SUGON

In this Vendor Showdown slot moderated by Addison Snell of Intersect360 Research and Rupak Biswas of NASA Ames Research Center, Li Yang presents a weather intelligence company affiliated with Chinese systems vendor Sugon. The company was among the first in China to migrate a physics-based numerical weather prediction model onto GPUs and has run it stably for over five years, with a mission of converting NWP output into industry-ready weather intelligence. Its core engine, SD3, features a non-hydrostatic dynamical core stable at kilometer scale, inline physics that couples physical processes directly with the dynamics to address gray-zone problems where traditional parameterization fails, and a clean split of fast and slow processes allowing larger time steps for lower compute cost at equal accuracy. Services include real-time, mid-term, and long-term forecasts and historical data via standard APIs, plus private deployment packages using Sugon's ScaleX super pod. Renewable energy is the primary market, with plant-level forecasts up to 46 days and real-time forecasts refreshed every 15 minutes for the next four hours to support power trading and dispatch; transportation is another focus, helping ports, railways, and roads avoid fog and storms. Datasets are published openly through the AWS open data program. In questions, Li notes the model produces a 10-day global forecast in 80 minutes, incorporates AI within the NWP loop including large language models for input and AI-based corrections, is CUDA-based but portable across NVIDIA and AMD GPUs, and reports an anomaly correlation coefficient of 0.6 validating 10-day forecasts.