ByteDance is designing its own server CPUs to power its AI infrastructure and support the large-scale deployment of agent platforms like Coze. The move comes as Intel and AMD have raised server CPU prices by 10% to 35% quarter over quarter in recent months, with lead times stretching to six months — pushing ByteDance to accelerate its in-house chip plans to control costs.
On the technical side, ByteDance is pursuing two CPU architectures simultaneously: Arm and the open-source RISC-V instruction set. Running both tracks in parallel lets the company evaluate which architecture offers the best performance per dollar for its data center workloads before committing to expensive tape-outs and mass production. Although the project is still in early stages, ByteDance has already approached multiple external partners to collaborate on chip design and is lining up foundry capacity.
The push for custom silicon is driven by the rapid shift toward AI inference. Running complex agent tasks demands more CPU horsepower and smarter scheduling. As demand for server processors surges, Intel warned Chinese customers in February that lead times had hit six months, and AMD CEO Lisa Su acknowledged that global CPU supply would remain tight. To cut computing costs and optimize for specific workloads, hyperscalers like Google, Amazon, and Microsoft have all launched their own CPUs. Nvidia is also accelerating its entry into the $200 billion CPU market, planning new Vera processors and AI systems integrating Groq inference technology to defend its dominance in the compute ecosystem.