Momenta CEO Introduces 'Large Models Define Chips' Strategy
Momenta CEO Cao Xudong unveiled the company's third core insight, 'Large Models Define Chips,' at the XHEART press conference. Alongside this announcement, XHEART launched the X7 Ultra, a custom chip designed specifically for large-scale autonomous driving models. This approach addresses the 'impossible triangle' of power consumption, automotive-grade reliability, and cost that generic chips face when handling massive AI models. Co-designing hardware specifically for large models is positioned as a critical step toward achieving scalable L4 autonomous driving and physical AI breakthroughs. The newly released X7 Ultra chip features a large-core NPU architecture with terabyte-per-second inter-core and intra-core data interaction rates. It also introduces China's first mass-produced Armv9 automotive-grade CPU, delivering a 20% performance boost and a 10% increase in energy efficiency.
## BACKGROUND
L4 autonomous driving requires vehicles to operate fully autonomously under most conditions without human intervention, often necessitating complex large language models with tens to hundreds of billions of parameters to handle unforeseen edge cases. Automotive-grade reliability refers to rigorous industry standards like AEC-Q series and ISO 26262 that ensure electronic components can safely endure harsh operational environments in vehicles over long periods.