CScale, a Silicon Valley startup working on fiber-optic interconnects for AI chips, has raised $145 million in venture capital with backing from Nvidia and Intel. The company, formerly known as CSpeed, disclosed the round on Wednesday and said it plans to ship its chips by 2028.

The round brings CScale's total funding to $188 million. Atreides Management, Valor Equity Partners and Premji Invest led the financing. Existing investors Sutter Hill Ventures and Maverick Silicon also participated.

The company's CEO is Martin Lund, who left Cisco Systems earlier this year, where he oversaw chip and hardware systems for the networking giant. CScale has released few technical details about its approach, but Lund outlined the company's core proposition in an interview: integrating lasers directly onto chips in a way that does not interrupt data flow when the lasers fail.

The problem CScale is attacking sits inside AI data center servers. Copper cables currently wire together the chips in those servers. Copper is fast and reliable, but it has limited reach. Those reach limits have forced chip designers such as Nvidia and Advanced Micro Devices to pack their processors tightly together inside servers. Tight packing, in turn, requires elaborate liquid cooling systems to handle the heat the chips generate.

Public companies and startups alike are pouring billions of dollars into replacing copper with optical connections that carry data as flashes of laser light rather than electrons. The obstacle has been reliability. Lasers fail frequently, so many firms in the field keep the laser separate from the chip, which allows the laser to be replaced when it breaks.

CScale's bet is that it can put the lasers on the chips themselves and engineer around the failure problem. Lund declined to give specifics of how the company achieves this. He did emphasize that the approach relies on established components rather than new physics.

"We're not using any exotic technologies," Lund said. "We're using proven technologies, integrating them together in a novel way, but preserving the same trusted fiber technology that has been out there and shipping in the industry for many, many years now."

The strategic logic for Nvidia's and Intel's participation is straightforward. Both companies sell the processors that need to be linked inside AI servers. Any interconnect technology that extends reach and reduces the thermal constraints of dense copper-wired clusters would change how their customers can design data centers. Longer optical links would loosen the physical packing constraints that currently dictate server architecture and drive cooling costs.

CScale enters a crowded field. The scramble to replace copper inside AI servers has drawn billions in investment across both public companies and startups, and CScale has so far distinguished itself mainly through the caliber of its investors and its leadership rather than through disclosed technical results. The company's 2028 shipping target places it several years out from commercial deployment, a timeframe that requires sustained capital — one reason the $145 million raise matters despite the company's limited public technical record.

Lund's move from Cisco to CScale also signals how the interconnect problem has shifted from a networking concern to a central bottleneck in AI compute. At Cisco, Lund ran the chip and hardware systems operation for one of the world's largest networking equipment makers. That background maps directly onto the engineering challenge CScale faces: moving data reliably between processors at scale.

For now, the key facts are these: $145 million new capital, $188 million total raised, Nvidia and Intel on the cap table, proven fiber technology recombined in what Lund calls a novel integration, and chips scheduled to ship by 2028. Whether the on-chip laser architecture survives contact with data center operating conditions will determine if CScale converts its funding and backing into a position in the AI interconnect market.