**ByteDance develops custom CPUs for AI inference amid chip supply crunch**
The AMW Read
Major hyperscaler entering custom CPU design updates the segment player map; silicon shift to inference and supply crunch carry cross-segment infrastructure significance.
**ByteDance develops custom CPUs for AI inference amid chip supply crunch**
ByteDance, the parent company of TikTok, is designing proprietary central processing units (CPUs) to support its growing AI infrastructure, three sources told Reuters. The company is pursuing two architectures simultaneously — one based on Arm and one on open-source RISC-V — and has approached external partners for design and manufacturing assistance. ByteDance currently relies on Intel and AMD for CPUs, but those suppliers have raised prices 10–35% quarter-over-quarter and extended delivery lead times to six months, according to the sources. The project is still at an early stage.
This story matters because it signals a structural shift in AI compute. The AI industry is moving from a training-dominated paradigm to an inference-heavy one, where agentic workloads place new demands on CPUs working alongside Nvidia GPUs. ByteDance joins Google, Amazon, and Microsoft in concluding that the economics of custom silicon outweigh the complexity of in-house design — a pattern of hyperscaler vertical integration into the chip stack that reshapes who captures value in AI infrastructure. The CPU shortage itself, flagged by Intel and AMD, points to a broader supply-demand imbalance that is forcing even large AI players to build rather than buy.
The move also updates the competitive landscape inside China: ByteDance is now investing in silicon sovereignty alongside its foundation-model and AI-agent ambitions (Coze platform). If successful, the company could reduce its exposure to both Intel/AMD price hikes and geopolitical supply-chain risk. However, the dual-architecture hedge (Arm + RISC-V) signals that the final design is far from settled, and manufacturing capacity at foundries remains a gating factor. This is a multi-year play with execution risk, but it confirms that the anti-Nvidia coalition is deepening.
