How China's ParaStor F9000 Became the First Domestic Storage to Top Global IO500 Rankings

Record‑breaking performance on the IO500 production benchmarks

The ISC 2026 conference announced that Sugon’s self‑developed ParaStor F9000 distributed all‑flash storage system secured the #1 position on two of the most stringent production‑grade IO500 lists: the full‑node ranking and the 10‑node ranking. The system achieved more than 247% improvement in bandwidth performance, over 126% in metadata performance, and a 171% increase in the 10‑node world‑record metric.

Why the IO500 production list matters

Unlike laboratory‑only benchmarks, the IO500 production list requires participating systems to run on real, continuously operating production clusters, handle heterogeneous data loads, and remain stable throughout a full year of service. Only mature, commercially‑deployed products qualify.

Real‑world deployments backing the numbers

ParaStor F9000 underpins the scaleX super‑cluster with tens of thousands of accelerator cards, supporting more than 60,000 cards in a national supercomputing internet. It currently runs over a hundred AI and high‑performance computing applications across domains such as artificial intelligence, autonomous driving, scientific simulation, and robotics.

1. AI4S scientific computing – 414.7 billion‑atom water simulation

Leveraging the domestic storage foundation, Longsun Kuangteng completed a 414.7 billion‑atom liquid‑water molecular dynamics simulation, setting a new record for simulation scale and accelerating research in weather forecasting, material design, protein analysis, and drug discovery.

2. Embodied‑intelligence robot – >500 GB/s aggregated read bandwidth

A dedicated storage solution for ZhiYuan robots provides more than 500 GB/s of read bandwidth, enabling rapid access to massive multimodal data for large‑model training and achieving millisecond‑level data retrieval for smooth real‑time interaction.

3. Autonomous‑driving data pipeline – 40% reduction in development cycle

The system supplies over 100 PB of domestically‑produced storage to a leading Chinese automaker, handling petabyte‑scale sensor data (cameras, LiDAR, radar) and delivering a complete “capture‑clean‑label‑train‑simulate” workflow. Intelligent tiered scheduling cuts total project cost by 40% and shortens the autonomous‑driving iteration period by the same margin.

Three distinguishing attributes of the platform

Label 1: The only Chinese vendor to win both production rankings simultaneously

Historically, high‑end storage benchmarks have been dominated by overseas giants. ParaStor F9000’s dual victory demonstrates that China’s end‑to‑end stack—from storage chips to distributed architecture and acceleration engines—now surpasses international peers in performance, stability, and large‑scale delivery.

Label 2: AI‑storage market share leader in China for two consecutive years

Deeply embedded in the AI storage lane, the solution serves clusters ranging from thousands to hundreds of thousands of cards, delivering native support for trillion‑parameter models. Measured results show a 50% reduction in model deployment time, a 50% boost in overall training efficiency, and a 60% cut in training cycles for trillion‑parameter models, eliminating common storage bottlenecks.

Label 3: Fully domestically designed hardware and software

ParaStor F9000 features a 100% Chinese‑developed stack with three core acceleration technologies:

XDS zero‑copy path directly connects storage to GPU/NPU memory, halving data‑load latency.

Burst‑Buffer local cache raises small‑file read/write performance tenfold and enables minute‑level checkpoint recovery.

KV Cache offload lifts single‑card inference concurrency tenfold, reduces inference latency by 80%, and removes the need for expensive HBM memory.

The system is also deeply optimized for vector databases, supporting RAG retrieval and long‑document question answering with high concurrency, and includes an on‑storage AI operator library that cuts cross‑node data transfer by 70%, freeing compute resources.

Implications of the global top‑ranking

Domestic storage technology has overtaken foreign competitors, establishing a self‑reliant high‑performance infrastructure foundation.

The platform fills the storage gap for massive AI clusters, ensuring that GPU utilization is limited only by compute, not by data movement.

High‑performance domestic storage becomes a core enabler for digital transformation across sectors—from national supercomputing research to autonomous driving, robotics, finance, energy, and healthcare—while reducing costs and preserving data security.

Conclusion

From following the leaders to competing on equal footing and now standing at the summit, China’s home‑grown storage has delivered a world‑shaking performance record. Sugon will continue to advance storage‑compute‑network co‑design, supporting ever larger AI innovations and frontier scientific exploration, keeping China at the forefront of global digital infrastructure.