$410M Went Into the Wires, Not the GPUs — DriveNets Raises a $410M Series D for AI Ethernet Fabric

AI money is moving from "the GPU" to "the wires between GPUs"

Talk AI and everyone looks at GPUs, NVIDIA, model size. But on June 1, big money went somewhere different. AI networking company DriveNets closed a $410M Series D, pushing total capital raised past $1B at an $8.5B valuation.

Here's why that's interesting: DriveNets doesn't make a model or a chip — it makes "the network that connects AI accelerators." Large-scale AI training lashes tens of thousands of GPUs into one cluster, and overall performance hinges on how fast and unclogged the data moves between those GPUs. However fast the chip is, if the "wire" linking them is the bottleneck, it's wasted. DriveNets solves that wire over Ethernet.

This deal captures AI-infrastructure capital shifting from "compute (GPUs)" toward "bottlenecks like networking and power." You can buy all the GPUs you want, but if you can't lash them together efficiently, you're just burning money — so capital is now eyeing the "connective tissue."

The players — DriveNets, and what "Ethernet fabric" even means

DriveNets is an Israel-based networking software company. It started in telco routing software, then pivoted toward "networks for AI data centers" as the AI boom hit. Its core product is the AI fabric — a networking layer that, in plain terms, binds many AI accelerators into one giant mesh.

Two keywords: Ethernet-based and heterogeneous. Large AI cluster networking has been dominated by proprietary tech like NVIDIA's InfiniBand. DriveNets aims to deliver that level of performance over commodity-standard Ethernet — which is more open and less vendor-locked. "Heterogeneous" means supporting environments where AI chips from different vendors (say, NVIDIA and AMD GPUs) run mixed in the same cluster.

The customer base is heavy: AI labs building foundation models, hyperscalers, NeoClouds (AI-native cloud upstarts), and large enterprises. DriveNets works with key ecosystem partners — AMD, Broadcom, Dell, Supermicro — to tighten integration between networking and compute.

What it looks like in numbers

Item	Detail	Note
Round	Series D	Closed June 1, 2026
Size	$410M	Total raised past $1B
Valuation	$8.5B	Post-money
Co-leads	Bessemer Venture Partners + Atreides Management	—
New investors	AMD, Red Dot Capital	AMD is strategic
Existing investors	Pitango, D1 Capital Partners	Follow-on
Backlog	$1B+ in business commitments	Company-stated
Profitability	Cash-flow positive since 2025	Company-stated

Two things stand out. First, AMD joining as a strategic investor. A chip company betting directly on a networking company signals that "a network that lashes our GPUs together efficiently" matters to its own business. Against NVIDIA's InfiniBand ecosystem, AMD wants to grow the open Ethernet camp to create an environment where its chips sell better.

Second, cash-flow positive since 2025. While most AI startups eat losses and burn cash, DriveNets is already making money and sitting on $1B+ in backlog. In an AI-infrastructure market shadowed by "bubble" doubts, a company proving itself with actual revenue and profit is exactly why investors stamped $8.5B on it.

Who gains — DriveNets, AMD, and the AI infra market

For DriveNets, $410M buys "inventory + faster deployment." The AI fabric is a hardware-software solution, so with demand surging, quickly scaling inventory and deployment capacity is market share. Already cash-flow positive with $1B in backlog, adding $410M gives it ammunition to expand aggressively.

For AMD, it's strategic positioning. In AI data centers, one of NVIDIA's strongest moats is the vertical integration of "GPU + InfiniBand network." By investing in an open Ethernet fabric like DriveNets, AMD helps grow an alternative stack — "AMD GPU + Ethernet fabric" — to crack NVIDIA lock-in. It's investing not just in great chips but in the ecosystem where those chips come alive.

For the AI infra market overall, it shows the investment theme moving. The last few years were all "how many GPUs can you secure." But stack tens of thousands of GPUs and utilization sags if networking, power, and cooling can't keep up. Capital is now flowing to those bottlenecks — deals in networking (DriveNets), power, and data (like robot training data) are coming in a row. The "second layer" of AI infrastructure is starting to absorb capital in earnest.

Historical parallels — the wins and limits of "the company that solves the bottleneck"

Tech booms often mint big winners among the "supporting cast that solves the bottleneck" next to the stars. But that seat isn't safe either.

Success — Cisco in the dot-com boom. The real money in the internet boom went not to website companies but to Cisco, selling the routers and switches that carried the traffic — the classic "sell picks during a gold rush." DriveNets eyes a similar position in the AI boom: whoever wins the model race, you still need a network to run those models.

Success — data infrastructure in the cloud era. In the cloud boom the spotlight went to SaaS apps, but the companies laying databases, observability, and networking beneath them (Snowflake, Datadog, etc.) grew into giants. Own the "infrastructure of the infrastructure," and you earn steadily regardless of who wins upstairs.

Limit — the standards-war risk. Networking carries a "standards war" trap. Between NVIDIA's InfiniBand and industry-consortium open Ethernet standards (like Ultra Ethernet), DriveNets' fate hinges on which becomes the de facto standard. However good the tech, if the market tilts to a different standard, it's risky. That's why locking in heavyweight allies like AMD and Broadcom matters — fighting as a camp, not alone.

Competitor counter-plays — NVIDIA, Arista, Cisco

NVIDIA is the biggest rival. It sells not just GPUs but InfiniBand (via the Mellanox acquisition) and its own Ethernet (Spectrum-X), so its weapon is selling "chip + network" as one vertical bundle. As the open camp grows, NVIDIA will counter with "we have our own Ethernet too (Spectrum-X)" and defend on performance and integration convenience.

Arista Networks is a powerhouse in data-center Ethernet switching — a direct rival. Arista is also pushing high-performance Ethernet hard for AI data centers, making "software-centric fabric (DriveNets)" vs. "hardware-switch leader (Arista)" a key contrast in approach.

Cisco, the dot-com-era champion, has a vast sales network and enterprise base. As AI networking grows, Cisco will leverage existing relationships to fold "AI data-center networking" into its portfolio. Upstart DriveNets must hold its ground among giants on the differentiator of being "more open and strong in heterogeneous environments."

So what actually changes — by persona

If you invest in or follow AI infra, watch the "bottleneck theme." The next opportunity after GPUs is in the networking, power, and data that connect and support them. DriveNets' $8.5B valuation and cash-flow positivity signal that this second layer is "a real market with real revenue."

If you run AI at scale or engineer it, remember "buy the GPUs and you're done" is false. The bigger the cluster, the more network efficiency governs total cost and speed — even more so in mixed, heterogeneous accelerator setups. If you want less vendor lock-in, it's worth knowing options like open Ethernet fabric.

If you're a solo developer or founder, direct impact is small but the lesson is clear. In a massive boom, the "star" seats may already belong to giants, but the "supporting role that solves the bottleneck" is always open. While everyone stares at models and GPUs, opportunity hides in connecting, supporting, verifying, and operating them. Selling picks is often safer than mining gold.

References

• PR Newswire — DriveNets Secures $410M Series D
• DriveNets Raises $410 Million Series D To Scale Ethernet Fabric (Pulse 2.0)
• DriveNets Scores $410 Million to Advance Its AI Fabric (Futuriom)
• DriveNets raises $410M more (Light Reading)
• AI networking challenger DriveNets secures $410M (SDxCentral)