Nvidia and Intel Capital Backed CScale
The investment aims to address optical reliability in the next generation of massive AI data centers.
Updated on Sept. 30, 2026 in Data Centers
Live Poll
Do you expect the cost of new AI-driven technology to make personal electronics more expensive?
Nvidia and Intel Capital have announced a strategic investment in CScale, a startup founded in 2023 to develop resilient interconnect architectures for AI infrastructure. The company focuses on mitigating optical failures within large-scale compute clusters.
Why it matters
As AI clusters scale to hundreds of thousands of accelerators, optical interconnect failures emerge as a continuous risk to system stability. Addressing these technical gaps is required to enable the gigawatt-class data centers projected to arrive by the end of the decade.
CScale architectures isolate optical failures at the interconnect layer to prevent them from cascading into full-scale compute interruptions. This approach integrates photonics—the use of light to transmit data—with specialized electronics and software management.
The players
Nvidia
A designer of graphics processing units and AI-specialized chips that maintains a central role in the hardware stack for modern machine learning.
Intel Capital
The venture capital division of Intel that invests in startups focused on cloud computing, silicon, and data center infrastructure.
CScale
A startup founded in 2023 specializing in optical interconnects designed to maintain uptime in hyperscale AI compute environments.
Sanjai Kohli
The founder and CTO of CScale who previously co-founded SiRF, a company recognized for advancements in GPS and semiconductor technology.
The details
CScale combines expertise in photonics, electronics, and software to design hardware that maintains data integrity during signal fluctuations. By localizing failures within the interconnect fabric, the system prevents a single optical component degradation from triggering a fleet-level compute outage. This design is intended to sustain performance across massive clusters where hardware errors are statistically inevitable.
Timeline
2014: Facebook acquired Inovi.
2023: CScale was founded.
September 2026: Nvidia and Intel Capital invested in CScale.
End of decade: Expected deployment of gigawatt-class AI datacenters.
The Tech Race
The investment underscores the growing industry focus on interconnect reliability as AI compute capacity grows exponentially. It places CScale within a competitive landscape where firms are racing to solve the throughput and stability challenges of massive, multi-accelerator systems.
This development serves as a foundation for future hardware reliability in large-scale AI clouds, though there is no immediate change for individual users. The startup's success will be measured by the eventual stability and uptime of the large-scale compute platforms that power global AI services.
The takeaway
The race to scale AI compute has moved beyond raw chip speed to the physical reliability of the interconnects that link them. Watch for future performance benchmarks from CScale as industry infrastructure trends toward the gigawatt-scale deployments expected by the end of the decade.
Further reading
For more on the infrastructure supporting modern machine learning, visit our Data Centers section.
Live Poll
Do you expect the cost of new AI-driven technology to make personal electronics more expensive?
