Nvidia-backed AI networking startup Upscale has closed a $190 million Series A-1 extension, bringing its valuation to $2 billion. The funding round underscores growing investor confidence in addressing a critical infrastructure challenge as AI training systems scale.

Upscale is developing custom silicon for AI cluster interconnects, targeting one of the key bottlenecks in large-scale training environments. Networking constraints have emerged as a limiting factor alongside GPU availability, as model training increasingly requires seamless data movement across hundreds or thousands of accelerators in distributed systems.

Nvidia's backing of the startup signals the semiconductor giant's strategy to expand beyond GPUs into the broader AI infrastructure stack. For data center operators and AI model developers, Upscale's custom interconnect solutions represent a potential path to improve cluster efficiency and reduce the time and cost of large-scale training runs.