ByteDance is significantly expanding its investment in artificial intelligence infrastructure, reflecting a broader acceleration by Chinese tech companies as they try to close the gap in the global AI arms race.

According to the South China Morning Post, ByteDance has raised its planned 2026 AI capital expenditure to more than 200 billion yuan, or about $30 billion, at least 25% above the 160 billion yuan plan initially discussed late last year.

The paper cited people familiar with the matter as saying the higher spending is mainly driven by the company’s continued deep bet on AI and rising memory chip costs. Notably, ByteDance has also increased the share of domestic AI chips in its procurement plan as part of the budget adjustment, a move aimed at managing geopolitical risks and aligning with Beijing’s push to promote homegrown semiconductors.

The expansion plan comes as ByteDance is also accelerating its overseas buildout. This week, Thailand’s government approved a $25 billion data infrastructure investment plan by ByteDance, a major increase from the previously announced initial plan of $8.8 billion and the company’s largest single project announced to date in Southeast Asia. Last month, the company also announced an additional 1 billion euros for data center construction in Finland, extending its digital infrastructure expansion in Europe.

ByteDance’s spending increase is substantial, but the gap with U.S. tech giants remains wide. Last week, Google and Microsoft each said annual capital expenditure would reach about $190 billion, Meta raised its capex forecast for this year to $145 billion, and Amazon kept its full-year guidance unchanged at about $200 billion.

According to a report released last December by UBS China internet analyst Wei Xiong, China’s internet giants spent a combined roughly 400 billion yuan on AI capital expenditure in 2024, only about one-tenth of their U.S. peers’ total. ByteDance’s newly raised 200 billion yuan budget, or about $30 billion, is already among the largest in China’s internet sector, but it still trails the spending of a single top U.S. tech company by several times.

China’s AI Ecosystem: A Collaborative Culture and Technological Self-Reliance Drive the Catch-Up

As capital spending accelerates, outside observers are also offering a new lens for understanding how China’s AI industry is catching up.

Nathan Lambert, a researcher at the Allen Institute for AI, recently completed an intensive 36-hour tour of Beijing’s AI circles, speaking in depth with frontline researchers from leading organizations including Zhipu AI, Moonshot AI, Meituan, Xiaomi, and 01.AI. In a long essay published afterward, he wrote that China’s AI labs present an ecosystem sharply different from Silicon Valley’s: competitors show mutual respect, and the atmosphere is far more collaborative than adversarial.

Lambert observed that ByteDance is one of the few major model players in China that has stuck with a closed-source approach, making it a focus of industry attention. DeepSeek, meanwhile, is widely seen as the lab with “the best research taste,” earning broad respect from peers. He also noted that non-AI-native companies such as Meituan and Ant Group are developing their own large models, based on a simple logic: large language models will become core to future technology products, so companies need to control them themselves.

On computing power, Lambert said NVIDIA remains the gold standard for training, while chip shortages are constraining the development speed of almost every Chinese lab. That assessment echoes ByteDance’s strategic move to increase its budget for domestic chips.

Lambert acknowledged at the end of his essay that the trip left him feeling he knew even less about China’s AI sector, and increasingly worried about whether the U.S. can maintain its lead in open-source models. “When I looked up from my laptop, I could always see clusters of cranes on the horizon,” he wrote. “That is clearly of a piece with China’s open-source spirit.”