From Expertise to Intelligence: NISCO Deploys World's First Swarm-Intelligence Steel Model at Scale
Reported by The Register and Hiru News2 sources ↓
NISCO says it has deployed the YuanYe Steel Large Model across multiple production scenarios. The company reports more than CNY100 million in annual cost reductions and a 50% shorter new-product R&D cycle.
Key points
- NISCO says YuanYe is deployed across ironmaking, steelmaking, rolling and quality inspection
- NISCO had launched 25 business agents by October 2026
- NISCO reports annual cost reductions of more than CNY100 million
- The company says new-product R&D cycles are 50% shorter
- NISCO says 100% of its key equipment is connected and digitally controlled
Sources
Details
The Register reports that Nanjing Iron & Steel, working with Huawei and ecosystem partners, has commercially deployed YuanYe across multiple steelmaking scenarios. NISCO describes YuanYe as a system combining AI models with metallurgical mechanism models, using operational data from 26 business domains. The company says it had launched 25 business agents by October 2026.
NISCO says its furnace models can predict temperatures and identify issues two hours ahead, with more than 90% accuracy, and have increased prime-grade molten iron yield by more than 10%. For rolling, the company reports a rolling-force prediction hit rate of 98%, up from 88%, and says the system supports annual output of more than 2.5 million tons of steel plates.
For metallographic inspection, NISCO says automation has raised inspection accuracy to more than 95% and shortened inspection time from 10 minutes to three to five minutes per sample. The company also reports that its energy and emissions agent predicts sulfur risks one hour ahead with more than 95% accuracy, and has cut natural gas co-firing consumption by nearly 50%.
Across its operations, NISCO says all key equipment is connected and digitally controlled, new-product R&D cycles are 50% shorter, and annual costs have fallen by more than CNY100 million. These figures are company-reported; the article does not provide independent measurements.
Written from coverage as of