CScale Exits Stealth With $145M From Nvidia and Intel to Build Failure-Proof Optical Links for AI Data Centers

A little-known Palo Alto startup has stepped out of the shadows with $145 million in fresh funding — and backing from two of the biggest names in chips — to solve one of the least visible but most expensive problems in artificial intelligence.

CScale exited stealth this week, announcing a $145 million Series C round that brings its total funding to $188 million. The round was co-led by Atreides Management, Valor Equity Partners and Premji Invest, with Sutter Hill Ventures and Maverick Silicon continuing their support. Nvidia and Intel Capital joined as the company's first strategic investors.

Founded in 2023 by chief technology officer Sanjai Kohli, who previously co-founded SiRF and later Inovi — acquired by Facebook in 2014 — the company is led by chief executive Martin Lund, who built Broadcom's switching business into a billion-dollar franchise and later led Cisco's Common Hardware Group. The startup employs roughly 85 people.

CScale is building optical interconnect for AI "scale-up" — the high-bandwidth, ultra-low-latency connections that let thousands of tightly coupled accelerators across dozens of racks behave like a single, much larger computer.

In the gigawatt-class AI data centers expected by the end of the decade, a single training cluster will demand continuous, perfectly synchronized communication between its chips.

The problem, CScale argues, is not bandwidth but reliability. A laser failure that is rare on any individual link becomes a continuous, fleet-level condition when a fabric spans hundreds of thousands of accelerators — degrading portions of the AI factory unless it can be contained.

The startup says its architecture is designed to isolate optical failures so they never interrupt ongoing compute.

"As AI scale-up domains extend across dozens of racks, optical interconnect becomes essential," Lund said in the announcement. "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."

Optical networking has become one of the hottest corners of AI infrastructure. CScale's most-cited rival, Lumilens, exited stealth in August with more than $700 million in funding at a $5.51 billion valuation — a round Atreides co-led as well.

Ayar Labs, an early leader in optical I/O chiplets, has raised $650 million this year, while Lightmatter was last valued at $4.4 billion and Celestial AI was acquired by Marvell for $3.25 billion.

The chip giants are also in the game directly. Broadcom and Nvidia are both shipping co-packaged optics switches, and Nvidia alone has poured billions into photonics companies in recent months, spreading its bets across multiple startups in the space.

CScale has disclosed few technical details beyond its reliability-first pitch, but its investor lineup — and Nvidia and Intel's decision to co-invest — signals that the market sees failure containment, not just raw speed, as the next frontier in connecting AI's ever-larger clusters.

The company says the new funding will accelerate development and commercialization of its interconnect offering. If its bet proves right, the defining technology of the AI factory era will not be the fastest chip or the brightest laser, but the one that keeps working when a laser inevitably fails.