Cornelis Networks is opening a new front in AI infrastructure: the Intel spin-off says it is moving into scale-up networking for rack-scale systems, raising $205 million and exploring a collaboration with Qualcomm. The important bit is strategic, not merely financial: Cornelis wants to give AI-platform builders an open alternative to NVIDIA’s proprietary NVLink.
Watch Desk analysis
What happened
CRN reports that Cornelis is expanding its Active Compute Fabric architecture beyond traditional scale-out networking. Scale-up networking links processors inside a rack, where fast communication is increasingly central to accelerator-heavy AI systems.
Cornelis says its architecture can add programmable compute to network interfaces and switches, allowing data to be processed as it moves through the system. It also plans to support open standards including UALink and Ultra Ethernet. Qualcomm executives described the approach as aligned with their data-centre accelerator and CPU plans, but the companies have not disclosed detailed collaboration terms.
Key findings
- A new NVLink alternative
Cornelis is targeting the high-speed links inside AI racks, where NVIDIA has built a powerful proprietary position. - $205m of fresh funding
The money is intended to fund future products, manufacturing and customer deployments. - Potential Qualcomm connection
The companies describe shared infrastructure goals, but there is no announced product, timetable or commercial agreement yet. - Programmable networking
Cornelis says its fabric can offload collective operations and adapt to changing AI workloads. - Open standards in the mix
The company plans to support UALink and Ultra Ethernet alongside its own architecture.
Why it matters
AI infrastructure is becoming a battle over the links between chips, not just the chips themselves. If companies building rack-scale systems can choose networking that works across different accelerators, they may gain more flexibility over supply, pricing and platform design.
That is the promise. The less glamorous test is whether Cornelis can turn an architecture announcement and a potential partnership into shipping products that match NVIDIA’s ecosystem, reliability and software support. AI hardware has many grand entrances; customers tend to wait for the loading dock.
Our read
This is a meaningful infrastructure move with a clear technical thesis: make the network more open and more computationally useful. Treat the Qualcomm connection as exploratory rather than signed business, and watch for product specifications, deployments and independent performance data before calling it a serious NVLink challenger.
What to watch
- Whether Cornelis announces a concrete Qualcomm product integration or customer deployment.
- Shipping dates and specifications for the scale-up products.
- Support for UALink, Ultra Ethernet and other open interconnect standards.
- Independent evidence that programmable networking improves AI utilisation or economics.
Discussion spark: Can an open rack-scale networking layer genuinely loosen NVIDIA’s grip, or will software and ecosystem support matter more than the interconnect standard?
Sources and evidence
- Cornelis Mounts Scale-Up Challenge To Nvidia, Teases Potential Qualcomm Tie-Up – crn.com (14 September 2026, 18:37 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.