Cornelis Networks raises $205M to challenge Nvidia's networking lock-in
Original: AI infrastructure company Cornelis raises $205M to chip away at Nvidia’s dominance
Why This Matters
Networking is one of the last chokepoints Nvidia controls end-to-end; breaking that grip attracts serious capital.
AI networking startup Cornelis announced a $205M funding round led by IAG Capital Partners on September 14, 2026. The Intel spinout's new Active Compute Fabric product lets GPUs process and transmit data simultaneously, targeting wasted GPU idle time and offering an open-architecture alternative to Nvidia's tightly integrated stack.
Cornelis, which spun out of Intel in 2020, is building networking fabric designed to fix one of AI infrastructure's most persistent inefficiencies: GPUs sitting idle while waiting for data to move across a cluster. Its new product, Active Compute Fabric, lets chips send and receive data concurrently rather than sequentially — a meaningful throughput gain at scale.
The $205M round was led by IAG Capital Partners. Cornelis has already begun shipping the current product generation and says a next-generation version is on track for later this year.
The competitive angle is deliberate. Nvidia's GPUs can technically run on third-party networking fabrics, but they're optimized for NVLink and InfiniBand — Nvidia's own networking layer — making it sticky and difficult for customers to mix hardware from other vendors. Cornelis counters with an open architecture that works with a range of GPU and accelerator hardware, lowering the barrier for operators who want to avoid full vendor lock-in.
The raise positions Cornelis alongside a broader wave of infrastructure startups — including custom silicon and memory bandwidth plays — each targeting a specific slice of Nvidia's end-to-end dominance rather than trying to out-GPU the market leader directly.