H3C Launches UniPoD S81000 X1 Super Node, Supporting 40 Domestic GPUs per Domain
Summary
H3C, under New Unisplendour Group, launched the rack-level fully interconnected UniPoD S81000 X1 super node, which supports up to 40 domestic accelerator cards per Scale-Up domain, achieving a single-level CLOS full-mesh topology via four dedicated switch trays with 448GB/s P2P bandwidth between any two cards. The 40 cards aggregate into a 5760GB uniformly addressable memory pool delivering 28P FLOPS at FP8 precision, compatible with standard 19-inch racks and using a liquid-cooling-first thermal architecture. H3C says that as AI agents penetrate enterprise workflows, per-task token consumption of 100,000 to one million tokens creates triple pressure on underlying compute resources.
Bubble analysis
This is a hardware product launch with no investment figures, valuations or revenue data, so it speaks only indirectly to whether AI is in an investment bubble. It shows that China's domestic compute supply side keeps iterating and vendors are betting on agent-driven compute demand growth — a side signal of industry momentum, but it says nothing about whether capital spending has run ahead of actual AI revenue.
Related signals
linked by shared tags
- Toshiba Plans to Double HDD Capacity for AI Data Centers, Investing About 60 Billion Yen#datacenter
- Behind the 4 Trillion Compute Network: What Does the Tech Foundation of 'New Quality Productive Forces' Look Like?#datacenter
- Qineng Technology's Datong Super Energy Complex Tianzhen Computing Park Phase I Tops Out#datacenter
- Bain: Global AI Industry Must Generate $6 Trillion in Annual Revenue by 2031 to Justify Data Center Value#datacenter
- Nvidia-Backed Firmus Plans Australian IPO at $30 Billion Target Valuation#datacenter