Upscale AI is a networking company building switches, both the chips and the systems and software around them, purpose-made to connect the graphics processors inside artificial intelligence data centers. Founded in 2025, it moved quickly to a $2 billion valuation, and in 2026 it raised $190 million with Nvidia joining the round as a strategic investor.
What Upscale AI builds
Training and running large models no longer happens on one machine. It happens across racks of thousands of GPUs that must behave as a single computer, trading data constantly, and increasingly the limiting factor is not the processors but the network stitching them together. Upscale builds both the scale-up switches that move data inside a rack and the scale-out switches that handle traffic between racks, tuned for AI workloads rather than general enterprise traffic.
The technical pitch centers on predictability. Its scale-up line, running on a custom chip the company calls SkyHammer, aims for what it calls deterministic latency so data-movement speeds can be known in advance and stalls avoided. The hardware supports open protocols such as UALink and an AI-tuned flavor of Ethernet, while its scale-out line builds on Nvidia's Spectrum-X silicon paired with an AI-tuned version of the open-source SONiC operating system.
Funding, founders, and competition
The financing, an extension of a $200 million Series A closed in January 2026, was led by Premji Invest and pushed total funding to roughly $500 million in under eighteen months. Nvidia, whose chips Upscale's switches are built to serve, joined as a strategic investor, deepening an existing partnership. The company was founded by chief executive Barun Kar and Rajiv Khemani, spun out of the Bitcoin-mining hardware firm Auradine, and has grown to around 200 employees.
The competitive backdrop is steep. Upscale faces Nvidia's own switches along with Broadcom, Arista Networks, and the incumbent enterprise leader Cisco. Its wager is that a clean-sheet design built around open standards, rather than a single vendor's proprietary stack, can win share as buyers look to avoid lock-in. Spending on AI data center switches is forecast to top $100 billion a year by 2030.
Why it matters
The first phase of the AI buildout rewarded whoever sold compute. The next phase is rewarding the less glamorous layers around it, the networking, memory, power, and cooling that decide whether all that compute runs efficiently. Picks-and-shovels businesses adjacent to a boom often outlast the boom itself, and investors are pricing that in.
There is also a lesson in speed and pedigree. Upscale reached a $2 billion valuation in roughly a year, raising half a billion dollars before shipping a commercial product, a pace made possible because its founders had built and sold networking silicon before and because a strategic backer can shorten the path to customer trials.
Frequently asked questions
What does Upscale AI make?
It designs networking switches, including the chips, systems, and software, that connect GPUs inside AI data centers so data moves between processors quickly and predictably.
How much has Upscale AI raised?
Its 2026 round added $190 million at a $2 billion valuation, bringing total funding to about $500 million in under eighteen months. The round extended a $200 million Series A closed in January 2026.
Why did Nvidia invest in Upscale AI?
Upscale's scale-out switches are built on Nvidia's Spectrum-X silicon and the company was already in Nvidia's partner network. Backing Upscale extends Nvidia's reach into the networking layer around its GPUs.
Who founded Upscale AI?
It was founded in 2025 by chief executive Barun Kar and Rajiv Khemani, spun out of the hardware firm Auradine. Both founders have deep backgrounds in networking and chip companies.
Who does Upscale AI compete with?
It competes with Nvidia's own switches as well as Broadcom, Arista Networks, and Cisco, betting that an AI-only design built on open standards can win buyers wary of vendor lock-in.