Momenta CEO Cao Xudong publicly reveals the third insight for the first time: large models define chips
At the XHEART New Core Voyage launch event, Momenta CEO Cao Xudong publicly revealed the company's third insight, "large models define chips," for the first time, stating that chips defined by large models deliver 10 times the performance of general-purpose chips. Scaled L4 requires models with 10 billion to 100 billion parameters, and general-purpose chips struggle to simultaneously meet power consumption, automotive-grade reliability, and cost constraints. On the same day, XHEART released the X7 Ultra chip designed specifically for large models, adopting a super-large-core NPU architecture and debuting mass production of an Armv9 automotive-grade CPU in China.
Body · AI translation
Momenta CEO Cao Xudong Publicly Reveals Third Insight for the First Time: Large Models Define Chips
2026-10-29 16:38 IT Home - Guilong
IT Home, October 29, on-site news: At today's XHEART Xinxin Hangtu launch event, Momenta CEO Cao Xudong publicly shared the company's third insight for the first time — large models define chips.
IT Home has learned that, centered on the goal of scaled L4, Momenta had previously formed two insights: solving long-tail problems through data-driven approaches, and obtaining massive amounts of data through large-scale mass production. At the end of 2022, Momenta further proposed that "large models define chips" and put it into action together with XHEART.
Cao Xudong said that chips defined by large models deliver 10 times the performance of general-purpose chips; scaled L4 requires models with 10 billion to 100 billion parameters, and general-purpose chips find it difficult to simultaneously satisfy the triple constraints of power consumption, automotive-grade reliability, and cost, which he called the "impossible triangle of general-purpose chips."
He believes that only by defining chips starting from the needs of large models is it possible to support 100-billion-parameter models under the above constraints, achieve scaled L4, and further drive breakthroughs in robotics and more physical AI applications.
According to an earlier report by IT Home today, XHEART Xinxin Hangtu CEO Li Zonglin officially released the all-new X7 Ultra large-model chip. This chip is designed specifically for large models, features a super-core NPU architecture, and boosts interconnect speeds by orders of magnitude to the TB/s level for intra-core data interaction compared with inter-core data interaction; it is the first in China to mass-produce an Armv9 automotive-grade CPU, with a 20% performance improvement and a 10% energy-efficiency improvement.