The Networking Layer That 800% ARR Growth Says You Can't Skip

01 /The Concept

The GPU shortage got the headlines. The networking bottleneck didn't.

Before a neocloud can sell a single hour of compute to a customer, the networking layer has to work: switches configured, tenant isolation enforced at hardware level, traffic separated so that one customer's workloads cannot bleed into another's. Hyperscalers solved this problem years ago by building proprietary automation internally. Every other operator has solved it the same way infrastructure operators have always solved network configuration – manually, slowly, by people with deep expertise, one rack at a time.

Alex Saroyan, Arsen Arakelyan, and Tigran Martirosyan founded Netris in 2018 – before the AI infrastructure boom, before neoclouds were a named category, before anyone outside networking circles cared about multi-tenant GPU cluster management. They built software that runs on network switches and automates the full lifecycle of GPU cluster networking: setup, configuration, operations, and tenant isolation enforced at hardware level rather than in software – the distinction that makes multi-tenancy both more secure and more operationally scalable. NVIDIA was impressed enough by an early demo that it began recommending Netris to its own customers.

When the AI infrastructure market accelerated in 2025 and 2026, Netris was already in position. ARR grew 800% in twelve months. A $15 million Series A from a16z closed in June 2026, led by Guido Appenzeller – former CTO of VMware's Cloud and Networking division and co-founder of Big Switch Networks – who joined the board. Thirty-five live GPU cluster deployments: Lightning AI, Foxconn-backed Visionbay, Hewlett Packard Enterprise, TensorWave, and Telus among them.

02 /Why It Matters

An 800% ARR year does not happen for a product that is nice to have. It happens for a product that is on the critical path – something that must exist before anything else can function.

The reason networking automation is on the critical path for neoclouds is that manual configuration at the scale of a GPU cloud serving dozens of enterprise customers simultaneously breaks. Not slowly, and not predictably. The multi-tenancy requirement is particularly hard: hardware-level tenant isolation requires specialized knowledge that most cloud operators don't have internally, and software-level isolation introduces security and performance tradeoffs that enterprise customers won't accept. Netris encodes eight years of that knowledge into deployable software.

The a16z thesis behind this investment is specific: the internal networking capabilities that hyperscalers built for themselves will, over time, need to exist as productized software available to every AI data center operator. The category didn't have a name until Netris named it. The company's ARR growth suggests the category now exists whether or not the name spreads.

03 /Opportunities

Netris defined the NAAM category – Network Automation, Abstraction, and Multi-Tenancy – and for now occupies it alone. The adjacent problem, compute lifecycle management for AI data centers, has not been productized the same way.

Networking is one layer. Compute scheduling (which workloads run on which nodes), node failure detection and automated failover, storage allocation, and capacity utilization visibility across heterogeneous clusters are others. None of these have an equivalent of Netris. The neocloud operator who discovered Netris for networking in 2025 is today managing workload queuing and cluster health with internal tooling, vendor dashboards, and spreadsheets.

The entry angle is vertical orchestration software for independent AI data center operators – the neoclouds at 50 to 500 GPU scale that lack the internal engineering depth of a hyperscaler. Netris solves their networking; nothing comparable solves their compute lifecycle. The qualification: this product requires deep knowledge of how GPU workloads actually behave at the scheduling layer, which means the founding team needs either hyperscaler infrastructure background or time embedded with neocloud operators as the first design partner.