CScale Secures $145M from Nvidia and Intel to Replace Copper Cables with Optical Interconnect

CScale Emerges from Stealth with $188 Million in Total Funding

CScale, formerly known as CSpeed, has emerged from stealth mode announcing $145 million in Series C funding, bringing its total venture capital funding to $188 million since its founding in 20231

1

3

3

. The Palo Alto-based scale-up interconnect startup secured strategic investments from industry giants Nvidia and Intel Capital as first-time investors, signaling strong confidence in its optical technology for AI data centers2

2

. Atreides Management led the round, with Valor Equity Partners and Premji Invest serving as co-leads, while existing investors Sutter Hill Ventures and Maverick Silicon also participated4

4

.

Tackling the Copper Cable Bottleneck in AI Infrastructure

CScale joins an intensifying race in Silicon Valley to replace copper cables with fiber-optic connections inside AI data center servers1

1

. Copper cables, while fast and reliable, have severe reach limitations that force chip designers such as Nvidia and Advanced Micro Devices to pack processors tightly together inside servers4

4

. This tight packing generates substantial heat, requiring elaborate liquid cooling systems that add complexity and cost to AI infrastructure. The company's optical interconnect technology is specifically designed to link thousands of AI accelerators across dozens of racks, enabling what the industry calls large scale-up cluster configurations2

2

3

3

.

Solving the Laser Failure Problem in AI Networking

Public companies and startups are pouring billions of dollars into optical connections that use flashes of laser light instead of electrons to carry data between AI chips1

1

. However, these lasers are prone to frequent failures, leading many firms to keep lasers separate from chips for easy replacement. CScale's approach differs fundamentally. CEO Martin Lund stated the company will integrate lasers onto chips in a way that does not interrupt data flow when lasers fail1

1

. "We're designing the interconnect for continuity. Lasers will fail. Compute shouldn't," Lund explained2

2

. The company describes its product as an "integrated light engine" capable of containing optical failures without disrupting compute operations3

3

.

Proven Leadership Driving AI Optical Interconnect Technology

Martin Lund, who left his position overseeing chip and hardware systems for networking giant Cisco Systems earlier this year, leads CScale as CEO1

1

. Lund previously built Broadcom's switching business and held senior roles at Microsoft and Cadence2

2

. Founder and chief technology officer Sanjai Kohli brings extensive experience, having co-founded SiRF Technologies, which brought GPS technology to the mass market, and later founded Inovi, which Facebook acquired in 20142

2

. The company currently employs approximately 85 people focused on developing this critical AI infrastructure technology2

2

.

Market Context and Timeline for Commercial Deployment

CScale enters a competitive landscape where optical interconnect startups have attracted substantial venture capital funding. Atreides Management, which led CScale's round, also backed Lumilens with $700 million in August, while Celero raised $275 million and Eliyan secured $145 million in Series C funding in July2

2

. "As scale-up systems move toward gigawatt-class deployments, interconnect bandwidth means nothing without system reliability," said Gavin Baker, managing partner and chief investment officer at Atreides Management. "Every optical failure is a compute failure. At this scale, failures are not hypothetical -- they are inevitable"3

3

. Lund emphasized the company uses proven technologies integrated in novel ways while preserving trusted fiber technology that has shipped in the industry for many years1

1

. CScale plans to ship its chips by 2028, positioning itself to address the growing demands of linking thousands of AI accelerators as data centers scale to gigawatt-class deployments1

1

3

3

.