Network Warfare: Cornelis Secures $205 Million to Break Nvidia’s Data Center Grip

Networking developer Cornelis announced a fresh 205 million dollar investment round led by IAG Capital Partners, positioning the firm to challenge Nvidia’s massive hold over high-performance computing networks.

Alongside the funding news, Cornelis launched Active Compute Fabric, a network architecture targeting a primary bottleneck in modern data centers: processing delay. Standard graphics processing units burn valuable compute cycles idling while waiting for data packages to travel across network cables. Active Compute Fabric fixes this latency issue by enabling processing units to transmit and process incoming data streams at the same time.

Cornelis originally spun out from Intel in 2020, positioning itself as an open-architecture competitor to Nvidia’s proprietary network hardware. Using an open network framework allows data center operators to link different graphics cards, custom silicon accelerators, and processing hardware under a single network architecture.

While Nvidia graphics cards work on alternative networking systems, Nvidia optimizes its hardware to run exclusively on proprietary software stacks, nudging corporate buyers toward its complete hardware ecosystem. Cornelis joins a growing movement of hardware startups aiming to break Nvidia’s monopoly by giving buyers open alternatives.

Cornelis already ships its core product line to commercial clients while engineering a next-generation architecture release scheduled for rollout later this year.

Reducing network delay solves a major cost driver for data center operators. Modern computing systems link thousands of individual processor chips across extensive server racks. When data transfers slow down, expensive processing chips sit idle, consuming power while waiting for input queues to clear.

Faster data transfers maximize processing output across every server rack. Data centers running active compute fabrics process complex software workloads faster while spending less on idle power draw.

Offering open network hardware also gives buyers crucial leverage against single-vendor lock-in. When companies buy proprietary hardware stacks, upgrading components forces them to replace entire server systems. Open standards allow engineers to swap out specific processors, upgrade graphics cards, and integrate custom silicon chips without replacing existing network gear.

As demand for raw compute power scales up, building open, high-speed data channels becomes essential. Cornelis’ open approach gives technology companies the flexibility to build custom computing setups without getting trapped inside a single manufacturer’s closed system.