❯ ByteDance AI chip chief Shi Yunfeng reportedly enters departure handover as chip adaptation work continues

A leadership handoverLatePost reported on November 11 that Shi Yunfeng, who leads ByteDance’s AI chip and DPU teams, has submitted his resignation and is handing over responsibilities. He joined in 2018 after roles at MediaTek, Marvell and HXT, and reports to ByteDance vice president and Volcano Engine head Yang Zhenyuan. No successor or handover completion date was disclosed.

A specialized chip strategyEstablished in 2019, the team moved from AI Lab into the Data organization before relevant teams began reporting directly to Yang this year. It follows an application-specific chip approach similar to Google’s TPU. LatePost says chips completed fabrication this year but performed below expectations, while work on a subsequent generation continues.

Software work remains essentialAda for Doubao and SeedChip for Seedance are undergoing operator adaptation, making model computations run correctly and efficiently on the hardware. Specialized chips cannot simply use the complete CUDA software ecosystem, so models, frameworks and underlying software need adaptation after fabrication. LatePost reports that the chips have not been deployed widely inside ByteDance, with migration and debugging requiring participation from product teams.

External technology remains part of the planByteDance is also investing and partnering externally, including with ReRAM memory specialist Xinyuan Semiconductor. LatePost says a proposed integration of another invested chip project into the Picoheart team has not closed. The personnel change does not itself end internal chip development: software adaptation, next-generation designs and outside technology integration remain at different stages.

▮ SIGNALGetting ByteDance’s chips into real workloads requires joint investment by hardware, model and software teams, making organizational coordination part of the deployment challenge.