CScale left stealth on 30 September with a $145m Series C, bringing its total raised to $188m. The round was co-led by Atreides Management, Valor Equity Partners and Premji Invest, with Sutter Hill Ventures and Maverick Silicon participating, and with investment from both Nvidia and Intel Capital.
That investor list is the first thing worth noticing. Nvidia and Intel are competitors in almost every market they share, and both are on the cap table of a company whose product sits between their accelerators.
What the company is building
CScale is developing an optical interconnect for AI scale-up networking — the links that let thousands of accelerators behave like one much larger computer. The company is building an integrated light engine for that job, and says it is designing the interconnect to contain optical failures without interrupting compute.
That second claim is the commercially interesting one. In a cluster spanning dozens of racks, a single failed optical link can stall a training run across the whole system. Containing a failure locally rather than letting it propagate is worth more, at gigawatt scale, than a marginal improvement in bandwidth.
Why now
The scaling of AI systems has moved the bottleneck. When a model fits in one rack, the interconnect is an implementation detail. When it spans dozens, the distance between chips, the power the links consume and the failure rate across thousands of them become the design constraint. Copper runs out of reach and budget before the rack count does, which is why optics has become an investable category rather than a component choice.
CScale was founded in 2023 and is headquartered in Palo Alto. Sanjai Kohli, its chief technology officer, is a founder. The company has published no performance figures and named no customers.
What to watch
Whether a named hyperscaler or neocloud says it is deploying the interconnect, and whether Nvidia’s own scale-up roadmap ends up treating CScale as a supplier or as something it would rather own.
