Another Startup Is Coming for Nvidia. This One Wants the Wires.

Nvidia’s crown does not sit only on its chips; a huge part of its lock on AI data centres is the networking that stitches thousands of those chips together, its InfiniBand and NVLink kit. On 14 September, a company called Cornelis Networks raised $205 million, led by IAG Capital Partners, and unveiled “Active Compute Fabric,” an open, GPU-agnostic networking layer aimed squarely at that unglamorous but lucrative corner of Nvidia’s empire. It also announced a collaboration with Qualcomm. The pitch: make the network itself do some of the compute, and do not tie customers to one vendor’s silicon.

The investment angle, both ways

The bull case is that networking is one of the most defensible parts of Nvidia’s grip and therefore one of the most valuable to crack. Every hyperscaler building giant AI clusters wants to avoid being locked into a single supplier for chips and the fabric between them, so an open, vendor-neutral alternative has a large and motivated customer base. Qualcomm’s involvement adds real credibility, and $205 million is enough to actually build rather than just pitch.

The bear case is the same one that buries most Nvidia challengers. Nvidia’s advantage is not only performance, it is a tightly integrated stack that already works, plus software the whole industry has standardised on. “Open and vendor-neutral” is a great slogan and a brutal engineering reality, and Cornelis is years from proving it at the scale the hyperscalers demand. Plenty of well-funded companies have promised to loosen Nvidia’s grip on the data centre and disappeared.

The takeaway

The useful frame: the smart money has worked out that you do not beat Nvidia head-on at chips, so it is probing the flanks, and networking is one of the softest. Cornelis is a clean bet on the thesis that AI infrastructure gets more open and less Nvidia-shaped over time. If you believe that, this is one way to express it; if you think Nvidia’s integrated stack is too entrenched to dislodge, it is another $205 million learning experience. (Not investment advice.)

Related: Positron AI’s tilt at Nvidia on inference.

Get the free weekly stack: the AI tools and moves that matter, hype filtered out.Subscribe free →
Scroll to Top