Hygon Unveils 'Intrinsic Security' Technology for Domestic Trillion-Parameter AI Model Training

According to ME News, on April 2 (UTC+8), at Higon Information’s 2026 Spring Technology Symposium, Higon officially unveiled multiple new achievements based on its “intrinsic security” concept and launched the annual version of its DCU software stack, providing computing power support for the training of multiple domestic large models with trillions of parameters. Higon Information Vice President Ying Zhiwei stated that the security challenges of computing power in the AI era have surpassed the capabilities of traditional solutions and must be addressed by breakthroughs at the most fundamental level of computing architecture. On a computing cluster comprising tens of thousands of Higon CPUs and DCUs, the Higon team has preliminarily achieved stable operation of a 10-trillion-parameter model through multiple technologies, including intrinsic security, operator optimization, compiler optimization, and integrated computing. Multiple benchmarks, including MLPerf, have reached international leading levels. (Source: Ifnar)