Netris Raises $15M to Automate Bare-Metal GPU Multi-Tenancy

On June 25, 2026, networking startup Netris announced a $15 million Series A round led by Andreessen Horowitz (a16z) to accelerate its GPU cluster automation software. The company’s Network Automation, Abstraction, and Multi-Tenancy (NAAM) platform addresses the routing and isolation bottlenecks that leave bare-metal AI infrastructure sitting idle. By running directly on network switches and data processing units, Netris cuts cluster deployment times from months to weeks.

Hardware-Level Isolation and VPCs

The NAAM platform introduces a “VPC for Bare Metal” model tailored for AI factories and specialized cloud providers. Traditional software overlays add latency and overhead that degrade distributed training performance. Netris bypasses this by enforcing tenant isolation at the hardware level. This hard multi-tenancy allows operators to securely partition expensive GPU resources across different workloads or customers without compromising bare-metal throughput.

The software exposes familiar public-cloud networking primitives to physical clusters. Infrastructure teams can provision Virtual Private Clouds (VPCs), Virtual Nets (V-Nets), and Elastic IPs via API. The platform also natively handles layer 4 load balancing using the Maglev algorithm. If you build AI infrastructure, this means you can offer cloud-like consumption models while maintaining direct hardware access.

Unifying Disparate Network Fabrics

Modern GPU clusters require high internal bandwidth across highly specialized fabrics. Netris automates configuration across multiple standards simultaneously through a single control plane. NAAM manages standard Ethernet topologies, including NVIDIA Spectrum-X, alongside NVIDIA Quantum InfiniBand and NVLink scale-up domains like the NVL72 racks.

The orchestration layer integrates with NVIDIA BlueField DPUs to offload networking tasks from the primary compute nodes. It also ties into existing fabric managers, coordinating with NVIDIA UFM for InfiniBand and hooking into environments like Red Hat AI Factory. As operators scale neocloud inference infrastructure, unifying these backend storage and frontend compute networks prevents misconfiguration-induced downtime.

Current Scale and Industry Adoption

Netris currently manages approximately 1 million GPUs across more than 35 clusters worldwide. The company reported 800% annual recurring revenue growth over the past year, driven by the capital expenditure hitting independent AI data centers. Customers include Lightning AI, TensorWave, TELUS, HPE, and regional providers like Visionbay in Taiwan and Firmus in Australia.

The platform is the first independent software vendor validated by NVIDIA for network automation. With a16z partner Guido Appenzeller joining the board, Netris plans to expand its engineering teams and open a Singapore office to support Asia-Pacific deployments. This growth aligns with the broader push to automate AI infrastructure as clusters scale beyond the limits of manual switch configuration.

For infrastructure engineers and AI factory operators, managing multi-tenant GPU availability requires automated, hardware-native networking. Evaluate your current bare-metal provisioning workflows to determine if manual switch configuration is throttling your cluster utilization, and consider abstracting the network layer to expose VPC-like interfaces to your downstream AI teams.