By Liam Mo and Fanny Potkin
BEIJING/PARIS, May 28 (Reuters) – Chinese tech giant ByteDance is developing its own central processing units (CPUs) to meet its growing artificial intelligence infrastructure needs, three people familiar with the matter said, as rising chip prices and prolonged supply shortages limit its expansion plans.
The move underscores the industry’s rapid shift toward “inference,” where AI models are deployed to perform agentic tasks that demand more from CPUs, working in tandem with the Nvidia-made graphics chips that have dominated the AI ​​boom.
The shift has created a CPU shortage in recent months, and global hyperscalers including Google, Amazon and Alphabet’s Microsoft are also developing their own custom CPUs to reduce costs and tailor performance to their specific workloads. It has also helped major CPU makers Intel and AMD emerge as major rivals to Nvidia’s AI dominance.
ByteDance, the parent company of short video platform TikTok, aims to deploy its proprietary CPU in its own servers and data centers to support internal operations, while preparing a massive launch of agent-based products, including its Coze platform, the first source said.
The Beijing-based company has approached several outside partners to help with the effort, and those partners are expected to contribute not only to chip design work but also to help secure manufacturing capacity at foundries, the sources added. The project is in an initial phase, the first source stated.
They declined to be identified as the plan is not public.
ByteDance did not respond to Reuters’ request for comment.
CPU SHORTAGE
ByteDance’s move puts it among a growing cohort of technology companies that have concluded that the economics of custom chips outweigh the complexity of designing them.
It is pursuing two chip architecture paths for its CPU development — one based on SoftBank-owned Arm and another on the open source RISC-V instruction set architecture — as it weighs which design best suits its long-term data center requirements, the sources said.
Developing two designs simultaneously is a common protection for tech giants, allowing them to test their options before committing to expensive large-scale production.
Arm did not immediately respond to a request for comment.
The push to develop proprietary silicon comes as Intel warned Chinese customers of server CPU delivery times of up to six months, Reuters reported in February.
Intel said last month that demand for its CPUs from artificial intelligence companies was so strong in the first quarter that it even sold chips it had originally canceled.