The AI boom has driven everyone to start a data center business with their uncle. However, starting a data center is not easy.
Even after you’ve solved the problem of securing your GPUs, network switches, and storage, you still need to configure everything to run and meet your customers’ varying needs. Preparing a data center to provide cloud computing services specialized for AI inference and training services can take several months. And the longer it takes to get to market, the higher the cost of keeping all those precious GPUs idle.
Network automation startup Netris claims Neocloud can solve that problem. In addition to providing software that runs on network switches, the company also provides a platform that connects to the switches to help neocloud operators reduce time to production by automating setup, configuration, and operations. The platform also provides network abstraction, so hardware configurations can be changed as needed, and servers and resources can be separated at the hardware layer, allowing Neocloud to serve multiple customers (multi-tenancy).
If that sounds like a solution to an obvious problem, it’s not. Until recently, data centers were primarily the domain of large infrastructure operators such as Equinix, NTT, Digital Realty, Oracle, Microsoft, AWS, and Google. These companies have largely taken care of network setup, configuration, and multi-tenancy by hiring engineers and building automation themselves. Small neo-cloud companies rarely have such resources at their disposal.
“As a GPU cluster operator, you have to change the configuration of all the links every day. In traditional data centers, we used something called SDN (software-defined networking) to do this, but SDN is a software technology, so it’s not good enough,” Netris CEO Alex Saroyan told TechCrunch. “For AI, it’s not just about software. The amount of traffic is so high that everything has to be hardware accelerated. So you need things like SDN, but fully hardware accelerated. That’s what we’re doing and what we’ve been doing for eight years.”
Saroyan said Netris’ platform is vendor-agnostic and compatible with networking equipment and standards used in both Nvidia and AMD server data centers.
The promise of this startup is believed by many, and one of them is Nvidia. Two years ago, the leading semiconductor manufacturing company was so impressed with a demonstration of Netris’ technology that it recommended the company to several of its customers. Netris currently runs on more than 35 GPU clusters around the world (approximately 1 million GPUs total), powered by Lightning AI, Foxconn, Visionbay, Hewlett Packard Enterprise, TensorWave, Telus, and more.
To build on its momentum, Netris has now raised $15 million in a Series A round from Andreessen Horowitz, TechCrunch has learned exclusively.
It’s worth noting that there’s no AI at work here. Saroyan said the company only uses previously developed algorithms to run and configure automation and operations.
“We started working on this long before AI came along. We understood this challenge early on and started developing this algorithm early on. AI isn’t deterministic, right? Sometimes it likes to do things on its own. It’s great for creative work, but when you’re changing thousands of switch configurations, you don’t need to be creative. You need to be very persistent and reproducible.”
a16z Partner Guido Appenzeller will join the firm’s Board of Directors. Going forward, Netris aims to use the funding to hire more engineers and sales staff, add support for more hardware vendors, and implement more features into its algorithms.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
