Netris has just secured $15 million in a Series A round led by a16z. The startup builds software that runs directly on network switches, with a clear mission: slash the time it takes for a neocloud to go from blueprint to live service. In a market where every day of delay means missed opportunities to train or serve models, Netris's promise touches a raw nerve.

Anatomy of a neocloud

Neoclouds are not just data centers. They are cloud providers built from scratch for AI workloads: thousands of interconnected GPUs, ultra-fast storage, and a network that must behave as a single nervous system. When standing up a new cluster, network provisioning is often the bottleneck – configuring switches, VLANs, security policies, and routing still relies on manual steps or ad-hoc scripts. Netris steps in with a platform that automates the entire configuration, trimming go-live time from weeks to hours. Its software sits directly on the network switches, treating them as programmable resources.

Networking as the first pillar of on-premise AI

For those evaluating on-premise deployment of LLMs, networking is as critical as it is underrated. Training distributed models across dozens of GPUs demands ultra-low-latency, high-bandwidth connections (RDMA over converged Ethernet or InfiniBand). Without an optimized network configuration, GPUs idle waiting for data, undermining the hardware investment. Software like Netris's – which strips away manual complexity – aligns with the need to control every layer of the stack, typical of organizations pursuing data sovereignty and cost predictability. The promise of shrinking time-to-value resonates well beyond pure neoclouds: research labs, manufacturing firms, and financial institutions managing private clusters face the same choke points.

The a16z thermometer and the infrastructure market

There’s a reason a16z is backing Netris. The Silicon Valley venture firm is pouring capital into so-called "AI infrastructure" – the layer of tooling that makes deploying compute scalable and repeatable. After years of attention focused on models and GPUs, the market is waking up to the fact that orchestration and network control software is the real enabler for moving from proof-of-concept to production services. The investment signals that the AI battle isn’t fought on silicon alone, but on the ability to make distributed hardware collaborate efficiently – a lesson that holds for hyperscalers, neoclouds, and on-premise deployers alike.

Beyond the cloud: what changes for those who choose local

Netris joining the a16z constellation strengthens an ecosystem in rapid evolution, where network automation platforms become indispensable companions for anyone operating GPU clusters. For teams weighing self-hosted architectures, the ability to shave activation time without being locked into vendor-specific switch configurations means gains in TCO and autonomy. At AI-RADAR we offer analytical frameworks to compare cloud versus on-premise trade-offs: networking is one of the chapters that, as Netris’s story shows, can no longer stay in the background.