Intel spinoff Cornelis raises $205M for AI networking aimed at Nvidia's stack

Cornelis, an AI infrastructure company that spun out of Intel in 2020, has raised $205 million in a funding round led by IAG Capital Partners, TechCrunch reported on Monday. The company paired the raise with the launch of Active Compute Fabric, a networking product built to fix a quiet but expensive problem in AI datacenters: GPUs sitting idle while they wait for data to arrive.

The product: overlapping communication with computation

According to TechCrunch, Active Compute Fabric is designed so that chips can process and send information at the same time. Instead of accelerators alternating between computing and waiting on the network, the fabric overlaps the two, recovering GPU time that would otherwise be lost to data movement.

The company is also further along than the average newly funded startup. TechCrunch reports that Cornelis has already started shipping the product and is at work on a new generation expected to be released later this year.

The pitch: an open alternative to Nvidia's stack

Cornelis is positioning itself against Nvidia by offering an open architecture, TechCrunch explains. Customers can run a variety of GPUs and accelerator hardware on the company's networking fabric rather than being tied to a single vendor's silicon.

That openness is the strategic point. As TechCrunch notes, Nvidia's chips can technically run on other networking fabrics, but they are optimized for Nvidia's own software, which makes buying the company's full stack — accelerators, networking and the software that ties them together — the easier and more enticing path for customers. A vendor-neutral fabric attacks that bundling logic from below: if the network no longer steers buyers toward one ecosystem, the rest of the stack becomes open to competition.

Cornelis is not alone in this effort. TechCrunch frames the company as part of a new wave of AI infrastructure businesses, each taking aim at one slice of Nvidia's hold on the market.

Why it matters

The size of the round matters as much as the technology. Two hundred and five million dollars is a serious commitment to a company whose product sits in the network layer rather than on the accelerator itself — evidence that investors see networking as a viable front in the effort to loosen Nvidia's grip on AI infrastructure.

The economics are straightforward. AI clusters are expensive precisely because they depend on thousands of accelerators working in parallel, and any time those chips spend waiting for data is capacity that has been paid for but not used. A fabric that recovers even part of that idle time lowers the effective cost of running models, a metric every AI operator is under pressure to improve.

The openness argument carries longer-term weight as well. Nvidia's dominance rests as much on the gravitational pull of its integrated software ecosystem as on raw chip performance. If open fabrics like Cornelis's prove credible, accelerator choice becomes a genuine procurement decision rather than a default — a shift that would ripple across the whole industry. With the product shipping now and a next generation due later this year, per TechCrunch, customers will not have to wait long to test that premise.