Chinese technology giants significantly increased AI capital expenditure in Q2 2026, with combined capex by Alibaba, Tencent and Baidu surging 105% year-on-year and 95% sequentially to 126 billion yuan, or approximately $19 billion. According to reports from The Economic Times, Alibaba's capital expenditure increased 75% from the previous year, while spending by Tencent and Baidu climbed 176% and 55% respectively. The surge was primarily driven by strong demand for AI inference and improved access to advanced chips, particularly Nvidia's H200 chips. As per Jefferies, the increase was mainly attributed to strong demand for AI inference—the process of using trained AI models to answer questions or generate content—and an opportunity to purchase Nvidia's H200 chips. This investment surge comes as global AI spending is projected to surpass $2 trillion in 2026, with the rapid rise of generative AI bringing the technology into everyday use for hundreds of millions of people worldwide.
Despite the substantial increase, China's capex intensity remains considerably lower than US counterparts. As reported by Jefferies, Chinese cloud companies' capex reached 176% of their cloud revenue compared with 130% for US peers in the latest quarter. However, Jefferies believes China's quarterly figure was distorted by irregular chip purchases and Tencent's heavy AI spending for internal use. Over the past four quarters, Chinese companies' capex was equivalent to 111% of cloud revenue, compared with 119% for US cloud providers, with any ratio above 100% warranting attention. The brokerage notes that China's spending remains considerably less intensive when measured against the companies' overall revenue and is also more volatile because chip purchases tend to occur when supplies become available. During the past four quarters, Chinese cloud companies spent around 16% of their revenue on capital expenditure, compared with approximately 27% for Amazon, Microsoft and Google.
Jefferies considers capital expenditure relative to cloud revenue a more useful measure, as much of the additional investment is directed towards AI infrastructure. According to the brokerage, US cloud capital expenditure still exceeded cloud revenue by 30% in the second quarter, and the gap has continued to widen. The firm is more concerned about the US because American AI models are largely closed-source, giving US hyperscalers a clearer route to monetising their infrastructure spending. Despite this advantage, US cloud spending remains stretched with ratios above 100%. The gap is wider when measured over a longer period, with US cloud companies' capex-to-sales ratio having more than doubled between the fourth quarter of 2024 and the second quarter of 2026, representing an increase of around 17 percentage points.
China may have purchased approximately 200,000 H200 chips, equivalent to about 25,000 HGX H200 servers, during the quarter, with purchases costing between $8 billion and $9 billion. Assuming ByteDance accounted for 30% of these purchases, Alibaba, Tencent and Baidu may have spent around $5-6 billion on H200 chips, accounting for 55-65% of the sequential increase in their capital expenditure. The remaining $3-4 billion was likely spent on domestically produced AI chips, with Jefferies identifying Huawei as the biggest potential beneficiary. The latest Chinese spending surge also came from a relatively low base, with China's previous peak capex-to-sales ratio being 14.5% in the fourth quarter of 2024. Even after the latest jump, the ratio has increased by about 10 percentage points from that level.
According to Jefferies, Chinese spending is expected to continue rising as inference demand increases and local technology companies purchase more domestic chips. However, the brokerage does not expect triple-digit growth every quarter or China to close the capex-intensity gap with the US soon. The firm emphasizes that investors should monitor whether capex continues to grow faster than cloud revenue over several quarters and whether demand for paid AI services can generate sufficient returns on additional investment. For China, the spending surge could benefit Nvidia in the near term and Huawei as local companies increase purchases of domestic chips. For US hyperscalers, the key question remains whether their closed-source AI ecosystems can convert rapidly rising infrastructure investment into proportionate cloud revenue. A single quarter of elevated expenditure does not necessarily indicate overspending, particularly in China, where purchases of restricted or scarce chips can produce sharp swings.
