China’s tech giants have doubled capital spending this year as they splurge on artificial intelligence infrastructure, despite US sanctions designed to limit the country’s progress in the crucial technology.
Alibaba, Tencent and Baidu had combined capital expenditure of Rmb50bn ($7bn) in the first half, compared with Rmb23bn a year earlier. The groups said the focus was on buying processors and infrastructure related to powering the training of large language models for AI, both their own models and those of others.
TikTok parent ByteDance has also increased AI-related spending, backed by a cash pile of more than $50bn and with the benefit of being privately held and relatively free of investor scrutiny, according to two people familiar with the matter.
“We’ll continue to invest in R&D and AI capex to ensure the growth of our AI-driven cloud business,” Alibaba chief Eddie Wu told investors this month. “It’s simply because we see a lot of unmet demand from many clients.”
Alibaba is buying processors to train its Tongyi series of AI models and rent out computing power to others. The Chinese tech giant’s capex in the first half totalled Rmb23bn, up 123 per cent from a year earlier.
“What we see when we’re making these kind of capex investments, as soon as we get a server up, that server is essentially instantly running at full capacity,” Wu said. “We can expect to see a very high ROI [return on investment] over these next quarters.”
Sales for the group’s cloud business accelerated during the second quarter, rising 6 per cent from a year earlier. Alibaba said AI-related product revenues had more than doubled year on year.
The spurt is in part being driven by investments in Chinese AI start-ups structured to bring in customers. Just under half of the $800mn it put into AI start-up Moonshot in February came in the form of vouchers to buy its cloud services.
While US export controls are cutting off access to Nvidia’s leading AI processors such as the H100 and upcoming Blackwell series, China’s tech giants can buy lower-performance processors such as Nvidia’s H20, which was designed not to exceed computing power thresholds set by Washington.
Analysts expect Nvidia to ship more than 1mn of the processors to Chinese tech groups in the coming months, at $12,000 to $13,000 a unit. ByteDance was a major customer, the two people familiar with the matter said.
Dylan Patel of chip research group SemiAnalysis estimates TikTok’s parent has been purchasing hundreds of thousands of H20s for its data centres in China, while also spending heavily to work with partners and build computing infrastructure in Johor, Malaysia.
“ByteDance is the largest Chinese purchaser of AI because they are investing heavily in China, Malaysia, and buying from US clouds,” said Patel.
Social media and gaming giant Tencent said capex rose to Rmb23bn in the first six months, up 176 per cent from a year earlier, in part “driven by investment in GPU and CPU servers”.
James Mitchell, chief strategy officer, said the company’s cloud business had benefited from a growing need to rent graphics processing units but on a smaller scale than the boom being experienced by US peers.
“You don’t have the same multitude of extremely well-funded start-ups trying to build large language models on their own in China. There are many small companies, but they’re capitalised for $1bn, $2bn,” he said. “They’re not capitalised at $10bn or $90bn” like in the US.
One person familiar with Tencent’s investment strategy said it was writing smaller cheques for AI groups because of lingering concerns about Beijing’s regulatory stance.
China’s longtime AI leader Baidu has been the most restrained on capex, spending Rmb4.2bn in the first half, up 4 per cent from a year earlier.
Overall, China’s Big Tech capex outlays still lag far behind those of their American peers. Alphabet, Amazon, Meta and Microsoft spent $106bn during the first half and pledged further investment in the coming months.
