Solving the Infrastructure Bottleneck for Neoclouds

Building a data center for AI training and inference is a complex, time-intensive process. Beyond the challenge of sourcing GPUs and storage, operators must configure network switches and manage multi-tenancy—tasks traditionally handled by large-scale providers like AWS or Google through massive internal engineering teams. Smaller 'neocloud' operators often lack these resources, leading to idle GPU clusters and delayed time-to-market.

Netris addresses this by providing a platform that automates the setup, configuration, and operation of network switches. By abstracting the network layer, the platform allows operators to manage hardware configurations dynamically and isolate resources for multiple customers without requiring manual intervention for every link change.

Hardware-Accelerated Automation vs. Software-Defined Networking

Netris distinguishes its approach from traditional software-defined networking (SDN). While SDN is a software-based technology, the high traffic demands of AI workloads require hardware-accelerated performance. Netris focuses on deterministic, repeatable algorithms that operate at the hardware layer, ensuring that the massive data throughput required for AI training is not bottlenecked by software overhead.

Notably, the company rejects the use of generative AI for these operations. CEO Alex Saroyan emphasizes that network configuration requires consistency and predictability rather than the creativity associated with LLMs. By relying on established, deterministic algorithms, Netris ensures the reliability required for managing thousands of switch configurations across large-scale GPU clusters.

Market Adoption and Scaling

The platform is vendor-agnostic, supporting networking equipment compatible with both Nvidia and AMD server ecosystems. This flexibility has led to adoption across 35 global GPU clusters, totaling approximately one million GPUs. Current clients include Lightning AI, Foxconn, and Hewlett Packard Enterprise. With $15 million in Series A funding from Andreessen Horowitz, the company plans to expand its engineering and sales teams, broaden hardware vendor support, and enhance its core automation algorithms.