ByteDance, the operator of TikTok, is accelerating development of a next-generation in-house central processing unit (CPU). As computing demand surges with the expansion of AI services, it is seen as a strategy to move away from a GPU-centric structure and increase the share of its own semiconductors.

TechRadar, an IT media outlet, reported on Monday that ByteDance is pursuing a plan to complete the design of its next-generation in-house CPU by early 2027 and begin mass production and apply it to internal services in the second half of that year. The company has already deployed an early version of its in-house processor in some systems since late 2025, the report said. It added that ByteDance is also reviewing ways to bring forward the development schedule for follow-up chips.

The move comes as ByteDance’s AI service expansion drives a sharp rise in computing demand. ByteDance is rapidly expanding its AI business with products such as the generative AI chatbot Doubao and the video-generating AI Seedance. As a result, the computing burden at data centres is also rising sharply, and the spread of agentic AI in particular is demanding a different computing architecture than before.

Existing generative AI has centred on GPU-based matrix operations. But AI agents significantly increase the share of general-purpose computing such as task planning and decision-making, memory management and software execution. That makes the role of CPUs supporting GPUs more important.

ByteDance appears to be reshaping its infrastructure to reduce dependence on external GPUs and use its own processors in response to these changes. The outlet said next-generation AI systems "require more coordination, decision-making, memory management and software work, and computing architectures are also shifting away from GPUs alone toward collaboration with general-purpose processors."

The development timeline could also move faster than expected. The industry is seen as reviewing ways to bring forward the 'tape-out' stage, the pre-production step after chip design is completed. ByteDance has not officially disclosed its semiconductor development plans, but it is known to be expanding its in-house chip capabilities by running multiple chip projects at the same time.

External cooperation is also being pursued during development. ByteDance is working with U.S. chipmaker Qualcomm to speed up design development and is also pushing to secure foundry production capacity, the report said. That shows that securing actual production volumes, as well as chip design, is a key task.

U.S. restrictions on semiconductor exports to China are also a factor spurring in-house chip development. The U.S. government has gradually restricted exports to China of high-performance AI chips such as Nvidia's H100 and H20. In this environment, large Chinese technology companies are accelerating in-house chip development to reduce reliance on external supply chains.

The market sees ByteDance's in-house chip strategy as something that could affect the semiconductor industry over the long term. If its in-house CPU is introduced in earnest, demand for some general-purpose processor suppliers such as Arm, Intel and AMD could decline. Nvidia, which is already facing the impact of regulations in the Chinese market, could also see the expansion of internal substitute chips emerge as a new variable.

The move also aligns with a broader trend among global big tech companies. Google is developing its own TPU chips, Amazon is developing Trainium and Graviton, and Microsoft is developing its Maia AI chip, seeking cost cuts and performance optimisation. ByteDance is also interpreted as pursuing a strategy to secure customised semiconductors based on large-scale AI services and boost long-term competitiveness.

Market attention is likely to focus on whether ByteDance will formalise its tape-out schedule. If the schedule is disclosed, it would be seen as a signal that ByteDance's semiconductor business is moving beyond simple research and development and becoming a full-fledged AI infrastructure strategy.