The Santa Clara, Calif., startup raised $190 million in a Series A-1 round, bringing its total funding to $500 million and its valuation to $2 billion—all in under 18 months, Fortune learned exclusively. Premji Invest, a $15 billion investment fund that previously backed one of the founders’ earlier exits, led the round. New investors include Nvidia, Salesforce Ventures, Temasek, and Seligman Ventures, alongside returning backers Mayfield, Tiger Global, StepStone, Maverick Silicon, and Prosperity7.

The opportunity Upscale is chasing is enormous. Spending on AI data center switches is forecasted to surpass $100 billion annually by 2030, according to Dell’Oro Group, as Microsoft, Google, Meta, and Amazon race to build out AI infrastructure. The five largest tech companies will spend roughly $660 billion to $690 billion on infrastructure in 2026 alone, nearly doubling what they spent in 2025.

“Legacy data center networks were designed for a pre-AI world, not for the massive, tightly synchronized scale-up required by modern AI workloads,” Umesh Padval, managing partner at Seligman Ventures, told Fortune.

Upscale is trying to solve an AI chip communication issue. Right now, the graphics processing units that power tools like ChatGPT mostly can’t talk to each other efficiently unless they’re all from the same manufacturer. It’s like if your laptop could only connect to the internet if every device in your house was also made by the same company. Upscale is building what it calls an open-standard networking fabric that lets chips from different manufacturers work together at full speed.

“We are trying to get those GPUs to talk the same language,” cofounder and CEO Barun Kar told Fortune.

Nvidia’s proprietary NVLink and InfiniBand technologies currently dominate AI data center networking and lock customers into Nvidia’s ecosystem. But Upscale cofounder and Executive Chairman Rajiv Khemani doesn’t see a contradiction.

“The world in the future is going to be a variety of heterogeneous AI systems,” he told Fortune. “It’s not an Nvidia or our solution. It’s a solution where everything coexists with each other.”

Lead investor Sandesh Patnam, managing partner at Premji Invest, sees the whitespace clearly. “If you look at everything that has been done over the last three or four years—if the compute layer itself was never built for GenAI workloads, look at everything else in that stack,” he told Fortune. “Networking, storage, caching—every single layer of the infrastructure was never built for this workload.” Premji Invest previously backed Khemani’s prior company, Innovium, through its late stages, then watched Marvell acquire it for $1.1 billion in 2021. 

Khemani is the serial entrepreneur anchoring investor conviction. Before Innovium, which was the only startup in a decade to meaningfully crack Broadcom’s lock on networking chip market share, he ran Intel’s network processing business and served as COO at Cavium.

Kar, the company’s CEO, was a founding team member at Palo Alto Networks and oversaw Juniper Networks’ entire Ethernet portfolio before the two joined forces at Auradine, a blockchain and AI chip startup where Upscale was incubated. 

Upscale’s $500 million in funding, however, won’t last forever in this business. Custom chip design at today’s most advanced manufacturing nodes costs hundreds of millions of dollars before a product ships. In addition, suppliers increasingly want companies to commit to, and pay for, production slots up to two years in advance. The company has also scaled to several hundred employees in under a year.

“To be able to go to our suppliers and lock in that supply, people are interested in seeing how strong our balance sheet is,” Khemani said.

And competitive pressure is mounting. Nexthop AI, a direct rival focused on custom AI networking for the same hyperscaler and neocloud customers, closed a $500 million Series B round in March at a $4.2 billion valuation. Broadcom and Nvidia have effectively unlimited R&D budgets. And the open networking standard Upscale is building on—backed by AMD, Intel, Google, Meta, Microsoft, and 80+ other companies—still needs to prove it can match Nvidia’s proprietary performance in real-world AI training.

Asked whether this is the run to an IPO, Khemani didn’t deflect. “An IPO would be an interesting milestone,” he said. “But we think it’s going to be much bigger than that.”