CScale Company Spotlight: Reliable Optics for AI Scale-Up

CScale is building optical interconnect for AI scale-up, the tightly coupled communication layer that lets accelerators across multiple racks operate like one large computer. Led by CEO Martin Lund and founder and CTO Sanjai Kohli, the Palo Alto company is designing an integrated light engine around a simple systems reality: in a sufficiently large AI factory, optical failures are inevitable.

The company's answer is not a promise that lasers never fail. CScale says its architecture is designed to contain a failure so the surrounding compute keeps working. That distinction matters because the cost of an interconnect problem is measured in idle accelerators, interrupted workloads, and expensive capacity that cannot do the job it was purchased to do.

CScale exited stealth in September 2026 with $188M in reported funding and approximately 85 employees globally. Its current hiring spans chip design, lasers, packaging, systems software, and operations, showing how many engineering disciplines have to meet before an optical architecture becomes dependable infrastructure.

About CScale

CScale was founded in 2023 by Sanjai Kohli and is headquartered in Palo Alto, California. The company previously operated under the CSpeed and CSpeedIO names. CScale now lists a presence in Palo Alto, Athens, Bangalore, Taipei, and Shanghai, giving the young company an international footprint across several semiconductor and engineering hubs.

The business is focused on AI scale-up connectivity. Scale-up systems connect processors closely enough to behave as a larger computing domain, rather than sending jobs between loosely connected machines. As those domains spread across dozens of racks and thousands of accelerators, the interconnect becomes part of the machine's economic core.

CScale describes its product as an integrated light engine for reliable optical connectivity at gigawatt scale. The company has not disclosed the bandwidth, latency, power, packaging, manufacturing, or compatibility details needed for an independent product comparison. Its public thesis is narrower and more useful: optical links must preserve continuity when failures become ordinary fleet behavior.

Why CScale Matters Right Now

AI infrastructure is moving from individual servers toward factories built from processors, memory, networking, power, cooling, and software that must cooperate under sustained load. More accelerators do not automatically create more useful compute. The system earns its keep only when those accelerators remain connected and productive.

That is why reliability changes the optical conversation. Bandwidth and reach remain essential, but they are not the whole commercial equation. If one failed component interrupts a wider scale-up domain, the damage is larger than the price of a replacement part. It reaches the utilization of the surrounding hardware and the schedule of the workloads running across it.

CScale is trying to turn a laser failure into a contained event instead of a compute event. CEO Martin Lund told Reuters that the company intends to integrate lasers onto chips while maintaining data flow when a laser fails. Reuters also reported that CScale plans to ship chips in 2028, making the next phase a test of whether the architecture can survive qualification, manufacturing, and field conditions.

Leadership Built Around Systems and Semiconductors

Martin Lund leads CScale after running Cisco's Common Hardware Group across silicon, hardware systems, and optics, including Silicon One. CScale also credits Lund with leading Metaswitch before its acquisition by Microsoft and with building Broadcom's switching business into a billion-dollar franchise. That background is relevant because CScale must translate technical architecture into a product that large infrastructure buyers can evaluate and trust.

Sanjai Kohli, CScale's founder and CTO, previously co-founded SiRF, which helped bring GPS into mass-market devices. He later founded Inovi, acquired by Facebook in 2014, and received the 2010 European Inventor Award. Kohli's track record sits directly in the hard middle between communications engineering and commercial systems.

The current CScale team page also shows operating depth across products, packaging, optics, mixed signal, system-on-chip design, systems software, and operations. That mix is less glamorous than a single product demo and more revealing. Optical-semiconductor businesses are coordination machines. Product design, packaging, reliability, software, supply chain, and customer qualification all have to arrive at the same time.

Funding the Move From Architecture to Product

CScale announced a $145M Series C on September 30, 2026, bringing its reported total funding to $188M. Atreides Management, Valor Equity Partners, and Premji Invest co-led the round. Sutter Hill Ventures and Maverick Silicon continued their support, while NVIDIA and Intel Capital joined as CScale's first strategic investors.

The investor group gives CScale financial runway and strategic proximity to the AI-compute ecosystem, but it does not prove customer adoption. The company has disclosed no named customer, production deployment, benchmark, or commercial agreement tied to the strategic investors. The Series C funds the work required to produce that evidence.

Semiconductor commercialization moves through design validation, packaging, manufacturing readiness, interoperability, customer qualification, and field reliability. A strong architecture can still fail if any one of those stages arrives late or cannot hold up under volume. CScale's financing is large because the distance between a credible idea and trusted infrastructure is long.

Hiring Shows Where the Work Is

CScale's careers page listed 20 open roles on September 30, 2026. The openings included 4 CMOS and electrical-chip positions, 3 laser roles, 8 packaging and operations roles, 4 systems and software roles, and 1 supply-chain position. Jobs were spread across Palo Alto, Bangalore, Athens, Shanghai, Taiwan, and Shijiazhuang.

The distribution matters more than the count. CScale is not hiring only for one photonics breakthrough. It is staffing verification, package design, fiber-array integration, thermal simulation, CAD and IT, software, printed-circuit-board validation, and supply-chain execution. That is the operating map of a company trying to make a complex hardware system manufacturable and supportable.

Hiring does not establish customer demand or production readiness, and CScale has not claimed either publicly. It does show where management believes the bottlenecks sit. The company is adding people across the interfaces where a lab design has to become repeatable hardware.

What CScale Still Has to Prove

CScale's public case begins with a sound systems observation: fleet-scale reliability cannot be reduced to the reliability of one component. The harder proof will be quantitative. Infrastructure buyers will need evidence that the architecture contains failures under realistic conditions, meets performance and power requirements, integrates into target systems, and can be produced consistently.

The company has not disclosed named customers, design wins, commercial deployments, revenue, valuation, performance benchmarks, or manufacturing partners. Those omissions do not defeat the thesis, but they define the diligence still ahead. CScale's planned 2028 shipment target leaves a meaningful execution window between today's financing and tomorrow's field data.

The wider signal is that AI infrastructure is becoming a continuity business. Compute density attracts the headlines, while uptime, serviceability, power, and network behavior determine whether the capital earns a return. CScale is betting that optical interconnect will be judged not only by how fast it moves data, but by how quietly it contains the failures that operators know will come.

That is a demanding place to build a company. It is also where infrastructure value tends to hide: inside the component nobody notices because the system kept working.