The company behind TikTok is said to be accelerating efforts to reduce reliance on external chip suppliers such as Intel and AMD, amid growing demand for computing capacity driven by AI systems and emerging agent-based applications.
The shift reflects a broader transformation in the AI industry, where “inference” workloads — AI systems executing real-world tasks — are increasingly placing heavy demand on CPUs, complementing the graphics processing units (GPUs) supplied by Nvidia, which have dominated the AI boom.
Rising CPU Demand and Industry-Wide Shortages
The surge in AI workloads has tightened the global CPU market, prompting hyperscalers such as Google, Amazon, and Microsoft to design their own custom processors to optimise performance and reduce long-term infrastructure costs.
ByteDance is now following a similar path, aiming to deploy its in-house CPUs within its servers and data centres to support internal systems and a planned expansion of AI-driven services, including its agent platform Coze.
“The Beijing-based company is targeting deployment of its proprietary CPU in its own servers and data centres to support internal operations,” one of the sources said, adding that the initiative remains at an early stage.
The company has reportedly engaged external partners to assist with chip design and to help secure manufacturing capacity at semiconductor foundries. These partners are also expected to contribute technical expertise as ByteDance navigates the complex process of chip development.
Dual Architecture Strategy: Arm vs RISC-V
ByteDance is pursuing two parallel CPU development paths — one based on Arm architecture and another on the open-source RISC-V framework — as it evaluates which design best aligns with its long-term data centre strategy.
Analysts say this dual-track approach is a common strategy among large technology companies, allowing them to test multiple architectures before committing to large-scale production.
Arm-based designs are widely used across mobile and server ecosystems, while RISC-V has gained traction as an open alternative that offers greater flexibility for custom chip development.
Arm did not immediately respond to a request for comment.
Supply Bottlenecks Push Tech Giants Toward In-House Chips
The move comes amid growing pressure on global chip supply chains. Intel has reportedly warned Chinese customers of server CPU delivery lead times of up to six months, highlighting ongoing constraints in the market.
In February, reports indicated that demand for server CPUs had outstripped supply, while Intel itself acknowledged last month that strong AI-related demand led it to sell even chips previously written off.
AMD chief executive Lisa Su also recently warned that the CPU market remains “tight,” with demand exceeding forecasts and supply constraints expected to persist.
ByteDance currently relies on Intel and AMD for its CPU supply, but rising prices have added further pressure. According to two of the sources, CPU costs have increased by between 10 per cent and 35 per cent quarter-on-quarter in recent months, accelerating ByteDance’s push toward self-sufficiency.
Intel has said it has adjusted pricing on some products to reflect sustained demand and higher material costs, while AMD declined to comment on pricing dynamics.
Global Tech Giants Race to Control Chip Supply
ByteDance’s strategy places it alongside global technology leaders increasingly investing in semiconductor design to reduce dependency on external suppliers.
Companies such as Google, Amazon, and Microsoft have already developed custom CPUs tailored to their cloud and AI workloads, seeking greater control over performance, efficiency, and cost.
At the same time, chipmakers are adapting to the shift. Nvidia is expanding beyond GPUs into CPUs, aiming to capture a larger share of the AI computing stack. Its upcoming “Vera” processors are designed to compete directly in the growing CPU market, which CEO Jensen Huang has described as potentially worth $200 billion.
The company also unveiled a new central processor and AI system based on technology from chip startup Groq in March, signalling its intent to defend its dominance across the evolving AI hardware ecosystem.
Strategic Shift Toward Inference-Driven AI
Industry experts say the broader trend driving these developments is the rapid rise of inference workloads — where trained AI models are deployed at scale to perform tasks such as automation, decision-making, and digital assistance.
Unlike training workloads that rely heavily on GPUs, inference requires a more balanced mix of CPUs and GPUs, increasing pressure on both segments of the semiconductor market.
As AI systems become more embedded in everyday digital infrastructure, competition for compute resources is expected to intensify further, pushing more companies toward custom silicon strategies similar to ByteDance’s emerging plan.