Cornelis Networks raises $205M, launches Active Compute Fabric to challenge Nvidia's AI dominance

Cornelis Networks Secures $205 Million to Challenge Nvidia's Dominance

Cornelis Networks, an Intel spin-off focused on AI infrastructure, announced Monday it has raised $205 million in a funding round led by IAG Capital Partners1

1

3

3

. The fresh capital will support the development of future scale-up networking and scale-out products, as well as manufacturing and customer deployments for its new CN6000 family4

4

. Spun out of Intel in 2020, the Wayne, Pennsylvania-based company is positioning itself as part of a new wave of AI networking startups emerging to break apart Nvidia's market dominance piece by piece1

1

.

Active Compute Fabric Targets GPU Idle Time Problem

Alongside the funding announcement, Cornelis Networks unveiled Active Compute Fabric, an open architecture that integrates programmable compute directly into data center networks3

3

. The technology addresses a critical bottleneck: much GPU time is wasted waiting for data to arrive. According to Cornelis, in a 100,000 GPU system, roughly half of all GPU hours are spent waiting for data, worth approximately $1.68 billion a year in wasted capacity and 500 GWh of power2

2

. Active Compute Fabric enables chips to process and send information simultaneously, allowing the network to operate on data as it moves through the system1

1

2

2

.

How Active Compute Fabric Reduces Network Overhead

The architecture performs workload-specific operations as data traverses the network rather than acting merely as a messenger. Chief Marketing Officer Brandon Draeger explained that in collective operations, partial results from many endpoints are combined inside the fabric, so a single reduced result lands at the destination instead of thousands of separate contributions3

3

. Active Compute Fabric can accelerate functions including KV cache offloading for storing model state across sessions, mixture-of-experts expert dispatch to reduce duplicate transfers, message passing interface offload for HPC applications, and in-fabric checkpointing for failure recovery2

2

. Pre-production simulations suggest this approach can reduce overall network traffic by up to 50%3

3

.

Open Architecture Offers Alternative to Nvidia's Closed Ecosystem

Cornelis Networks is competing with Nvidia by offering an open architecture, meaning customers can use a variety of GPU and accelerator hardware with Cornelis' networking fabric1

1

. Built on open standards such as Ethernet, UALink for scale-up networking, and Ultra Ethernet for scale-out, organizations can implement Active Compute Fabric with their existing compute architectures3

3

. This contrasts with Nvidia chips, which technically can run on other networking fabrics but are optimized to run on Nvidia's own software, making it more enticing for customers to use Nvidia's full GPU stack1

1

. CEO Lisa Spelman stated the company is seeing growing demand from customers for an open alternative that gives them more choice in how they build their AI infrastructure3

3

.

Strategic Collaboration with Qualcomm Signals Rack-Scale AI Push

Cornelis Networks announced a strategic collaboration with Qualcomm Technologies at the AI Infra Summit, aimed at shaping the future of networking for rack-scale AI data centers3

3

4

4

. While details remain limited, the partnership signals potential integration of their technologies. Qualcomm plans to debut its first rack-scale platform this year with its AI200 accelerator chip4

4

. Tony Pialis, general manager of Qualcomm's data center business, said Cornelis' vision for an open, programmable fabric aligns with the need to improve GPU utilization and AI economics with a more integrated approach across compute, memory and networking4

4

.

Entry Into Scale-Up Networking Challenges NVLink

Cornelis' expansion into scale-up networking represents a direct challenge to Nvidia's proprietary NVLink interconnect technology, which enables high-speed connections between processors inside rack-scale systems4

4

. As Nvidia expands its influence through NVLink Fusion tech, rival networking vendors including Cornelis Networks and Delos Data are scrambling to bring alternative interconnects and switches to market2

2

. The company's new 800 Gbps-capable CN6000-series switches and NICs signal its entry into this arena2

2

. Cornelis has already started shipping its product and is working on a new generation expected to be released later this year1

1

.

What This Means for AI Workloads and Data Center Economics

The launch of Active Compute Fabric addresses a fundamental shift in AI infrastructure requirements. Spelman noted that AI infrastructure is reaching a point where faster endpoints alone are not enough, and the fabric has to become an active part of the compute system3

3

. An executive at a large U.S. systems integrator told CRN that many enterprise customers would welcome new data center suppliers who embrace open standards because it could potentially help them with product choice, supply and pricing4

4

. The ability to work agnostically across different vendors carries cost implications, availability implications, and reduces dependencies4

4

. Watch for how quickly Cornelis can deliver on its promise to reclaim wasted GPU capacity and whether its open approach gains traction against Nvidia's established ecosystem.