneer SiRF. With teams in Palo Alto, Athens, Bangalore, Taipei, and Shanghai, CScale is poised to challenge the status quo in AI networking, betting that "lasers will fail. Compute shouldn’t."