Upscale AI, a Santa Clara-based AI networking infrastructure startup, said Monday it raised $190 million in new financing, drawing Nvidia into a round that values the company at $2 billion and underscores investor interest in the networking layer behind large AI data centers.

The financing was structured as a Series A-1 extension and brings Upscale AI’s total funding to $500 million, according to the company. Premji Invest led the round. New investors included Nvidia, Salesforce Ventures, Seligman Ventures and Temasek, while Maverick Silicon, Mayfield, Prosperity7 Ventures, StepStone Group and Tiger Global returned as existing backers.

The round gives Upscale fresh capital for a market that has quickly become central to the economics of artificial intelligence. AI clusters are no longer limited by accelerator supply alone. As models grow and inference workloads move into production at scale, the network that links GPUs, custom AI chips, memory and storage can determine whether expensive hardware is fully utilized or sits idle while data moves between systems.

Upscale says it is building a full-stack portfolio spanning silicon, systems and software. Its pitch is that AI data centers need networking infrastructure purpose-built for synchronized workloads, rather than general-purpose data center networks designed for earlier generations of cloud computing. The company describes its platform as an open-standard AI fabric intended to support heterogeneous compute, meaning systems that may use accelerators or components from multiple suppliers.

That open-standards positioning is important because much of the current AI infrastructure market has developed around vertically integrated stacks. Nvidia’s GPUs, switching technologies and software ecosystem have helped define the modern AI buildout, while cloud providers and AI labs have also invested in custom chips and internal infrastructure. Upscale’s strategy is to make the network layer more interoperable, giving large infrastructure buyers a way to connect different types of compute while reducing dependence on any single vendor’s closed architecture.

Nvidia’s participation therefore carries a dual signal. The company remains the central supplier to many AI data center deployments, but its investment in Upscale suggests continued interest in the broader ecosystem needed to support AI workloads at cluster scale. For Upscale, having Nvidia in the round adds strategic credibility, even as the startup’s open-networking message is partly aimed at a market seeking more flexibility than proprietary stacks can offer.

Reuters reported that Upscale builds hardware, systems and software that connect AI chips, memory and storage across one fast network, helping large AI models train and run more efficiently with fewer delays. The company said it will use the new capital to expand the business and accelerate delivery of its advanced AI-native networking technology.

The company’s own announcement said customer evaluations and deployments are underway across both scale-out and scale-up networking environments, with engagement from multiple hyperscalers and leading neocloud infrastructure providers. Hyperscalers typically operate the largest cloud platforms, while neocloud providers have emerged to supply specialized GPU and AI compute capacity to startups, enterprises and model developers.

Those customers are under pressure to deliver more compute capacity while controlling power, latency and utilization costs. A single AI cluster can include thousands of accelerators that must exchange data continuously during training or inference. When networking is slow, lossy or difficult to coordinate across different hardware, system performance can fall short of the headline capacity implied by the number of chips installed.

Upscale CEO Barun Kar framed the problem as a cluster-scale infrastructure issue. In the company’s release, Kar said AI infrastructure is being redefined at cluster scale and described networking as one of the most critical bottlenecks. He said the investment would help Upscale scale the business as it works with neocloud and hyperscale infrastructure providers seeking open and interoperable AI infrastructure.

The financing also reflects a broader shift in AI venture investing. After an early wave of funding into foundation models and application-layer startups, capital has increasingly moved into the physical and systems infrastructure needed to support AI deployment. That includes data centers, power systems, optical interconnects, memory, specialized chips and networking technology. Upscale sits squarely in that infrastructure stack, where technical execution is costly and product cycles can be longer than in software-only startups.

Fortune reported that Upscale has reached $500 million in funding in less than 18 months and cited the cost of advanced custom chip development and supply commitments as reasons such companies need deep balance sheets before products fully scale. That is a central challenge for AI hardware and systems startups: they must spend heavily on engineering, manufacturing preparation and customer validation before revenue ramps in a market dominated by large incumbents.

Upscale’s founders bring prior networking and semiconductor experience to that challenge. Executive Chairman Rajiv Khemani previously founded Innovium, a networking chip company acquired by Marvell in 2021. Kar has held senior networking and security technology roles, including experience tied to Ethernet products and infrastructure companies. Investors are betting that experience can translate into a new generation of networking infrastructure for AI-era data centers.

The competitive field is already crowded. Nvidia, Broadcom and other established infrastructure suppliers have major engineering budgets and entrenched relationships with cloud customers. Startups are also entering the market with specialized approaches to AI networking, switching and interconnects. For Upscale, the main test will be whether its open-standard fabric can match the performance, reliability and operational simplicity that large customers require in production AI environments.

Still, the size of the financing shows that investors see networking as a major value pool in the AI buildout. AI models may capture the public attention, but the ability to train and serve them depends on whether data centers can move data fast enough across increasingly dense clusters. Upscale’s latest raise positions the company as one of the better-funded Silicon Valley startups trying to solve that problem at the silicon and systems level.

For the U.S. technology sector, the round is another sign that AI infrastructure competition is widening beyond GPU supply. As cloud companies, enterprise AI users and specialized compute providers search for more flexible architectures, open networking could become a key battleground. Upscale’s new funding does not prove that its architecture will win, but it gives the company more runway to test that thesis with the customers building the next wave of AI data centers.