Cornelis Networks
Cornelis Networks provides an active compute fabric for AI and HPC clusters that performs computation on data while it is in transit, rather than merely forwarding packets. The CN5000 Omni-Path fabric delivers 400 Gb/s ports with in-network collective acceleration, while the CN6000 extends to 800G multi-protocol support including RoCEv2 and Ultra Ethernet. The company designs every layer from silicon through software, with OEM validation across more than 50 server configurations.
- Artificial Intelligence
- Hardware
- Enterprise Software
- Semiconductor
Funding
Raised to date
$254MRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
across 3 rounds
+7$205M · Series C
Announced Sep 2026
Disclosed cumulative: $254M
Founders
Founder details are not available yet.
Product
Problem
AI and HPC clusters often suffer from communication bottlenecks that leave expensive GPUs idle while waiting for data movement, synchronization, and collective operations. Traditional passive networks only forward packets between endpoints, forcing congestion control and collective communication to run on host accelerators, which reduces overall system utilization and slows time-to-solution for distributed workloads.
Solution
Cornelis Networks provides an active compute fabric that acts on data while carrying it, turning the network from passive transport into a programmable compute layer. The architecture spans scale-up and scale-out with one open design, enabling congestion control, collective acceleration, and workload-specific operations to execute in the fabric rather than on accelerators. The CN5000 Omni-Path fabric delivers lossless, congestion-free scale-out networking with 400 Gb/s ports, while the CN6000 extends the path with 800G multi-protocol support including RoCEv2 and Ultra Ethernet. The company designs every layer from silicon through software, tuning the full fabric as one system to deliver higher message rates, lower latency, and better performance per networking dollar than competing solutions.
Target Audience
Primary customers are national laboratories, enterprise HPC and AI centers, OEMs, and compute partners running demanding AI training, inference, and high-performance computing workloads that require low-latency, high-throughput interconnects.
Features
- CN5000 SuperNICs with one or two 400 Gb/s ports on a low-profile x16 PCIe Gen 5 adapter
- Director Class Switches with 576 ports at 400 Gb/s, delivering up to 230.4 Tb/s in a two-tier fat tree configuration
- Active Compute Fabric with three cumulative levels: active transport with credit-based flow control, active acceleration with native in-network collectives, and active compute with programmable Distributed Compute Fabric capabilities
- CN6000 multi-protocol SuperNIC supporting Omni-Path, RoCEv2, and Ultra Ethernet on PCIe 6.0
- In-network collective operations including AllReduce, AllGather, and ReduceScatter with UEC conformance and in-network quantization
- KV Cache Acceleration pipeline for disaggregated serving and MoE Expert Dispatch with content-addressed deduplication for inference workloads
- OEM validation across more than 50 server configurations with over 10 partners, including published solution briefs and reference architectures