ByteDance is planning to boost its artificial intelligence capital spending to over 200 billion yuan (about $28 billion) this year — a more than 25 percent increase from the 160 billion yuan preliminary budget discussed late last year. And according to a report from the South China Morning Post, a large chunk of that money is going to domestic AI chips.
Meanwhile, Chinese AI startups like Zhipu AI are advancing their own chips and models, recently releasing their GLM-5V-Turbo model, which competes with offerings from the likes of Anthropic.
The direct reason for the budget increase is rising memory chip costs and the expansion of ByteDance's AI business. But there's a key backdrop to this shift in computing procurement: although the US government has approved Nvidia to export its H200 chips to China, Chinese regulators have not yet given the green light for local companies to import them. So to avoid geopolitical risk and respond to policy signals, buying more domestic chips has become the pragmatic choice.
At the same time, ByteDance's overseas computing infrastructure buildout is accelerating. This week, the Thai government approved a massive $25 billion data infrastructure investment from ByteDance — far exceeding the company's original $8.8 billion five-year plan. Last month, it also announced an additional €1 billion data center investment in Finland.
For scale: last year, all Chinese internet giants combined spent about 400 billion yuan (~$56 billion) on AI capital expenditure. This year, Google or Microsoft alone will spend roughly $190 billion each.