ByteDance has opted to increase its capital expenditure to RMB 1.6 trillion by 2026. This unlisted Chinese technology company is making a more aggressive bet on the future.
According to the Financial Times, ByteDance plans to expand its hundreds of billions of dollars in investment in the AI sector next year. This trend is becoming increasingly evident as China’s leading technology groups strive to catch up with their American competitors.
According to two people familiar with the matter, the Beijing-based tech company has preliminarily planned to reach a capital expenditure of RMB 160 billion (approximately USD 23 billion) in 2026. However, ByteDance did not respond to requests for comment.
This figure will exceed its RMB 150 billion investment in AI infrastructure this year. About half of the amount will be used to procure advanced semiconductors for developing AI models and applications, the sources said.
They also stated that, despite whether China's top technology groups can continue to obtain$NVIDIA (NVDA.US)$They also stated that despite uncertainties over whether China's top tech groups can continue to access chips, ByteDance has still allocated a budget of RMB 85 billion for spending on AI processors next year.
The Financial Times reported that ByteDance is one of the largest builders of AI infrastructure in China, aiming to become a global leader in the field.
However, compared to major U.S. tech companies, its scale of investment remains relatively modest.$Microsoft (MSFT.US)$、 $Alphabet-A (GOOGL.US)$ 、$Amazon (AMZN.US)$and$Meta Platforms (META.US)$This year, they have collectively invested more than USD 300 billion to build data centers supporting AI models and products.
U.S. chip export control measures have led ByteDance,$Alibaba (BABA.US)$among other companies, to shift toward developing models with lower computational demands, lower costs, and higher efficiency.
This month, U.S. President Trump lifted a ban, allowing NVIDIA to sell its H200 processors to "approved customers" in China. Although the performance of these chips is lower than its most advanced products, related sales remain uncertain.
A person familiar with the plans said that if the sales are approved, ByteDance and other Chinese technology groups have expressed interest in purchasing large quantities of H200 chips. The individual added that ByteDance plans to place a trial order for 20,000 H200 chips, with each unit potentially costing around $20,000.
According to multiple sources, if ByteDance is able to purchase more H200 chips without restrictions, it may significantly increase its capital expenditure scale by 2026.
Meanwhile, the company continues to invest billions of dollars overseas in leasing data centers for training AI models and serving customers outside of China. Such leasing agreements are typically not counted as capital expenditures but are classified as operating costs.
Although ByteDance’s open-source model “DouBao” still lags behind Alibaba's “Tongyi Qianwen” and DeepSeek in independent benchmark tests, the company dominates in consumer-facing AI applications.
According to data from analytics firm QuestMobile, the “DouBao” chatbot has surpassed DeepSeek in terms of monthly active users and downloads, making it the most popular product in China. Meanwhile, ByteDance is also fiercely competing with Alibaba in the enterprise market through its Volcano Engine cloud services.
Goldman Sachs stated that these products have made ByteDance’s AI services the most widely used AI services in China. Analysts at the investment bank also noted that in October, ByteDance's daily token usage surged to over 30 trillion, compared to Google’s 43 trillion during the same period—a metric used to gauge consumer adoption of AI services.
A ByteDance investor stated: "With$Alibaba (BABA.US)$ 、 $TENCENT (00700.HK)$Compared to other large Chinese technology companies such as [omitted], ByteDance’s status as a privately held company provides a significant advantage. This allows it to make aggressive investments in the AI field more flexibly while adhering to long-term strategies.
Editor/Doris