The picks-and-shovels play for the AI infrastructure gold rush just got a $15M vote of confidence from Andreessen Horowitz.
The Summary
- Netris raised $15M Series A led by a16z to accelerate deployment of AI-focused data centers ("neoclouds")
- The company's software runs directly on network switches, cutting the time and complexity of standing up new compute infrastructure
- This matters because GPU scarcity has spawned dozens of new cloud providers, and none of them want to wait 18 months to go live
The Signal
The AI compute shortage created a new category: neoclouds. These aren't AWS or Azure. They're purpose-built GPU farms from players like CoreWeave, Lambda Labs, and a hundred others you haven't heard of yet. They exist because hyperscalers can't build capacity fast enough to meet demand from companies training foundation models.
But there's a problem. Standing up a data center is hard. The networking layer, specifically, is a nightmare of manual configuration, vendor lock-in, and tribal knowledge. Traditional approaches take 12-18 months from breaking ground to serving customers. Netris compresses that timeline.
"Neoclouds need to move at startup speed with enterprise reliability, and the networking stack wasn't built for that."
Here's how it works:
- Netris software runs on commodity network switches (think Nvidia Spectrum, Broadcom Trident)
- It abstracts the underlying hardware into a software-defined control plane
- Operators get a single interface to configure routing, load balancing, and security policies across thousands of switches
- Deploy time drops from months to weeks
The a16z bet isn't just on Netris. It's on the thesis that AI infrastructure will fragment further before it consolidates. Every AI lab building frontier models needs dedicated compute. Most will lease from neoclouds, not hyperscalers. Those neoclouds need to differentiate on speed and price, not on how well they can configure BGP routing tables.
The network layer is the last major bottleneck in AI infrastructure buildout. Chips are constrained but production is ramping. Power and cooling are solvable with money. Networking has been the silent tax: hard to staff for, hard to automate, hard to scale. Netris is betting that software can eat this part of the stack the same way it ate compute and storage.
The timing matters. We're entering the phase where second-tier AI companies (the ones training 10B-100B parameter models, not the frontier labs) need compute but can't get in line at AWS. Neoclouds serve that market. If Netris can standardize networking for this tier, they become infrastructure for infrastructure.
The Implication
Watch for neocloud proliferation to accelerate. If deployment friction drops, more operators will enter the market. That means more price competition for GPU clusters, which is good for anyone training models or running inference at scale. It also means the hyperscaler moat on AI workloads isn't as deep as it looked six months ago.
For builders: if you're planning AI infrastructure spend in the next 12 months, shop around. The neocloud tier is maturing fast, and capabilities are converging with the big three at a fraction of the cost.