Cornelis Networks Raises $205M, Skips a Valuation Number

Cornelis Networks, a networking-chip maker that spun out of Intel in 2020, said Sept. 14 it raised $205 million led by IAG Capital Partners. Neither Cornelis nor IAG disclosed a valuation for the round, according to TechCrunch's report and Cornelis's own announcement.

"AI infrastructure is reaching a point where faster endpoints alone are not enough," said Lisa Spelman, Cornelis's chief executive. "The fabric has to become an active part of the compute system." IAG Capital Partners' Joel Whitley put a number on the opportunity his firm is betting on: "Open-standard scale-up and scale-out networking for AI represents more than $55 billion of opportunity by 2030," he said, a figure that is IAG's own market estimate rather than a disclosed revenue projection for Cornelis itself.

What the money is funding

The round came with a strategy collaboration with Qualcomm Technologies, announced at the AI Infra Summit, where the two companies said they would design networking and accelerator silicon together rather than bolting a network onto a finished chip. "As AI systems continue to scale, moving data efficiently across the rack becomes just as important as the compute itself," said Tony Pialis, executive vice president and general manager of Qualcomm's data center business.

Product	Status	Availability
CN5000 switch	Shipping	Now
CN6000 switch	Sampling with customers	Wider availability in Q4 2026

Cornelis is selling an alternative to Nvidia's approach, in which the network is built to run on Nvidia's own software stack. Cornelis's Active Compute Fabric is built on open standards instead: UALink and ESUN for connecting chips within a rack, and Ultra Ethernet for connecting racks to each other, according to SiliconANGLE's coverage of the launch. The pitch is that a customer can mix GPUs and accelerators from different vendors on the same fabric, rather than being locked into one supplier's hardware.

The waste Cornelis says it is selling a fix for

Cornelis's own modeling, which it says covers 100,000-GPU systems, found that up to half of GPU hours go to chips sitting idle while they wait for data, a gap the company valued at $1.68 billion a year in lost compute capacity and 500 gigawatt-hours of power, per SiliconANGLE's report. That estimate is Cornelis's, not an independently audited figure, and networking is a comparatively small share of what a data center spends on AI hardware: Cornelis puts it at roughly 15 percent of total system cost, arguing that share still determines how much value the other 85 percent actually delivers.

What the announcement doesn't name

Cornelis did not name a single paying customer for either switch, disclose revenue, or say how the $205 million compares to money it has raised since leaving Intel. That puts the round in different company than Ayar Labs, which priced a $5 billion secondary sale this month on a specific number, or Fab2's $500 million Series A, priced at a $3.7 billion valuation: both of those AI-infrastructure hardware rounds came with a price attached, even if the number itself was contested. Cornelis's did not. The company's next public test is whether CN6000 actually reaches the wider availability it targeted for the fourth quarter of 2026, and whether the Qualcomm collaboration produces a named joint product rather than a keynote appearance.