Upscale AI raises $190 million as Nvidia joins the round. Upscale AI, a year-old startup building networking chips purpose-made for artificial intelligence data centers, said Monday it has raised $190 million in fresh capital and is now valued at $2 billion. The round drew a notable new backer in Nvidia, the dominant supplier of the graphics processors that Upscale's hardware is designed to connect. The financing arrived as an extension to a $200 million Series A the company closed in January, and it pushes Upscale's total funding to roughly $500 million in under 18 months. Premji Invest, the India-based fund that has previously backed CrowdStrike and Sysdig, led the round. Salesforce Ventures, Seligman Ventures and Temasek joined as new investors, alongside returning backers including Mayfield, Tiger Global and Prosperity7 Ventures. Nvidia, whose chips Upscale's switches are built to serve, came in as a strategic investor. Why networking became the bottleneck. The story behind the raise is a structural shift in how AI gets built. Training and running large models no longer happens on a single machine. It happens across racks of thousands of GPUs that must behave as one giant computer, trading data constantly. When the chips wait on each other, expensive silicon sits idle. Increasingly, the limiting factor in an AI cluster is not the processors themselves but the network stitching them together. That is the gap Upscale is chasing. A data center relies on two broad classes of switch. Scale-up switches move data between servers packed inside the same rack. Scale-out switches handle traffic between racks. Upscale is building both, tuned specifically for AI workloads rather than the general-purpose enterprise traffic that older switch designs were optimized for. The technical pitch centers on predictability. The math behind a single AI response runs in sequence, so one delayed calculation stalls everything downstream. Those delays often trace back to GPUs swapping data more slowly than expected. Upscale says its scale-up line delivers what it calls deterministic latency, meaning data-movement speeds can be known in advance and unexpected stalls avoided. The capability runs on a custom chip the company calls SkyHammer, which supports open protocols built for this kind of traffic, including UALink, a standard that lets GPUs read one another's memory as if it were local, and ESUN, an AI-tuned flavor of Ethernet. On the scale-out side, Upscale is leaning directly on Nvidia. In March it previewed a switch line built around Nvidia's Spectrum-X silicon, paired with SONiC, an open-source network operating system Microsoft originally created for its own cloud. Upscale has built an AI-tuned version of SONiC meant to make the notoriously fiddly software easier to run. The Nvidia investment, then, is less a surprise than a deepening of an existing relationship; Upscale was already a member of Nvidia's partner network. "AI infrastructure is being redefined at cluster scale, and networking is one of the most critical bottlenecks," Chief Executive Barun Kar said in announcing the round. The company says its hardware is being evaluated by several hyperscalers and so-called neocloud operators, the new wave of GPU-rental providers, and that the new money will go toward turning those trials into commercial deployments. A crowded, fast-moving market. Upscale is a newcomer in a category suddenly thick with money. The startup was founded only in 2025 by Kar and Rajiv Khemani, spun out of Bitcoin-mining hardware firm Auradine, and has already grown to around 200 employees. Its founders carry deep networking pedigrees: Kar previously worked at Palo Alto Networks, Juniper and Motorola, while Khemani held senior roles at chipmakers Innovium and Cavium and earlier at Intel. The competitive backdrop is steep. Nvidia sells its own Spectrum-X switches. Broadcom and Arista Networks dominate the merchant switching market. Cisco, the company Upscale's investors openly compare it to, remains the incumbent giant of enterprise networking. Upscale's wager is that a clean-sheet design free of legacy baggage, 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, according to Dell'Oro Group, a number large enough to support several winners. "The progress we've seen since our initial investment has only deepened our conviction," said Sandesh Patnam, managing partner at Premji Invest, in explaining why the firm led the extension. What it means for founders and operators. For founders watching the AI infrastructure boom, Upscale is a useful case study in where value is migrating. The first phase of the 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 actually runs efficiently. Picks-and-shovels businesses adjacent to a gold rush often outlast the rush itself, and investors are clearly pricing that in. There is also a lesson in speed and pedigree. Upscale went from founding to a $2 billion valuation in roughly a year, raising half a billion dollars before shipping a commercial product. That pace is possible because the founders had built and sold networking silicon before, and because a strategic backer like Nvidia can shorten the path from prototype to customer trials. Domain credibility and the right partner on the cap table remain among the few things that let a hardware startup move at software speed. The risk worth naming: valuations in AI infrastructure are running well ahead of revenue, and a switch that is still in customer evaluation has not yet proven it can win against entrenched giants. For operators evaluating vendors, the open-standards angle is the part to scrutinize closely, since interoperability is exactly what reduces the lock-in that makes infrastructure bets expensive to reverse. Frequently asked questions. What does Upscale AI actually make? How much has Upscale raised and what is it worth? Why did Nvidia invest in a networking startup? Who does Upscale compete with? Why does AI networking matter so much right now?

Upscale AI readies Skyhammer scale-up networking tech, raises new funding. Jun 23, 2026 4 mins Upscale AI raised $190M in new funding to support AI networking scale up and scale out architecture. Skyhammer is its switch silicon designed to connect GPUs and XPUs at very high speed. AI continues to fuel demand for new networking technology and the vendors that build it. Upscale AI, which emerged from stealth in September 2025 with a $100 million seed round, has raised an additional $190 million in a Series A-1 extension. Total funding now stands at $500 million, and its current valuation is $2 billion. The round was led by Premji Invest and includes new investors Nvidia, Salesforce Ventures, Seligman Ventures, and Temasek. Existing backers - Maverick Silicon, Mayfield, Prosperity7 Ventures, StepStone Group, and Tiger Global - also participated. The Santa Clara-based company describes itself as a pure-play AI networking company focused on the backend networks that connect GPUs and XPUs for hyperscalers and neoclouds. Upscale builds AI networking hardware across two tracks. Scale-up networking connects GPUs and XPUs within a cluster. Scale-out networking extends connectivity across nodes. Since the company emerged from stealth it has expanded its efforts, inking a partnership with Nvidia on SpectrumX Ethernet. Upscale has also announced that it is developing its own custom scale-up switch ASIC, code-named Skyhammer. "The world that we see is going to be a heterogeneous AI compute, and we want to be a player in that heterogeneous AI compute by providing a network that supports Nvidia-based infrastructure as well as... other GPU and XPU vendors," Rajiv Khemani, co-founder and executive chairman of Upscale AI, told Network World. Skyhammer: Upscale's purpose-built scale-up silicon. Skyhammer is Upscale AI's flagship effort on the scale-up side of AI networking: a purpose-built switch silicon designed to connect GPUs and XPUs at very high speed inside AI systems. Khemani said that unlike commodity data center chips repurposed for AI, Skyhammer is being developed specifically for AI scale-up use cases and is tightly coupled to Upscale's broader full-stack strategy, which spans silicon, systems and software. Khemani declined to share detailed timelines, but he said Upscale expects to reveal product details on Skyhammer later this year, with actual deployment synced to when GPU and XPU vendors are ready. "The Skyhammer product doesn't work by itself," he explained. "It works in conjunction with XPUs and GPUs, and so for us to be deployed, the XPUs and GPUs need to incorporate scale-up capabilities to interoperate with us." Nvidia, Spectrum X, and strategic capital. Nvidia sits at the center of Upscale AI's story, both as a technology partner and now as a strategic investor. On the product side, Upscale has aligned its scale-out roadmap with Nvidia's Spectrum X Ethernet platform to build AI-optimized network fabrics that connect GPU and XPU nodes across the data center. The Spectrum X-based systems are expected to reach the market later this year, with Khemani noting that products are already in customer labs and are booked for early deployments. Network World smart answers. Explore related questions. That technical collaboration expanded into a capital relationship, with Nvidia joining Upscale's latest funding round. Software, token efficiency and what comes next. The goal for many organizations today is to optimize AI token usage. That's a topic that is often referred to as 'tokenmaxxing,' and it's an area where Khemani noted that networking has a role to play. Upscale is building a full-stack offering that includes software for both large hyperscalers and emerging neocloud providers. Khemani emphasized that there is not a single right answer to the challenge of token efficiency. He sees it as workload-dependent, and changing as the industry moves from simple prompts to agents onto more complex loops and multi-step applications. "I don't think we have a choice other than to make sure that AI infrastructure is used in as optimal a fashion as it can," Khemani said. "So, you have to build an infrastructure that is somewhat flexible to adapt to whatever AI applications require." From our editors straight to your inbox. Get started by entering your email address below. Contributing Writer Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer, and has been known to spend his spare time immersed in the study of the Klingon language and satellite pictures of Area 51. He has pulled Token Ring, configured NetWare and has been known to compile his own Linux kernel. He consults to industry and media organizations on technology issues.