DriveNets is selling the less glamorous side of AI infrastructure: the network that keeps expensive GPUs from waiting on data. CEO Ido Susan framed the problem bluntly: the most expensive idle asset today is a GPU waiting on the network.
The company raised $410 million in Series D funding at an $8.5 billion valuation, bringing total funding close to $1 billion. Bessemer and Atreides led the round, with Pitango, D1, Red Dot and AMD also participating.
Why networking is becoming strategic
At large cluster scale, GPU performance depends on how well thousands of chips communicate. DriveNets uses Ethernet and end-to-end optimization across libraries, transport, network cards and fabric scheduling to keep those clusters full.
The company says it has been cash-flow positive since 2025 and holds more than $1 billion in orders and projects. Susan said the new capital is mainly for inventory so DriveNets can meet AI infrastructure demand already in the pipeline.
AMD is the signal
AMD's participation matters because AI data centers are choosing between Nvidia's tighter InfiniBand ecosystem and a more open Ethernet camp. For AMD, better open networking reduces the lock-in that can keep customers inside Nvidia's stack.
Analyst Alan Weckel of 650 Group expects the AI networking market to exceed $200 billion by the end of the decade. DriveNets is betting that as clusters become heterogeneous, the wire between accelerators becomes as strategic as the accelerator itself.
Sources: PR Newswire, Calcalist, Globes, Tech Startups, CocoLoop; checked the $410 million Series D, $8.5 billion valuation, nearly $1 billion total funding, $1 billion-plus backlog and $200 billion AI networking market estimate.