Analysis
CScale emerged from stealth with $145 million in Series C funding to build optical interconnects for gigawatt-scale AI data centers, bringing its total funding to $188 million, SiliconANGLE reported. The Series C was co-led by Atreides Management, Valor Equity Partners and Premji Invest, with Nvidia and Intel Capital joining as strategic investors, Tech Startups confirmed.
Solving Copper's Bandwidth Ceiling
CScale was founded in 2023 by electrical engineer Sanjai Kohli, with long-time data center executive Martin Lund as CEO. Its interconnect links the accelerators that make up an AI cluster -- potentially thousands of chips spread across dozens of racks -- by moving data between them optically instead of over copper, which runs out of bandwidth and consistent low latency at the scale next-generation AI data centers are targeting. The company says it's designing the system so optical failures stay contained to a single link rather than taking down a cluster's training run.
“## Solving Copper's Bandwidth Ceiling CScale was founded in 2023 by electrical engineer Sanjai Kohli, with long-time data center executive Martin Lund as CEO.”
A Crowded, Well-Funded Field
CScale is a late entrant into a category with real money and real exits already in motion:
- Ayar Labs -- raised $500 million in a Series E backed by Nvidia and AMD in March 2026, then added $150 million more, bringing 2026 funding to roughly $650 million at a valuation above $5 billion. Widely seen as the category leader.
- Lightmatter -- raised a $400 million Series D in 2024 at a $4.4 billion valuation; competes directly on co-packaged optics for AI chips.
- Celestial AI -- pursued a similar photonic-fabric approach but agreed to be acquired by chipmaker Marvell for $3.25 billion rather than scale independently, a sign of where consolidation in this category may be heading.
CScale's $188 million is a fraction of what Ayar Labs alone has raised this year, meaning it's entering as a smaller, newer player against incumbents with multi-year head starts and hyperscaler design wins already in motion.
What the Headline Misses
Neither SiliconANGLE nor Tech Startups' reporting names a signed customer or deployment for CScale's interconnect -- the company is launching with funding and a thesis, not disclosed revenue or a hyperscaler commitment. Nvidia and Intel Capital's participation reads as a hedge as much as a conviction bet: strategic investors in fast-moving infrastructure categories routinely back more than one horse, and their money here doesn't preclude them from also deepening ties with Ayar Labs or Lightmatter. The AI data center debt financing Pulse has tracked elsewhere this week, including Lambda's $1 billion loan, shows how much capital is chasing the same underlying bottleneck -- compute interconnect and power -- from different angles.
Whether CScale differentiates on reliability, cost, or simply gets acquired the way Celestial AI did will likely be decided well before its optical links ship at the gigawatt scale its pitch describes.
Why Gigawatt-Scale Changes the Math
Today's largest AI training clusters run in the hundreds of megawatts; the generation CScale is targeting -- gigawatt-class campuses -- implies single sites consuming as much power as a small city and linking accelerator counts an order of magnitude beyond current deployments. At that scale, even small latency or bandwidth losses in copper interconnects compound across thousands of simultaneous links, which is the specific failure mode optical interconnects are pitched to fix. None of the four optical-interconnect players -- CScale, Ayar Labs, Lightmatter or the former Celestial AI -- has publicly disclosed a live gigawatt-scale deployment yet, meaning the entire category is still racing to prove the technology at a scale that barely exists outside of blueprints from Microsoft, Meta and OpenAI's own data center plans.