Netris raised $15M in a Series A led by Andreessen Horowitz to expand network automation software that helps AI neoclouds get GPU clusters online faster. The eight-year-old startup is already deployed across more than 35 GPU clusters worldwide, covering roughly 1 million GPUs at operators including Lightning AI, Foxconn, Hewlett Packard Enterprise, TensorWave, Visionbay, and Telus. a16z partner Guido Appenzeller is joining the Netris board as part of the round.
The pitch is narrow and concrete: shave the months it takes a new neocloud to configure switches, isolate tenants, and start serving paying inference and training workloads. Every day a GPU cluster sits idle waiting on network setup is a day the operator is burning capital on hardware that cost tens of thousands of dollars per accelerator. Netris sells software that runs on the switches themselves plus a control platform that automates setup, ongoing configuration, and multi-tenant isolation at the hardware layer.
Until the neocloud wave, network plumbing at this scale was largely the problem of Equinix, NTT, Digital Realty, Oracle, Microsoft, AWS, and Google — operators with the engineering benches to build their own automation. The new entrants chasing AI demand do not have those benches. They have racks of Nvidia and AMD GPUs, a customer list, and a clock ticking on amortization.
Key facts
- 01Netris raised a $15M Series A led by Andreessen Horowitz, with a16z partner Guido Appenzeller joining the board.
- 02The platform is deployed at more than 35 GPU clusters globally, covering roughly 1 million GPUs.
- 03Customers include Lightning AI, Foxconn, Visionbay, Hewlett Packard Enterprise, TensorWave, and Telus.
- 04Nvidia began recommending Netris to customers two years ago after seeing a demo of its technology.
- 05The company has spent 8 years building hardware-accelerated network automation, predating the AI boom.
Netris CEO Alex Saroyan frames the technical gap as a mismatch between traditional software-defined networking and the traffic volumes AI clusters generate. SDN moves configuration through software control planes that were never built for the east-west traffic patterns of distributed training jobs. Netris keeps the abstraction model of SDN but pushes execution down into hardware-accelerated paths on the switches.
The platform is vendor-agnostic across the networking gear that Nvidia and AMD server fleets typically sit behind, which matters because neoclouds rarely standardize on a single switch vendor. Customers get an abstracted topology view, the ability to reconfigure links without manual switch-by-switch work, and tenant isolation at the hardware layer so a single cluster can serve multiple paying customers without performance bleed.
Nvidia took notice two years ago. After seeing a Netris demo, the chipmaker began recommending the startup to its own customers — a meaningful endorsement in a market where Nvidia's reference architectures effectively set the standard for how neoclouds get built. The customer roster that followed reads like the second tier of the AI infrastructure stack: contract manufacturers, regional telcos, and specialty GPU clouds racing to catch the hyperscalers.
Notably, Netris does not use AI inside its own product. Saroyan said the company relies on deterministic algorithms it developed before the current AI cycle began, on the view that switch configuration is a domain where creativity is a liability and repeatability is the entire point. It is a quietly contrarian position for a company selling into AI infrastructure.
The Series A will fund hiring across engineering and sales, broader hardware vendor support, and expanded functionality in the automation algorithm. The capital is modest by 2026 infrastructure-round standards — Cerebras and CoreWeave have raised orders of magnitude more — but Netris is selling software into a market where the buyers are spending nine and ten figures on GPUs and need every piece of the stack to just work.
The risk is that the neocloud boom itself thins out. If GPU pricing softens, if Nvidia's direct cloud partnerships consolidate demand, or if hyperscalers undercut the second tier on price, the addressable customer base for Netris could shrink faster than its product roadmap can broaden. The 35-cluster footprint is real traction, but it is concentrated in a customer segment whose long-term economics are still being tested.
The bet a16z is making is structural rather than cyclical: every new AI data center, regardless of who operates it, needs hardware-accelerated network automation, and the operators who lack hyperscaler-grade engineering teams will keep buying rather than building. If the neocloud category survives the next 18 months of GPU pricing volatility, Netris is positioned as the default plumbing layer for the operators who outlast the shakeout — and that is a more defensible business than chasing the model layer with another fifteen million dollars.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
