CScale, a Silicon Valley startup backed by Nvidia and Intel, has raised $145 million to move optical links closer to AI processors — and to keep data flowing even when the lasers carrying it die.

The round was led by Atreides Management, Valor Equity Partners and Premji Invest, and brings CScale's total funding to $188 million. CEO Martin Lund, a former hardware executive at Cisco Systems, says the company aims to ship chips by 2028.

The commercial problem CScale is attacking sits inside the AI server rack. Today's AI servers mostly connect chips within a box using copper cables. Copper is cheap and dependable, but it cannot carry ultra-fast signals far without degradation. That physical limit pushes designers like Nvidia and AMD to pack processors tightly together and lean harder on complex cooling — a layout constraint that shapes the entire bill of materials for AI hardware.

Optical links, which transmit data as pulses of light through fiber, move far more data over longer distances. The industry's direction of travel is clear: bring optics inside the server and, eventually, right up to the chip package itself.

The snag is reliability. Lasers are components that wear out. If a laser failure forces a full swap of a tightly integrated part, downtime and maintenance costs spike. That risk is why many current designs keep lasers in replaceable modules rather than integrating them onto the package.

CScale claims it can put lasers directly onto chips while maintaining traffic when a single laser fails — converting what would be an outage into a manageable performance hit. That claim, unproven until silicon ships, is what the $145 million is buying time to demonstrate.

Why the 2028 timeline matters

CScale's schedule tests a specific question for data center operators: is "lasers fail" a downtime problem, or a design constraint that engineering can absorb?

If the approach works, near-chip optics becomes easier to live with. Fewer truck rolls, less forced replacement of integrated parts, and higher real-world uptime would remove one of the biggest practical barriers to swapping short-reach copper links inside racks for optical ones sitting closer to Nvidia and AMD processors.

For chipmakers, server builders and suppliers of cables, transceivers and cooling gear, such a shift would redraw where money gets spent across the AI hardware stack. Operators could stop treating interconnects as a hard physical limit dictating dense layouts and heavy cooling, and start weighing optics as an architectural choice judged on reliability, serviceability and total system cost.

Whether CScale meets its 2028 shipping target will determine if laser-on-package moves from a reliability gamble to a standard option in AI server design.