Upscale AI raises $190M from Nvidia, Salesforce Ventures for AI networking infrastructure, hits $2B valuation

Upscale AI, a startup building networking infrastructure for large AI clusters, has raised $190 million in fresh funding from backers including Nvidia and Salesforce Ventures, pushing its valuation to $2 billion as investors pour money into one of the least glamorous but most important layers of the AI stack.

The new financing is an extension of Upscale AI’s Series A, bringing the Santa Clara-based company’s total funding to $500 million. Premji Invest led the round. Temasek and Seligman Ventures joined as new investors, with existing backers including Maverick Silicon, Mayfield, Prosperity7 Ventures, StepStone Group, and Tiger Global returning for more.

The pitch behind Upscale is simple enough: AI systems have gotten far better at generating text, images, and code, but the infrastructure underneath them is still full of bottlenecks. Training frontier models and serving large inference workloads require large numbers of chips, memory, and storage systems to work in sync. The network connecting those machines has become one of the biggest pressure points in the modern AI data center.

That is the opening Upscale wants to exploit.

The company is building what it describes as a full-stack AI networking platform spanning silicon, systems, and software. Its broader goal is to connect accelerators, memory, and storage via an open-standard fabric designed for large-scale AI workloads, rather than relying on networking architectures originally built for older data center workloads.

That focus has turned AI networking into one of the hotter corners of the infrastructure market. As hyperscalers and neocloud providers race to add more compute, they are running into a basic problem: it is no longer enough to buy more GPUs. Those chips need to communicate efficiently across large clusters, and the cost of poor networking shows up as lower utilization, slower training runs, and wasted infrastructure spending.

Upscale says it is already working with multiple hyperscalers and neocloud infrastructure providers, with evaluations and deployments underway across both scale-out and scale-up environments. The company did not disclose customers, though the new funding suggests investors believe the market for AI networking is large enough to support new challengers beyond incumbent players.

That matters because Upscale is trying to wedge itself into a market dominated by giants. Nvidia has its own tightly integrated AI systems strategy. Broadcom, Cisco, Arista, and a long list of data center suppliers are all chasing the same AI infrastructure dollars. Upscale’s bet is that customers will want open, interoperable networking gear that can tie together increasingly mixed AI environments without locking them into a single vendor’s stack.

Barun Kar, CEO of Upscale AI, framed the company’s pitch around the pressure large AI clusters are putting on existing data center designs.

“AI infrastructure is being redefined at cluster scale, and networking is one of the most critical bottlenecks. Upscale AI is building a high-performance, open-standard AI fabric purpose-built for large-scale, synchronized workloads,” Kar said in a statement. “This investment reinforces our vision and enables us to scale the business as we engage with leading neocloud and hyperscale infrastructure providers to meet growing demand for open, interoperable AI infrastructure.”

Investors pour $190M into AI infrastructure startup Upscale AI to tackle one of AI’s biggest bottlenecks

Investors are clearly buying into that thesis. In a statement announcing the deal, Premji Invest managing partner Sandesh Patnam said the firm had grown more confident in Upscale since its initial investment. Salesforce Ventures, one of the new backers in the round, pointed to growing demand for scalable systems that avoid vendor lock-in. Seligman Ventures made a similar argument, saying that open-standard fabrics will matter more as AI clusters span increasingly heterogeneous environments.

The size of the round shows how expensive this fight has become. Building AI infrastructure companies is capital-intensive work, especially when custom silicon, systems design, and supply commitments must be aligned long before products reach broad deployment. Upscale is not selling a lightweight software layer. It is trying to build core plumbing for AI data centers at a time when the largest buyers in the market are spending tens of billions of dollars to expand capacity.

That spending wave has created a rush of startups trying to carve out a slice of the AI stack beneath the model layer. Some are going after chips. Others are building storage, power, cooling, or data center software. Upscale is planting its flag in networking, betting that one of the biggest winners of the AI boom may not be the company with the smartest model, but the one that helps all those expensive machines talk to each other without slowing down.

For now, investors seem willing to fund that bet in a big way.