Upscale AI raises $190 million to build AI data center networking tech

Upscale AI has raised $190 million in new funding, giving the Santa Clara, California, startup more capital to build networking technology for AI data centers. The round matters because the company is trying to loosen Nvidia’s grip on the systems that connect chips used to train and run advanced AI models.

The Series A-1 financing brings Upscale’s total funding to $500 million and values the company at $2 billion, Fortune reported. Premji Invest led the round, with Nvidia, Salesforce Ventures, Temasek and Seligman Ventures joining as new investors.

Mayfield, Tiger Global, StepStone, Maverick Silicon and Prosperity7 also returned as backers, according to Fortune. The company reached the $500 million funding mark in less than 18 months.

A bet on open AI networking

Upscale is building what it calls an open-standard networking fabric for AI infrastructure. The goal, according to cofounder and CEO Barun Kar, is to help graphics processing units from different makers communicate more effectively.

Today, GPUs used for AI workloads often work best when paired with networking technology from the same supplier. Nvidia’s NVLink and InfiniBand products dominate that part of AI data centers, according to Fortune, and keep many customers tied closely to Nvidia’s hardware and software system.

Rajiv Khemani, Upscale’s cofounder and executive chairman, told Fortune that he expects future AI systems to use a mix of technologies rather than one vendor’s products alone. He said Upscale’s approach is meant to let different systems coexist.

Umesh Padval, managing partner at Seligman Ventures, told Fortune that older data center networks were not designed for the tightly coordinated scale required by modern AI workloads. Sandesh Patnam, managing partner at Premji Invest, told Fortune that networking, storage and caching were among the infrastructure layers not originally built for generative AI.

Demand is rising, but costs are high

Dell’Oro Group forecasts that annual spending on AI data center switches will top $100 billion by 2030. The Futurum Group estimates that the five largest technology companies will spend about $660 billion to $690 billion on infrastructure in 2026, nearly twice their 2025 spending, as Microsoft, Google, Meta and Amazon build out AI capacity.

Upscale’s funding also reflects the cost of competing in advanced chip and networking hardware. Fortune reported that custom chip development at leading-edge manufacturing nodes can cost hundreds of millions of dollars before a product reaches customers.

Suppliers are also asking companies to reserve and pay for manufacturing capacity as much as two years ahead of time, according to Fortune. Khemani told Fortune that suppliers want to see the strength of Upscale’s balance sheet before committing capacity.

The company has grown to several hundred employees in under a year, according to a company press release cited by Fortune.

Experienced founders, strong rivals

Investors are also backing the founders’ track records. Khemani previously led Innovium, which Marvell acquired for $1.1 billion in 2021, and earlier ran Intel’s network processing business and served as chief operating officer at Cavium, according to Fortune.

Kar was a founding team member at Palo Alto Networks and later oversaw Juniper Networks’ Ethernet portfolio, Fortune reported. Upscale was incubated at Auradine, a blockchain and AI chip startup where Kar and Khemani had worked together.

The company faces rivals with deep pockets. Fortune reported that Nexthop AI, which targets similar hyperscaler and neocloud customers, raised $500 million in March at a $4.2 billion valuation. Nvidia and Broadcom remain entrenched competitors, and the open standard Upscale is using still has to show it can match Nvidia’s proprietary systems in real AI training environments.

Khemani told Fortune that an initial public offering could be a milestone for Upscale, while saying the company’s ambitions go beyond that.

This story draws on original reporting from Fortune.