Cornelis Networks Raises $205 Million to Tackle AI Networking Bottlenecks
The Intel spin-off launches 'Active Compute Fabric' and partners with Qualcomm to reduce GPU idle time in AI clusters.
AI infrastructure startup Cornelis Networks has secured $205 million in new funding to accelerate the deployment of a network architecture designed to eliminate data bottlenecks in large-scale AI clusters. The funding round was led by IAG Capital Partners.
Alongside the capital injection, the company launched its "Active Compute Fabric," an open network architecture built to support both scale-up and scale-out AI workloads. To further its reach into data center optimization, Cornelis also announced a strategic collaboration with Qualcomm Technologies focused on optimizing rack-scale AI networking.
The Shift to Active Networking
Cornelis Networks originated as a spin-off from Intel's Omni-Path business in 2020. Since its inception, the company has positioned itself as an open-standard alternative to the proprietary ecosystems currently dominating the high-performance computing (HPC) and AI sectors, most notably Nvidia's InfiniBand.
The company is targeting the "passive pipe" problem, where expensive GPUs waste significant processing cycles waiting for data synchronization and transport. By transforming the network from a simple conduit into an active participant in the compute process, Cornelis aims to integrate programmable compute directly into the fabric. This allows the network to handle operations such as gradient compression and KV cache assembly while data is in transit.
"AI infrastructure is reaching a point where faster endpoints alone are not enough," said Lisa Spelman, CEO of Cornelis. "The fabric has to become an active part of the compute system."
Industry Implications
As AI clusters grow in complexity, the primary bottleneck has shifted from raw compute power to the efficiency of data movement. By reducing the time GPUs spend idle, Cornelis intends to increase the return on investment for organizations spending billions on accelerator hardware.
This approach directly challenges the vertical integration of the current market leader. By utilizing an open architecture, Cornelis seeks to break the lock-in associated with proprietary networking stacks, offering a more flexible path for data center operators to scale their AI capabilities without being tied to a single vendor's hardware ecosystem.
The Mechanics of the Fabric
According to Brandon Draeger, Chief Marketing Officer at Cornelis, the Active Compute Fabric changes how data is delivered to the destination. "The payload does not arrive the way it left," Draeger explained. "In a collective operation, partial results from many endpoints are combined inside the fabric, so a single reduced result lands at the destination instead of thousands of separate contributions."
What's Next
Market observers will now look to the results of the Qualcomm partnership to see how effectively these optimizations translate to real-world rack-scale deployments. While the funding provides a significant runway, the ultimate success of the Active Compute Fabric will depend on its adoption rate among hyperscalers and enterprise AI labs currently reliant on established proprietary standards.