ByteDance is building its own central processing units to underpin its expanding AI infrastructure, three people familiar with the project told Reuters in an exclusive report published May 28. The effort is driven by a combination of surging chip prices and persistent supply shortages that have begun constraining the company’s growth plans. The Beijing-based company has approached multiple external partners who are expected to assist with both chip design and securing manufacturing capacity at foundries. According to the sources, ByteDance is currently evaluating two CPU architecture tracks in parallel — one based on SoftBank-owned Arm and another on the open-source RISC-V instruction set — as it determines which design best fits its long-term data center needs. The project remains at an early stage, sources said. ByteDance did not respond to Reuters’ request for comment. The CPU push reflects the broader industry shift from AI training toward inference, where models are deployed to carry out agentic tasks that place heavier demands on CPUs working alongside Nvidia’s GPUs.
The move places ByteDance alongside a growing class of tech companies that have concluded the economics of custom silicon outweigh its design complexity. Google, Amazon, and Microsoft have each developed proprietary CPUs for their cloud infrastructure, and the same inference-driven CPU shortage has helped Intel and AMD emerge as meaningful challengers in the AI compute market. For Chinese tech companies, U.S. export controls that restrict access to Nvidia’s most advanced chips add an additional layer of urgency. ByteDance’s chip efforts date back to at least 2022, and it previously worked with Broadcom on an AI ASIC in 2024. Separately, Bloomberg reported this week that ByteDance had agreed to procure millions of custom AI ASICs from Qualcomm to support its AI agent software “Dou+,” with TSMC handling manufacturing — a deal that underscores how ByteDance is simultaneously pursuing multiple chip supply strategies.