Upscale AI Lands $190M to Build AI Networking

The big picture: Upscale AI, a Santa Clara-based startup developing networking infrastructure for large AI clusters, has secured $190 million in an extension of its Series A round. Backers include Nvidia, Salesforce Ventures, Premji Invest, Temasek, and Seligman Ventures, pushing the company’s valuation to $2 billion and bringing its total funding to $500 million.

Why it matters:

• Bottleneck relief: While AI systems excel at generating content, the underlying infrastructure, particularly networking, remains a significant bottleneck for training frontier models and serving large inference workloads.
• Efficiency gains: Inefficient networking in large AI data centers leads to lower GPU utilization, slower training runs, and increased infrastructure spending, highlighting the need for specialized AI networking solutions.
• Open standards: Upscale AI aims to provide open, interoperable networking gear, offering an alternative to vendor-locked ecosystems and enabling customers to connect increasingly mixed AI environments.

How it works:

• Full-stack platform: Upscale AI is building a comprehensive AI networking platform that spans silicon, systems, and software, addressing the entire networking stack.
• Open-standard fabric: The company’s core technology connects accelerators, memory, and storage via an open-standard fabric specifically engineered for the demands of large-scale, synchronized AI workloads.
• Hyperscaler engagement: Upscale AI is actively engaged with multiple hyperscalers and neocloud infrastructure providers, with evaluations and deployments underway in both scale-out and scale-up environments.

The catch: Upscale AI is entering a market dominated by established giants like Nvidia, Broadcom, Cisco, and Arista, all competing for AI infrastructure dollars. Building custom silicon and systems for core data center plumbing is capital-intensive, requiring significant investment and long lead times before broad product deployment.