Upscale AI Raises $190 Million at a $2 Billion Valuation With Nvidia Joining, Betting Its Open UALink Switch Can Break NVLink's Grip
Nvidia joined a $190M round backing a startup building an open scale-up switch to rival its own NVLink fabric.
Editor's Note ·
- Correction:
- The article quotes Sandesh Patnam, managing partner at Premji Invest, as saying "The progress we've seen since our initial investment has only deepened our conviction," attributed to SiliconANGLE. That quote does not appear in the cited SiliconANGLE report or any other cited source. The SiliconANGLE report contains no Patnam quote; the only sourced Patnam statement is from Upscale AI's January Series A announcement and reads differently: "Their ability to deliver proprietary-grade performance while embracing open standards reflects both strong technical conviction and real-world operating insight." The quoted sentence attributed to Patnam in this article should be treated as unverified.
Overview
Upscale AI Inc., a data center networking startup, said on June 22 that it had raised $190 million in fresh funding at a $2 billion valuation, according to SiliconANGLE. The round was led by Premji Invest, and one of its new backers is Nvidia — the same company whose grip on AI cluster networking Upscale is trying to loosen. Nvidia joined alongside Salesforce Ventures, Seligman Ventures, Temasek and several existing backers, SiliconANGLE reported.
The money is an extension of a $200 million Series A that the company first closed in January, SiliconANGLE reported, bringing the Palo Alto company to roughly half a billion dollars in total funding raised in under 18 months.
What We Know
Upscale AI was incubated by Auradine, a blockchain and AI compute infrastructure company, and founded in 2024 in Palo Alto, California, according to Network World. It emerged from stealth with $100 million in seed funding co-led by Mayfield and Maverick Silicon, Network World reported. The January Series A, which totaled $200 million, was led by Tiger Global, Premji Invest and Xora Innovation, with participation from Intel Capital and Qualcomm Ventures among others, according to the company’s announcement.
The company is led by Chief Executive Officer Barun Kar and Executive Chairman Rajiv Khemani, per the Series A announcement.
At the center of the pitch is SkyHammer, a custom chip Upscale describes as a scale-up solution, according to the Series A announcement. “Scale-up” networking is the tightly coupled fabric that links accelerators inside a single rack or pod so they behave like one large processor — the domain Nvidia serves with its proprietary NVLink and NVSwitch silicon. The Register reported that Upscale aims to challenge Nvidia’s dominance of switches for rack-scale AI systems and to compete with the NVSwitch chips inside Nvidia’s NVL72 racks.
Rather than a closed design, SkyHammer is built around open standards. The Register reported that the chip uses a memory semantic-based load-store network architecture and includes acceleration for collective communication similar to Nvidia’s SHARP, and that it will support both UALink and the competing ESUN protocol. The company also supports the SONiC open-source network operating system, according to SiliconANGLE.
UALink is the open interconnect at the heart of Upscale’s bet. The UALink 200G 1.0 specification defines a low-latency, high-bandwidth interconnect between accelerators and switches, delivering a 200G-per-lane scale-up connection for up to 1,024 accelerators within an AI computing pod, according to the UALink Consortium. The consortium positions the standard explicitly as an alternative to Nvidia’s proprietary NVLink, and its membership includes AMD, Broadcom, Cisco, Google, HPE, Intel, Meta, Microsoft and Synopsys, Network World reported.
Upscale is not only building a scale-up switch. Its scale-out switches — the broader fabric that connects racks and pods across a data center — are based on Nvidia’s Spectrum-X chip series, according to SiliconANGLE, an arrangement that helps explain Nvidia’s willingness to invest in a would-be rival.
“AI infrastructure is being redefined at cluster scale, and networking is one of the most critical bottlenecks,” Kar said, according to SiliconANGLE. “The progress we’ve seen since our initial investment has only deepened our conviction,” said Sandesh Patnam, managing partner at Premji Invest, SiliconANGLE reported.
What We Don’t Know
Upscale’s hardware is being evaluated by multiple hyperscalers and neocloud operators, according to SiliconANGLE, but neither that report nor the company’s Series A announcement names a committed production customer. The company said its networking solutions are slated to ship in 2026, per the Series A announcement, though detailed SkyHammer specifications and pricing have not been disclosed.
Analysis
The round captures a tension running through the AI networking market. Nvidia’s NVLink and NVSwitch remain the default fabric for the largest GPU clusters, and the UALink consortium — backed by AMD, Intel, Meta, Microsoft, Google and others per Network World — exists precisely to give buyers an open alternative. That a well-funded startup can reach a $2 billion valuation targeting scale-up switching, while simultaneously licensing Nvidia’s Spectrum-X for its scale-out products and taking Nvidia’s investment, underscores how a single vendor can be both the incumbent to beat and a partner in the same buildout.