Upscale AI has secured $190 million in a Series A-1 funding extension, bringing the Santa Clara-based startup’s total capital raised to $500 million and establishing a company valuation of $2 billion. The networking infrastructure firm, which focuses on backend connectivity for GPUs and XPUs, attracted new investment from Nvidia, Salesforce Ventures, Seligman Ventures, and Temasek, with Premji Invest leading the round. The funding arrives as the company prepares to scale its specialized AI networking hardware, including a custom switch ASIC currently under development.
The company, which emerged from stealth in September 2025 with an initial $100 million seed round, is positioning itself as a pure-play provider for hyperscalers and neocloud operators. According to company leadership, the influx of capital will support the development of both “scale-up” networking—designed to link processors within a single cluster—and “scale-out” networking, which manages connectivity across broader data center nodes. The round also saw continued participation from existing backers, including Maverick Silicon, Mayfield, Prosperity7 Ventures, StepStone Group, and Tiger Global.
Strategic Alignment with Nvidia and Spectrum-X
Nvidia’s participation in the funding round marks a deepening of a relationship that began on the product development front. Upscale AI has integrated its scale-out roadmap with Nvidia’s Spectrum-X Ethernet platform, a move intended to build network fabrics optimized for AI workloads. Rajiv Khemani, co-founder and executive chairman of Upscale AI, stated that these systems are currently in customer labs and are already booked for early deployments, with a broader market release expected later this year.
This technical collaboration serves to address the increasing demand for high-speed connectivity in heterogeneous AI computing environments. By aligning with Nvidia’s ecosystem, Upscale aims to ensure its networking hardware remains compatible with the most widely used GPU infrastructures while maintaining the flexibility to support other XPU vendors. The strategic investment from Nvidia effectively cements the startup’s role in the hardware supply chain for large-scale AI deployments.
Developing the Skyhammer Switch Silicon
A central component of Upscale AI’s hardware strategy is the development of a custom scale-up switch ASIC, internally code-named Skyhammer. Unlike commodity data center chips that have been adapted for AI tasks, Khemani describes Skyhammer as purpose-built silicon designed specifically for the low-latency, high-speed requirements of AI scale-up use cases. The chip is intended to function as a critical link between GPUs and XPUs within a cluster.
Upscale has not yet disclosed a definitive timeline for the release of Skyhammer, noting that the deployment schedule is inherently linked to the readiness of GPU and XPU vendors. Because the switch silicon is designed to work in conjunction with these processors, its rollout must be synchronized with the availability of hardware that supports the necessary scale-up interoperability. The company plans to reveal further product specifications later this year as the integration process with partner hardware advances.
Addressing Token Efficiency and AI Infrastructure
Beyond hardware, Upscale AI is developing a full-stack offering that incorporates software designed to optimize AI token usage—an industry challenge often referred to as “tokenmaxxing.” Khemani notes that as AI applications evolve from simple prompts to complex, multi-step agentic loops, the requirements for underlying network infrastructure change accordingly. He argues that there is no single solution to token efficiency, as the optimal approach remains highly dependent on the specific workload.
%20(1).webp "Addressing Token Efficiency and AI Infrastructure")
The company’s focus on a full-stack strategy—integrating silicon, systems, and software—is intended to provide the flexibility required to adapt to these shifting AI demands. By building an infrastructure that can be optimized for various application requirements, Upscale aims to provide hyperscalers and neocloud providers with the tools to maximize the utility of their compute resources. The company maintains that ensuring AI infrastructure operates in an optimal fashion is a necessary evolution for the industry.
What Comes Next for Upscale AI
Following the completion of the Series A-1 extension, Upscale AI is expected to focus on the deployment of its Spectrum-X based systems and the continued R&D of the Skyhammer silicon. Industry observers and prospective customers can expect further technical disclosures regarding the Skyhammer architecture later this year. The company has not provided a specific date for these disclosures, stating that updates will be shared as the product roadmap aligns with the broader ecosystem of GPU and XPU manufacturers.
As Upscale AI continues to scale, its ability to successfully bridge the gap between legacy data center networking and the unique demands of AI compute clusters will remain a focal point for investors and industry partners alike. For ongoing updates regarding product availability and technical specifications, stakeholders are encouraged to monitor official announcements from the company. We invite readers to share their thoughts on the evolution of AI networking and the role of specialized silicon in the comments section below.
Keep reading