How New Networking Tech Is Powering AI GPU Clouds

Two startups are shaking up how AI clouds handle networking. Netris and Upscale AI both raised big funding rounds this year. Their goal is to speed up and streamline the complex networking behind GPU-heavy AI systems.

Netris, based in Santa Clara, raised $15 million in a Series A round led by Andreessen Horowitz. The company launched in 2018 and focuses on automating GPU cloud networking. Their platform cuts weeks off the time it takes new cloud operators to go live. That’s a huge boost for companies racing to deploy AI infrastructure fast.

Alex Saroyan, Netris’s CEO, says their system uses hardware acceleration because AI workloads generate massive traffic. “For AI, software is not okay,” he says. “Everything must be hardware accelerated.” Their approach skips AI itself in the product. Instead, it relies on deterministic algorithms developed before the AI boom.

Nvidia took notice early. They saw a demo from Netris two years ago and began recommending the startup to their customers. Today, Netris’s platform runs at over 35 GPU clusters worldwide. These clusters have roughly one million GPUs in total. Netris grew its annual recurring revenue by 800 percent last year. Their customers include Lightning AI, Visionbay, Hewlett Packard Enterprise, TensorWave, and Telus.

Andreessen Horowitz partner Guido Appenzeller led the funding round. He joined Netris’s board and praised the startup’s role in AI networking. He said, “GPU clusters run across many fabrics at once, and legacy automation was never built for that.” Appenzeller added, “Netris is the platform AI cloud operators standardize on to solve it.”

Upscale AI’s Big Bet on AI-Only Networking Hardware

Upscale AI raised $190 million in a Series A-1 round this year. That brought their total funding to $500 million and pushed their valuation to $2 billion. Premji Invest led the round, joined by Nvidia, Salesforce Ventures, Temasek, and Seligman Ventures.

Founded in 2025 and spun out of Auradine (formerly Velaura AI), Upscale AI builds custom chips designed purely for AI workloads. Their SkyHammer switch chip line is an AI-only networking solution. The founders, Barun Kar and Rajiv Khemani, come from backgrounds in high-performance, power-efficient chip design.

Kar, Upscale AI’s CEO, explained the problem: “AI infrastructure is being redefined at cluster scale, and networking is a critical bottleneck.” Their approach uses open standards like ESUN and SONiC to build AI fabrics from the ground up. This avoids legacy networking baggage and aims for better performance across AI clusters.

Nvidia’s $190 million investment highlights their strategic interest in AI networking hardware. Deepti Chandra, Upscale AI’s VP for product and marketing, says the funding will help scale the business and accelerate technology delivery. Khemani also noted the investment supports both immediate goals and long-term ambitions.

Why AI Networking Demands New Solutions

Every era of computing needs a new networking foundation. First came virtualized data centers. Then the cloud. Now AI requires its own network design. That’s according to Guido Appenzeller, who helped lead the Netris funding.

He points out that AI workloads generate huge traffic across many fabrics simultaneously. Legacy networking automation wasn’t built to handle such complexity. That’s why startups like Netris and Upscale AI are building specialized platforms and hardware.

For AI cloud operators, time to market matters. Netris’s automated platform slashes setup time by weeks. At the same time, Upscale AI’s chips promise better hardware acceleration tuned for AI traffic. Together, these advances could reshape how GPU clouds power AI services.

As AI models grow larger and more demanding, networking will remain a key bottleneck. The startups leading the charge are backed by major players and investors. Their work could soon become the new standard for AI cloud infrastructure worldwide.