BREAKINGVIEWS-China's slim AI budgets are self-reinforcing​​​

By Robyn Mak

HONG KONG, Oct 9 (Reuters Breakingviews) - Artificial intelligence will transform industries and economies. Yet while fears are growing in the United States that spending on advanced chips and data centres is inflating an investment bubble, China may end up with the opposite problem: not spending enough to make breakthroughs at the frontier.

Technology giants in the People's Republic are rapidly loosening their purse strings. Alibaba 9988.HK shares surged to their highest level in nearly four years last month when boss Eddie Wu said demand for AI infrastructure "far exceeded" the company's expectations.

Alibaba now looks set to be one of China's most aggressive AI investors: Wu wants to establish the $430 billion company he leads as a "full-stack AI service provider" and one of the only "five or six" supercomputing platforms in the world. He's aiming for "artificial general intelligence", or a level of AI that matches or surpasses human cognitive abilities. That lofty goal is common among U.S. tech giants but in China it is a rare ambition, shared by cross-town upstart Deepseek.

A nalysts at Morningstar reckon Alibaba's capital expenditure will average roughly 15% of revenue over the next three years, implying a total of around $71 billion. That is less than the absolute and relative amounts Alphabet GOOGL.O, Amazon.com AMZN.O and Microsoft MSFT.O will each spend this year alone, per Visible Alpha. The trio will splash out an average of $94 billion, or 21% of sales.

Other major Chinese firms including TikTok-owner ByteDance and Meituan 3690.HK are upping their investments too. The $800 billion social media and gaming group Tencent 0700.HK flagged capital expenditure to be in a "low teens" percentage of revenue this year - up from less than 5% it devoted just two years earlier.

Overall, AI capital expenditure in the People's Republic may hit 700 billion yuan, or $98 billion in 2025, Bank of America analysts estimated in August. But that's just one-fifth of annual spending consultancy Bain & Company expects to see in the United States each year over the rest of the decade.

There are several good reasons why the Chinese are spending less.

In the People's Republic, enterprises have been slow to adopt IT solutions. In cloud, for example, where most AI models exist, the world's second-largest economy accounts for just a tenth of global sales, Jefferies research notes. For the current fiscal year, Microsoft's Azure and Amazon's AWS units are forecast to each bring in over $100 billion of sales, easily outstripping the $21 billion at Alibaba's Cloud Intelligence Group, Visible Alpha estimates show.

By 2030, the U.S. duo will generate more cloud computing sales combined than the estimated $440 billion for the entire Chinese market, according to Visible Alpha and Bank of America figures. It helps that Microsoft and Amazon have a strong presence in Europe and the Middle East b ut Alibaba has global ambitions too. It plans to open its first data centres in Brazil, France and the Netherlands, with additional facilities planned for Mexico, Japan, South Korea, Malaysia and Dubai over the coming year.

In China, unlike in the United States, there also is a race to the bottom on prices that will cap returns on AI investments even as productive use cases emerge. It's not just in cloud computing, where market leader Alibaba is battling it out with a dozen or so rivals ranging from telecoms-to-chips champion Huawei to state-owned mobile carriers. Low-cost and free chatbots and agents are available to companies and individual users, proliferating in what one Tencent executive called "a war of a hundred models".

Take Alibaba's popular open-source Qwen model. In May, the company slashed the Qwen-Long model price for developers by 97% to 0.0005 yuan per a thousand tokens. A month later, ByteDance cut its Doubao model prices by 63% to as low as 2.6 yuan, roughly $0.35, per one million tokens, Reuters reported. And more recently, DeepSeek saidit was halving prices on its software tools. To compare, OpenAI's GPT-5 charges an average, or "blended", $5.63 per one million tokens, assuming a ratio of one input token to one output token.

U.S. export controls on chips also mean Chinese companies are relying more on less-powerful domestic alternatives for computing power. That means businesses have to buy more semiconductors to achieve the same outcomes as Western peers, or invest in finding creative workarounds and shortcuts. Some, like Alibaba and Huawei, are pouring resources into developing their own processors.

These hardware constraints, plus the price wars and slow AI adoption, all amount to profit killers. Alibaba's cloud operating profit margin, using the company's preferred metric of earnings before interest, tax and amortisation, is expected to reach 9.2% in 2027, an increase of less than one percentage point from this year. That's less than half the operating profit margin at Alphabet's cloud solutions unit, and less than a quarter of Amazon AWS', according to Visible Alpha.

Little wonder shareholders are less enthusiastic about the earnings potential of China's AI companies than their U.S. peers. Even after more than doubling this year, Alibaba's stock fetches less than 20 times forward earnings, per LSEG, well below the average 27 times for the Big Four U.S. hyperscalers.

In addition, China's conflicting goals may be holding back how much companies spend. Policymakers want to achieve technological self-sufficiency and be an AI leader. But they are also warning against "disorderly competition" and a "follow-the-crowd" approach in the sector. To avoid the overcapacity prevalent in industries from solar panels to electric vehicles, Beijing wants to develop AI in a coordinated way. Alibaba, Tencent and others bruised by a years-long regulatory crackdown are likely to toe the line - a sharp contrast to, say, Sam Altman's OpenAI's inking a web of deals and tie-ups worth $1 trillion to realise his ambitions.

These pressures explain why most Chinese tech firms are opting to quickly commercialise pragmatic applications across factories, schools, hospitals, cities and more, rather than spending larger sums to achieve higher AI capabilities.

China's measured approach to spending means it is less likely than the United States to face an AI bubble-induced economic crash. But the Asian behemoth risks falling behind in shaping this generation's most important technology.

Follow Robyn Mak on X.