• Virtual data networking specialist DriveNets has already proven itself in major telco networks
• Now it has its eye on the burgeoning AI factory networking opportunity
• It has raised $410m, with AMD among its new investors

Virtual routing system specialist DriveNets has raised a staggering $410m Series D funding round to help take its software-based networking solutions into AI factories and datacentres to tackle key infrastructure efficiency challenges.   

DriveNets has long been a disruptor in the wide area data networking sector, successfully competing with major IP network infrastructure names, such as Cisco, Nokia and HPE (Juniper Networks) to secure deployments for its Network Cloud stack at major network operators, such as AT&T, Comcast and KDDI.  

And for the past few years, using the same tech foundations, it has been developing  Ethernet-based ‘AI fabric’ solutions designed to address some of the main networking issues encountered in large-scale AI infrastructure deployments run by hyperscalers, AI developers, neoclouds, large enterprises and more. 

And now the company believes it can meet the needs of AI infrastructure companies that need to find every possible way to maximise the return on their substantial investments in the graphics processing units (GPUs) sourced from Nvidia (the clear market leader) and AMD. 

“This financing round marks a pivotal step in scaling our company to meet the surging demand for large-scale AI infrastructure,” stated Ido Susan, CEO and co-founder of DriveNets. “The most expensive idle asset in the world right now is a GPU waiting on the network. We’re applying a decade of high-performance networking expertise to enable our customers to achieve higher utilisation, reduce cost per workload, and scale their AI operations efficiently – on any AI accelerator they choose.” (More on the relevance of that multivendor architecture later.) 

DriveNets’ AI fabric solutions have been developed for scale-up, scale-out, and scale-across architectures, as well as for the front-end and storage connectivity requirements associated with large-scale AI compute clusters. (Scale-up refers to connectivity within an individual rack/machine, scale-out to connectivity between racks, and scale-across to connectivity between AI factory sites.) 

According to DriveNets, the networking optimisation capabilities of its technology can address “two fundamental constraints in AI infrastructures today: Large GPU clusters operating below peak efficiency due to network bottlenecks and reliability challenges; and slow cluster bring-up time (‘idle capex’), especially in multi-vendor environments.”

The Ra’anana, Israel-based vendor notes that “some of these optimisations are developed in collaboration with leading AI accelerator vendors, such as the recently published validated reference architecture for AMD-DriveNets-based clusters that maximises GPU utilisation, reduces cost-per-token, and enables rapid deployment and efficient end-to-end scaling.”

It will come as no surprise, then, that chip giant AMD participated in the new round of funding, which was led by Bessemer Venture Partners and Atreides Management but also included existing investors Pitango and D1 Capital Partners as well as new investors Red Dot Capital and, of course, AMD.

The new round takes the total funding raised by DriveNets to $1bn: The company also claims to have more than $1bn in secured business and to have been cash-flow positive since 2025. DriveNets says it will use the new funding to ”scale inventory to support its growing AI fabric pipeline and expand its “heterogeneous AI infrastructure solutions”. 

Heterogeneous AI refers to a disaggregated approach to AI factory networking that mirrors the open architecture approach that DriveNets has helped to promote in wide area networking.

The vendor explains: “The recent AI infrastructure spending shift from training to inference is expected to drive the adoption of heterogeneous AI architectures that bring infrastructure costs down and optimise power utilisation. Heterogeneous AI architecture uses multiple AI accelerators from multiple vendors within the same cluster, each best for a different stage or task within the AI training or inference process. The compute resources in the cluster are orchestrated to provide the best overall performance and power utilisation, to reduce cost per million tokens and maximise tokens per watt. DriveNets’ AI fabric is uniquely positioned to support heterogeneous multivendor AI environments due to its ability to perform full-stack optimisation for any AI accelerator in the cluster, maximising the performance and utilisation of the entire cluster.”

For more on what the company means by heterogeneous AI, and why it believes CPMT (cost per million token) is the appropriate measure of AI infrastructure economic efficiency, rather than the traditional TCO (total cost of ownership), see this blog. 

Its lead investor clearly believes this approach will pay off. Adam Fisher, a partner at Bessemer Venture Partners (BVP) noted: “Every shift in compute produces a new networking giant. Cisco wired the internet. Arista wired the cloud. Nvidia wired single-vendor AI. DriveNets is wiring what comes next: Heterogeneous AI. This is why BVP led DriveNets’ $410M Series D, an existing portfolio company we’ve backed since its Series A.”

And there’s a lot riding on this trend, according to Alan Weckel, founder and technology analyst at 650 Group. “AI networking [spending] is on track to surpass $200bn by the end of the decade, driven by the shift from single-vendor stacks to multivendor and later heterogeneous AI infrastructures. DriveNets enters this phase with a strong combination – tier-one service provider reliability, validated AMD reference design, and the inventory position to deliver into a supply-constrained market. That positions the company well as open Ethernet becomes the foundation of next-generation AI infrastructure,” said Alan Weckel, founder and technology analyst at 650 Group.

So what’s next? DriveNets notes that, as part of its efforts to encourage uptake of its AI fabric solutions, it is “now working with leading AI vendors, such as AMD, Broadcom and others, to tighten the integration between networking and compute in multivendor AI environments, maximising cluster performance and GPU utilisation to substantially improve token economics.” The company’s head of product marketing, Dudy Cohen, noted earlier this year that DriveNets was already “participating in proofs-of-concept in some of the largest AI clusters in the world (we’re talking LARGE, very large, clusters!).”

It should also be noted that DriveNets faces the not inconsiderable task of persuading AI factory operators to use its Ethernet-based networking fabric rather than Nvidia’s proprietary InfiniBand AI datacentre networking solution. 

- Ray Le Maistre, Editorial Director, TelecomTV