ByteDance is pursuing development of its own central processing units to underpin growing artificial intelligence infrastructure needs, three people familiar with the matter said. The initiative is intended to help the Beijing company support an expanding rollout of agent-based services while addressing surging CPU prices and an industry-wide shortage that has constrained expansion plans.
Those familiar with the effort said ByteDance aims to install the custom processors in its own servers and data centres to support internal operations and a large-scale deployment of agent-based products, including its Coze platform. The project remains at an early stage, the first source said, and the plan has not been publicly disclosed.
To advance the programme, ByteDance has engaged several external partners. These collaborators are expected to play roles beyond chip design, assisting in efforts to secure manufacturing capacity at foundries for any subsequent production runs, the sources added. The company declined to comment through available channels.
The move reflects an industry shift toward inference workloads - deployments of AI models to carry out agentic tasks that place different demands on CPUs than earlier model stages. As graphics processing units supplied by Nvidia have dominated the initial AI surge, the growing emphasis on inference has increased demand for CPUs and contributed to shortages in recent months.
Major cloud operators and hyperscalers have responded by creating their own custom central processors to control costs and tailor performance for specific workloads. Google, Amazon and Microsoft are among those developing bespoke CPUs, and major CPU suppliers such as Intel and AMD have also moved into more prominent roles as challengers in the AI hardware stack.
ByteDance is developing two CPU architecture tracks in parallel, the sources said. One architecture is based on Arm, owned by SoftBank, and the other uses the open-source RISC-V instruction set architecture. Developing two designs concurrently is a commonly used hedge by large technology firms, enabling testing of alternatives before committing to costly, large-scale manufacturing.
Supply-side pressures and rapid demand growth have been highlighted by major chipmakers. Intel has warned Chinese customers of server CPU delivery lead times of up to six months, and reported strong first-quarter demand from AI customers that led it to sell chips it had previously written off, according to the people familiar with the situation. AMD's chief executive has said the global CPU market is tight, with demand outpacing forecasts and supply constraints expected to persist.
ByteDance currently purchases server CPUs from Intel and AMD, but the company has seen significant price increases for those components recently. Two of the sources said quarter-over-quarter price hikes ranged from about 10% to as much as 35% in recent months. Those increases were cited as a catalyst for accelerating the push toward in-house alternatives.
Intel said it had updated prices for some of its products to reflect sustained demand, higher component and material costs, and evolving market dynamics. AMD did not provide an immediate response to a request for comment.
Meanwhile, Nvidia is expanding beyond graphics processing units into the CPU market. Its chief executive has positioned the firm's new central processors, known as Vera, as access to a sizeable market opportunity. In March, Nvidia unveiled a new central processor and an AI system built on technology from a chip startup specialising in inference, as it seeks to defend its position in the AI chip landscape.
Sectors impacted: Data centre operations, cloud infrastructure, semiconductor manufacturing, enterprise software that relies on agent-based AI deployments.
Context limitations: Sources described the project as early-stage and anonymous; no public timetable, detailed technical specifications, or production commitments were provided.