Nexo Earn with Nexo
ByteDance develops custom CPUs to support AI infrastructure amid chip shortages

ByteDance develops custom CPUs to support AI infrastructure amid chip shortages

TikTok's parent company is designing its own processors as CPU prices from Intel and AMD surge by up to 35%, joining a growing club of tech giants building custom silicon.

ByteDance, the company behind TikTok, is building its own CPUs. The reason is straightforward: buying them from Intel and AMD has become painfully expensive.

CPU prices from major suppliers have climbed 10-35% quarter-over-quarter, according to Reuters, and supply constraints have stretched expansion timelines to the breaking point. For a company projected to spend RMB 160 billion (roughly $22.8 billion) on AI-related capital expenditures in 2026, those price hikes add up fast.

The custom silicon play

ByteDance is currently in early-stage development, evaluating both Arm and RISC-V architectures for the new chips. The company is also seeking partnerships for design and manufacturing, which suggests it’s not trying to build a full-stack semiconductor operation from scratch.

Advertisement

The target use case is internal. These CPUs are meant to power ByteDance’s own servers and data centers, with a particular focus on supporting agent-based AI products like its Coze platform.

Nearly half of that $22.8 billion AI budget is reportedly earmarked for advanced semiconductor development.

ByteDance joins the club

ByteDance isn’t pioneering this approach. It’s following a well-worn path carved by Google, Amazon, and Microsoft, all of which have invested heavily in custom CPU designs to optimize cost and performance for AI workloads.

Google has its Axion processors. Amazon has Graviton, now in its fourth generation, powering a massive portion of AWS compute. Microsoft has its Cobalt chips.

The architectural choice between Arm and RISC-V is telling. Arm is the proven option, already battle-tested in cloud environments by Amazon and others. RISC-V is the open-source upstart, offering more design flexibility and no licensing fees, but with a less mature ecosystem. The fact that ByteDance is evaluating both suggests the project is still early enough that fundamental design decisions haven’t been locked in.

For ByteDance specifically, US export restrictions on advanced semiconductors to China have created persistent uncertainty around supply chains for Chinese tech companies. Building custom silicon, even if manufactured through partners, gives ByteDance more control over its technological destiny in an increasingly fragmented geopolitical landscape.

What this means for investors

The immediate implication for semiconductor investors is nuanced. On one hand, ByteDance designing custom CPUs means potentially fewer purchases from Intel and AMD over the long term. On the other hand, custom chip design requires manufacturing partners, and the short list of companies capable of fabricating advanced chips is very short indeed.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.

ByteDance develops custom CPUs to support AI infrastructure amid chip shortages

ByteDance develops custom CPUs to support AI infrastructure amid chip shortages

TikTok's parent company is designing its own processors as CPU prices from Intel and AMD surge by up to 35%, joining a growing club of tech giants building custom silicon.

ByteDance, the company behind TikTok, is building its own CPUs. The reason is straightforward: buying them from Intel and AMD has become painfully expensive.

CPU prices from major suppliers have climbed 10-35% quarter-over-quarter, according to Reuters, and supply constraints have stretched expansion timelines to the breaking point. For a company projected to spend RMB 160 billion (roughly $22.8 billion) on AI-related capital expenditures in 2026, those price hikes add up fast.

The custom silicon play

ByteDance is currently in early-stage development, evaluating both Arm and RISC-V architectures for the new chips. The company is also seeking partnerships for design and manufacturing, which suggests it’s not trying to build a full-stack semiconductor operation from scratch.

Advertisement

The target use case is internal. These CPUs are meant to power ByteDance’s own servers and data centers, with a particular focus on supporting agent-based AI products like its Coze platform.

Nearly half of that $22.8 billion AI budget is reportedly earmarked for advanced semiconductor development.

ByteDance joins the club

ByteDance isn’t pioneering this approach. It’s following a well-worn path carved by Google, Amazon, and Microsoft, all of which have invested heavily in custom CPU designs to optimize cost and performance for AI workloads.

Google has its Axion processors. Amazon has Graviton, now in its fourth generation, powering a massive portion of AWS compute. Microsoft has its Cobalt chips.

The architectural choice between Arm and RISC-V is telling. Arm is the proven option, already battle-tested in cloud environments by Amazon and others. RISC-V is the open-source upstart, offering more design flexibility and no licensing fees, but with a less mature ecosystem. The fact that ByteDance is evaluating both suggests the project is still early enough that fundamental design decisions haven’t been locked in.

For ByteDance specifically, US export restrictions on advanced semiconductors to China have created persistent uncertainty around supply chains for Chinese tech companies. Building custom silicon, even if manufactured through partners, gives ByteDance more control over its technological destiny in an increasingly fragmented geopolitical landscape.

What this means for investors

The immediate implication for semiconductor investors is nuanced. On one hand, ByteDance designing custom CPUs means potentially fewer purchases from Intel and AMD over the long term. On the other hand, custom chip design requires manufacturing partners, and the short list of companies capable of fabricating advanced chips is very short indeed.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.