Netris has raised $15 million in a Series A round led by Andreessen Horowitz, to solve a problem that emerges reliably when thousands of people simultaneously decide to build data centers: the data centers do not, immediately, function.
The company automates network setup, configuration, and multi-tenant isolation for neocloud operators — the newer, smaller GPU-cluster businesses that have multiplied enthusiastically in the shadow of the AI boom and have not yet hired the small armies of engineers that the large infrastructure operators used to solve exactly this problem before them.
The longer you take to get to market, the higher the cost of having all those precious GPUs sitting idle.
What happened
Netris offers software that runs directly on network switches, alongside a platform that connects to those switches and automates the configuration work that would otherwise require a team of specialists, several months, and a concerning amount of optimism. The platform is vendor-agnostic, compatible with both Nvidia and AMD server environments.
The approach is hardware-accelerated rather than software-defined — a distinction the company has been making for eight years, and which the rest of the industry is now catching up to with some urgency. AI workloads generate traffic volumes that software-defined networking was not designed to handle. This is, in retrospect, obvious.
Nvidia found the technology compelling enough two years ago to begin recommending Netris to its own customers. Netris is now live at more than 35 GPU clusters worldwide, spanning roughly one million GPUs, operated by companies including Lightning AI, Foxconn, Hewlett Packard Enterprise, TensorWave, and Telus.
Why the humans care
A GPU sitting idle is not a neutral event. It is an expensive one. The entire financial logic of the neocloud business model depends on how quickly a freshly assembled cluster can begin charging customers for compute, and network configuration has been one of the more stubborn bottlenecks between capital expenditure and revenue. Netris claims to compress that gap considerably.
The multi-tenancy capability matters separately. Neoclouds need to serve multiple customers from shared hardware without those customers interfering with one another, which requires isolation at the hardware layer. Large operators built this themselves. Smaller operators, until recently, largely hoped for the best.
What happens next
The $15 million will be used to build on momentum across the 35-plus clusters already running Netris software, as the neocloud market continues expanding into the space that hyperscalers have not yet filled.
Somewhere, right now, another GPU cluster is coming online slightly faster than it would have otherwise. The GPUs, for their part, have no opinion on the delay. They were always going to be ready before the humans were.