a16z just backed the company that makes GPU clouds actually work

Your AI data center can buy 10,000 GPUs.

But if you can't wire them together in weeks instead of months, those GPUs are expensive paperweights.

Netris just raised $15M from a16z to solve the problem nobody talks about.

Every GPU server carries at least 23 network connections across Ethernet, InfiniBand, NVL72, and DPUs.

Every time you add a tenant, resize a cluster, or remove a customer — thousands of switches need coordinated reconfiguration.

One misconfiguration takes the cluster down. Or leaks tenant data across boundaries.

Legacy SDN can't handle it. It's a software technology running on hardware that demands hardware-accelerated speed.

Netris built NAAM — Network Automation, Abstraction, and Multi-Tenancy.

Software that runs directly on the switches themselves.

35+ live GPU clusters worldwide. About 1 million GPUs.

800% ARR growth in 12 months.

Customers include Foxconn's Visionbay (Taiwan's largest GPU cluster), HPE, Lightning AI, TensorWave, and Telus.

Nvidia recommended Netris to its customers two years ago after a single demo.

a16z partner Guido Appenzeller is joining the board. The same Guido Appenzeller who helped revolutionize datacenter networking at VMware.

IDC projects AI infrastructure will hit $758 billion by 2029.

The bottleneck isn't GPUs. It's the network fabric connecting them.

Audit your AI infrastructure deployment timeline.

If your network configuration is still manual, you're burning GPU capital every day those machines sit idle waiting for switches to be wired.

The neocloud wars just shifted from who has the most GPUs to who can wire them fastest.

SOURCE: https://techcrunch.com/2026/06/25/netris-raises-15m-series-a-from-a16z-to-help-ai-neoclouds-go-live-faster/
VERIFIED: TechCrunch, Netris official blog, The Next Web
SIGNAL: The AI infrastructure buildout bottleneck has moved from GPU procurement to network automation — and a16z is betting the next layer of infrastructure value sits in the networking fabric, not the silicon.