China’s largest internet companies are spending heavily on artificial intelligence while accepting a near-term hit to cash flow. UBS sees a different stage of the cycle ahead. In a September 2 interview with the South China Morning Post, Kenneth Fong, UBS’s head of China internet research, said platforms with vast data, distribution networks, and user bases could take a larger share of AI profits in two to three years.
The argument is not that today’s investment bill has disappeared. It is that the profit pool may migrate as constraints change. Fong said the current advantage lies upstream, where capacity remains scarce. If that shortage eases, he expects the balance of pricing power to shift toward consumer and enterprise platforms that can distribute AI products at scale and draw on large pools of data and users.
That forecast arrives as spending is already testing the financial discipline of China’s biggest tech groups. SCMP reported that Tencent almost tripled second-quarter capital expenditure to 52.8 billion yuan, and its cash flow fell 13.8 billion yuan below zero in the first such quarter reported by SCMP. Alibaba, which owns SCMP, reported a June-quarter free-cash outflow of 44.7 billion yuan after a quarterly outlay of 67.7 billion yuan.
Infrastructure Spending Creates Today’s Pressure Point
The figures illustrate the short-term tension in the Chinese AI cycle. A company can improve model offerings, cloud capacity, and agent products while still showing weaker near-term free cash flow because the necessary computing and infrastructure spending comes first. In UBS’s framing, the question is not whether that expenditure is costly. It is whether the firms that pay for it will eventually convert their installed infrastructure, customer relationships, and distribution channels into recurring revenue.
Tencent and Alibaba are relevant examples because both own large consumer ecosystems as well as cloud and enterprise operations. Their spending is not occurring in isolation from products. Tencent has been extending its model and agent portfolio, including the Hy4 release for China’s AI agent race. Alibaba has also continued to expand Qwen across cloud services, developer tools, and workplace products.
The investment figures, however, should not be read as a direct measure of AI profitability. Capital expenditure supports more than one activity, and the available SCMP report does not assign each yuan of spending to a particular model, cloud region, or product line. It instead captures a broader strategic choice: companies are prioritizing capacity and AI infrastructure at a time when investors remain cautious about macroeconomic conditions and spending intensity.
This distinction matters because revenue from an AI product does not necessarily translate quickly into free cash flow. Training, inference, data-center operations, product engineering, and customer acquisition can all occur before a platform finds its most durable pricing structure. UBS’s two-to-three-year window forecasts that transition, not a company earnings target.
Data and Distribution Are UBS’s Case for Downstream Gains
Fong’s thesis turns on the idea that supply constraints currently favor infrastructure providers, but that a more abundant capacity environment would make distribution and user access more valuable. Internet platforms already have channels to place AI features in search, messaging, commerce, cloud software, content, and workplace workflows. They also have data generated by those services, subject to the legal and product limits that govern its use.
This is why the analysis is broader than a prediction about one model vendor. A successful foundation model can attract developers, but a large platform can also bundle AI into existing services or use it to improve retention, advertising tools, merchant operations, and cloud offerings. The eventual winner, in UBS’s view, is likely to be determined partly by which companies can turn AI capability into distribution rather than simply by who spends the most on computing first.
Alibaba’s recent activity reflects that distribution challenge. Its QwenWork expansion placed workplace agents into a competitive field where users already have established collaboration and cloud tools. Tencent faces a similar task with agents attached to products such as WorkBuddy, WeChat, and enterprise software. Neither company has disclosed a single, standardized measure of AI profit that would make the comparison simple.
The UBS forecast therefore offers a useful lens rather than a definitive conclusion. It identifies data, users, and access to customers as prospective advantages after capacity becomes less scarce. It does not establish when that moment will arrive or how much of the resulting profit will flow to any one company.
Forecasts Are Not Earnings Guidance
The report’s timing is important. Investors have watched Chinese technology shares react sharply to signs of AI spending, model progress, and macroeconomic weakness. A forecast that profits may emerge in two to three years addresses the logic of investment, but it cannot remove execution risks in the intervening period.
The hurdles are visible in the same figures. Tencent’s capex increase and negative free cash flow show the financial cost of scaling infrastructure. Alibaba’s outlay and free-cash outflow demonstrate the same problem from another direction. Both groups must keep models and cloud services competitive while finding uses that customers will pay for at volumes large enough to justify the upfront expense.
There is also a policy and technology dimension. Regulators are pressing AI providers to meet labeling, content, and data obligations as products move into more consumer and enterprise settings. Competition is intensifying among China’s technology groups, model developers, and cloud suppliers. A platform’s historical user base may be an advantage, but it does not guarantee that a user will adopt a new agent or that a business customer will standardize on one cloud service.
For now, UBS’s conclusion should be understood precisely: the firm expects a possible transfer of AI pricing power from constrained upstream infrastructure toward downstream platforms over a two-to-three-year horizon. That is a view about the structure of the market, not evidence that the transfer has already occurred.
China’s leading companies are still laying the financial and technical foundation for that outcome. Their next results will show whether higher spending is producing more than model headlines and product launches. The central test is whether the firms with the greatest reach can make AI an everyday paid service across the businesses that already connect them to hundreds of millions of users.