Upscale AI Hits $1 Billion Valuation With $200M Funding to Challenge Nvidia and Broadcom

Upscale AI Valuation Tops $1 Billion as Startup Takes on Cisco, Broadcom

Upscale will use the fresh funding to expand its engineering operation, which currently has 150 employees and is expected to double by the end of the year. Upscale AI Inc., a networking startup seeking to challenge Cisco Systems Inc. and Broadcom Inc., raised $200 million in a funding round that values the company at more than $1 billion. The round was led by Tiger Global Management, Premji Invest and Xora Innovation, Upscale said on Wednesday. Other investors included Maverick Silicon, StepStone Group, Mayfield, Prosperity7 Ventures, Intel Capital and Qualcomm Ventures. The fundraising comes four months after the Santa Clara, California-based company announced a $100 million seed round. Upscale aims to develop networking equipment that can better cope with the onslaught of artificial intelligence data. Cloud-computing systems that train and run AI programs currently struggle with slowness and data loss because of network congestion. Though Cisco, Broadcom and Arista Networks Inc. are tackling the same problem, Upscale looks to gain an edge by building a new networking system designed for AI -- rather than adapting older technology. "AI networking is one of the most urgent and unaddressed bottlenecks in the system," Barun Kar, Upscale's chief executive officer, said in an interview. "A faster chip cannot just deliver breakthroughs if the network can't keep up." The company is already talking to partners and potential customers, including data center giants, specialized cloud-computing providers and so-called sovereign AI projects backed by governments. It plans to have a product available later this year, said Rajiv Khemani, the company's executive chairman. The round had more interested investors than Upscale could accommodate, said Navin Chaddha, managing partner at venture firm Mayfield, which co-led the earlier round. It happened so quickly after the previous fundraising because investors reached out to the company, he said. Mayfield looks for companies that solve the biggest pain points for customers, he said. "You can solve a problem, but if it's not the No. 1 problem for me, you're wasting my time," Chaddha said. Khemani co-founded and ran Innovium, which was bought by Marvell Technology Inc., and served on the leadership team at Cavium, another Marvell acquisition. Kar was an early employee at Palo Alto Networks Inc. and served as senior vice president of engineering. Upscale was spun out of Auradine, which was founded in 2022 by Kar and Khemani and focused on AI infrastructure. It raised more than $300 million from some of the same investors, including Mayfield. Upscale will use the fresh funding to expand its engineering operation. It has 150 employees and will probably double that number by the end of the year, Kar said. The company needs to use the funds to get a working product out and win over clients, Mayfield's Chaddha said. He's known Kar and Khameni for about 15 years and has confidence they can turn Upscale into a dominant player. "It's a real market," he said, adding that the key part will be gaining enough share against rivals like Broadcom.

Upscale AI raises $200M to challenge Nvidia's NVSwitch

Plans to swing SkyHammer silicon into UALink switches later this year AI networking startup Upscale AI on Wednesday announced it has raised $200 million in Series A funding to challenge Nvidia's dominance of switches for rack-scale AI systems, putting it in competition with the likes of Cisco and AMD. Founded last year, with backing from the likes of Intel, AMD, and Qualcomm, Upscale is developing a new class of scale-up network switches specifically designed to compete with the NVSwitch chips found in Nvidia's NVL72 racks. NVLink is a high-speed interconnect technology that Nvidia developed to abstract memory and compute resources from multiple GPUs so they appear as a single logical resource. The tech debuted in 2024, and ever since, the likes of AMD and Cisco have tried to create alternatives. But their efforts so far, like Ultra Accelerator Link (UALink) and Ethernet for Scale-Up AI Infrastructure (ESUN) remain immature. The first UALink-based rack systems from AMD will hit the market later this year, but they'll be tunneling the protocol over Ethernet. Purpose-built UALink switches capable of competing with Nvidia's NVSwitch simply aren't ready yet. Upscale aims to change that with its SkyHammer custom ASICS. "Instead of retrofitting legacy systems, what we are re-imagining is what scale truly means in AI networking," Barun Kar, Upscale AI CEO told El Reg. "The heart of this architecture is essentially to scale up. It's meant for AI workloads and nothing else." The startup isn't ready to reveal technical details about SkyHammer, but has said it will offer its first silicon as a standalone ASIC for hyperscale integration, and as an integrated switch blade and rack. While we don't have enough information about the chip to draw any comparisons to NVSwitch 6 or Broadcom's Tomahawk 6, Kar tells us it is using a memory semantic-based load-store network architecture, and will feature acceleration for collective communication similar to Nvidia's Sharp. The platform will also support both UALink and the competing ESUN protocol. To make the whole thing manageable at scale, Upscale is working to extend support for the SONiC network operating system (NOS). Originally developed by Microsoft, the open source NOS is both widely deployed and well understood by hyperscale customers. For the moment, Upscale is primarily focused on scale-up networking products, but long-term it plans to expand its offering to more traditional scale-out switches. For this, Kar tells us, the company is still evaluating its options and may lean on third-party IP from partners. With an additional $200 million from Tiger Global, Premji Invest, and Xora Innovation's coffers in hand Upscale is now looking to expand its engineering, sales, and operations teams in anticipation of shipping its first AI networking productions later this year. "We are already partnering with hyperscalers and GPU vendors who have validated the architecture. That is complete. Now the focus with this funding is to turn innovation into deployment," Kar said. ®

Upscale AI raises $200M to develop scale-up AI networking chips

Upscale AI Inc., a developer of networking chips for artificial intelligence clusters, has raised $200 million in funding. The startup disclosed in its announcement of the round today that Tiger Global, Premji Invest and Xora Innovation were the lead investors. They were joined by the venture capital arms of Intel Corp. and Qualcomm Technologies Inc. along with several others. The investment brings Upscale AI's outside funding to more than $300 million. An AI cluster comprises multiple racks that can each contain several dozen servers. Those servers exchange data with one another via a switch built into the host rack. A rack switch's technical attributes often differ significantly from those of other networking devices, such as those used to connect different racks with one another. Upscale AI is developing a chip optimized for scale-up networking, the task of linking together the hardware components inside a rack. The processor, which is called SkyHammer, will offer deterministic latency. That means the amount of time a piece of data requires to travel between one rack component to one another can be predicted with a high degree of certainty. AI models process data using calculations that must be carried out in a specific sequence. As a result, a delay in one calculation often requires all the subsequent processing steps to be postponed. The ability to predict network latency ahead of time avoids unexpected data transmission delays that might slow down AI workloads. Upscale AI says SkyHammer will also generate real-time telemetry. Telemetry, or technical data about a system, is needed for not only troubleshooting but also configuration tasks. Administrators can analyze telemetry about the state of a network device to find ways of optimizing its performance. SkyHammer is compatible with multiple open-source networking technologies. The list includes UALink and ESUN, two projects that seek to harness Ethernet for scale-up networking use cases. ESUN, the newer of the two initiatives, launched last year with backing from Nvidia Corp., Broadcom Inc. and other major industry players. SkyHammer will also support a networking technology called UEC. Whereas ESUN is designed to link together the components inside a rack, UEC is geared toward connecting different racks with one another. It can power AI clusters with up to 1 million chips. "As AI systems scale, interconnect efficiency has become a defining driver of performance and economics -- not just raw compute," said Premji Invest managing partner Sandesh Patnam. Upscale will use its newly raised funding to enhance its technology and hire more go-to-market professionals. The company plans to launch its first SkyHammer-products later this year. Those products reportedly include a high-radix switch that can power a large number of connections.