According to The Information, ByteDance is developing a new AI inference chip with a design philosophy similar to that of the Language Processing Unit (LPU) by U.S. chip company Groq, aiming to run trained AI models at a lower cost. ByteDance is collaborating with Shanghai-based memory chip company XinYuan Semiconductor to explore integrating the latter's RRAM memory technology into the new chip. Sources familiar with the matter said the new chip design may not use HBM to reduce reliance on key components subject to U.S. export controls. ByteDance's move aims to lower inference costs, enhance the autonomy of its AI infrastructure, and address uncertainties in external supply.
