ByteDance is moving forward with an internal artificial intelligence chip project aimed at handling AI inference tasks and has held talks with Samsung Electronics over manufacturing, two people familiar with the matter said. The company expects to receive sample chips by the end of March, the sources added.
According to those briefed on the plans, ByteDance intends to produce at least 100,000 units of the new processor this year. One source said the company is exploring a phased scaling approach that could increase output to as many as 350,000 units over time. Negotiations with Samsung reportedly include terms that would provide access to memory chip supplies, which one source said are unusually constrained amid the global build-out of AI infrastructure - a factor that makes the partnership potentially valuable.
A ByteDance spokesperson issued a statement saying that information about the company’s in-house chip project is inaccurate, without providing further detail. Samsung declined to comment.
The reported effort would mark a notable step in ByteDance’s longer-running push to develop chips tailored to its AI workloads. The company began hiring chip-related personnel at scale at least as far back as 2022 and has since expanded its investments across AI infrastructure and model development.
Earlier collaboration discussions have also been described. ByteDance has worked with a U.S. chip designer on an advanced AI processor, with previous plans indicating manufacturing could be outsourced to Taiwan Semiconductor Manufacturing Company. The current initiative carries the internal codename SeedChip, and sits within a broader strategy that the company has described as channeling resources into AI from chips to large language models.
Seed, a unit founded in 2023 to develop AI models and promote their application across the business, is part of that effort. One source said ByteDance plans to spend more than 160 billion yuan on AI-related procurement this year. The same source added that more than half of that budget is earmarked for purchasing Nvidia chips, including H200 models, while also funding the in-house chip program.
Within the Chinese technology sector, several companies have already moved to design or release their own AI processors. The report noted that Alibaba introduced its Zhenwu chip for large-scale AI workloads recently, and that Baidu both sells chips to outside clients and intends to list its chip unit Kunlunxin in the near term. Those moves place ByteDance behind some peers in bringing a self-developed processor to market, according to the sources.
ByteDance executives have signalled the breadth of the company’s AI push internally. One executive, Zhao Qi, told employees at a January all-hands meeting that the firm’s AI investments would benefit all divisions. Zhao, who oversees the Doubao chatbot and its overseas counterpart Dola, acknowledged that the company’s AI models trail global leaders such as OpenAI, but said the company would continue to back AI development throughout the year, a person briefed on the meeting said.
The chip project is part of a calculated allocation of engineering and procurement resources within ByteDance’s broader strategy to apply AI across its products, which span short-form video, e-commerce, and enterprise cloud services. While the company has not yet launched its own processor commercially, the current manufacturing discussions and planned sample timeline outline a concrete near-term pathway for testing and initial volume production.
Summary of developments
- ByteDance is developing an AI inference chip and has held manufacturing talks with Samsung.
- The company expects sample chips by end-March and plans at least 100,000 units this year, with a possible ramp to 350,000.
- Negotiations reportedly include access to constrained memory supplies; ByteDance plans significant AI procurement spending, including Nvidia H200 purchases.