ByteDance is moving deeper into AI hardware through a collaboration with Broadcom, according to Reuters sources. The TikTok parent company is working with the US chipmaker on a custom AI processor designed around both performance needs and US export regulations.
The project comes as generative AI becomes more important to social media products and as access to advanced AI chips becomes a competitive issue. For ByteDance, the central question is practical: how to get enough compute for AI work without depending entirely on the same supply channels as larger international rivals.
A custom chip built around constraints
The planned processor is a 5-nanometer chip. It is intended to comply with US export regulations and is expected to be manufactured by TSMC in Taiwan.
That combination matters because ByteDance faces political restrictions that limit its access to AI chips compared with international competitors. A custom processor gives the company a route to high-performance AI infrastructure that is shaped by those restrictions from the start.
The goal is not only access. The project also seeks to reduce costs. AI systems require large amounts of compute, and chip spending can become a major pressure point when a company is researching generative AI and bringing that technology into consumer products.
Production is slated to begin next year. The source article also notes that ByteDance and Broadcom already collaborate in other areas, giving the new AI processor effort an existing business relationship to build on.
Why ByteDance needs more AI compute
ByteDance is researching generative AI and implementing the technology in TikTok. One recent example mentioned in the source is digital avatars.
Generative AI is compute-intensive. If these capabilities become central to social media services, companies with stronger access to AI hardware can move faster, test more ideas, and support more demanding AI features. The source frames this as a risk for ByteDance: without enough compute, the company could fall behind.
Reuters reports that ByteDance spent $2 billion on Nvidia chips last year. That is described as a small fraction of what social media competitor Meta has invested. Meta is also identified as one of Nvidia's biggest known customers alongside Microsoft.
This context shows why a custom AI chip project is strategically useful. ByteDance can still buy chips, but a dedicated processor could help it secure supply and manage costs in a market where the most capable hardware is both essential and difficult to obtain.
Rivals are building their own paths
ByteDance is not the only company trying to reduce dependence on outside AI chip supply. The source article points to several competitors that are also developing custom hardware or considering deeper chip strategies.
- Meta's "Artemis" chip is set to enter production this year and will be used in the company's data centers for AI model inference.
- Microsoft has developed Azure Maia for AI work and Azure Cobalt, an Arm-based CPU, for its cloud infrastructure.
- OpenAI is considering developing its own AI chips and is working with RainAI, a company creating a neuromorphic processing unit (NPU).
Mark Zuckerberg recently announced that Meta plans to deploy 340,000 Nvidia H100 GPUs and around 600,000 total GPUs by year-end. That scale illustrates the compute race around AI services, especially for companies that expect AI to become a larger part of their products.
The source also says OpenAI CEO Sam Altman is reportedly planning to raise significant funding to build a global chip infrastructure as an alternative to Nvidia. Taken together, these moves show that leading AI and social media companies are not treating chips as a background procurement issue. They are becoming part of core strategy.
The bigger hardware shift
The ByteDance-Broadcom project fits into a wider pattern: companies with large AI ambitions want more control over the hardware that powers their models and products. Nvidia remains central to the market described in the source, but major customers are also looking for alternatives, supplements, or custom designs.
For ByteDance, the pressure is sharper because access is not only a matter of budget. Political restrictions limit the Chinese company's access to AI chips compared with international rivals. That makes a compliant custom chip more than a cost-saving measure; it is a way to keep AI development moving inside a constrained environment.
The success of the effort will depend on whether the chip can deliver useful performance, reach production as planned, and support the AI workloads ByteDance needs for products such as TikTok. The source does not provide technical performance claims, so the most grounded takeaway is strategic rather than benchmark-based.
ByteDance is trying to secure compute before it becomes an even bigger bottleneck. Broadcom brings chipmaking expertise, TSMC in Taiwan is expected to manufacture the processor, and the design is being shaped to fit US export regulations. In a market where AI features increasingly depend on hardware access, that combination could become important for ByteDance's next phase of generative AI work.