How significant is the "H20 Recovery"? The strategic differences between Tencent and Alibaba, the revenue and capital expenditure of major cloud companies

For investors focused on Chinese tech giants, Nvidia's recent plan to resume supply of the H20 chip to China is undoubtedly the market's focal point. However, JP Morgan's latest analysis has poured cold water on this enthusiasm: while it is a positive incremental factor, it is far from a "game changer."

According to news from the Wind Trading Desk, on July 16, JP Morgan published a research report stating that in the early stages of this AI race, Alibaba's "shovel-selling" model seems to create financial returns for shareholders more quickly, while it may be a short-term negative for Tencent. Previously, on July 15, Goldman Sachs published a research report that predicted Alibaba and Tencent's capital expenditures would bottom out in the second quarter of 2025 and begin to recover quarter-on-quarter in the second half of the year.

JP Morgan believes that investors should not overestimate the direct boost to cloud vendors' revenues from the resumption of H20 supply, but should instead take a deeper look at each company's AI strategies, capital expenditure plans, and their actual impact on profit margins:

• Short-term financial impacts are significantly differentiated: Due to vastly different strategic paths, this news is a short-term positive for Alibaba but a short-term negative for Tencent. As a "shovel seller," Alibaba's AI investments can be monetized more quickly through cloud services; whereas Tencent's focus on free AI application development will directly erode its short-term profit margins due to huge capital expenditures and operating costs.

• Capital expenditure (Capex) is a key sentiment indicator: Don't expect cloud vendors' revenues to surge dramatically; the real highlight is the steady growth of capital expenditures. JP Morgan predicts that after the inventory boom in 2024, the industry's capital expenditure growth rate will slow in 2025 but will still maintain stable growth, which is a core financial indicator shaping investors' confidence in the industry.

• Demand has not "exploded exponentially": The real demand for AI computing power is "moderate growth" rather than "exponential explosion," especially among external public cloud customers lacking native AI killer applications. Therefore, the recovery of chip supply will not immediately translate into a significant upward revision of cloud vendors' revenues.

Strategic Differentiation: Alibaba's "Shovel Seller" vs. Tencent's "Application Developer"

Although both Alibaba and Tencent are involved in external sales of computing power and internal development of AI applications, their strategic focuses are fundamentally different, leading to disparities in the financial impact of the H20 supply recovery.

JP Morgan believes that the resumption of H20 supply is positive for Alibaba, helping it better serve external developers, and in the early stages of AI development, Alibaba's return on AI investment should be higher than Tencent's:

"Shovel Seller" Alibaba:

• Alibaba is more inclined to play the role of a "shovel seller" in the AI "gold rush," pursuing maximization of short-term financial returns, as its AI capital expenditures can be directly monetized through its profitable cloud business, converting into cloud revenue.

• This "cloud-first" strategy means that investing in AI can bring about incremental profits, albeit not significant. It is predicted that Alibaba Cloud's revenue growth for the fiscal year 2026 will be 22%. Data shows that the monthly active users of Tencent's Yuanbao APP surged from 2.9 million in December 2024 to 42 million in March 2025, stabilizing around 40 million in June, ranking among the top three in the industry. At the same time, Tencent has embedded Yuanbao into its ecosystem, including WeChat, Tencent Docs, and QQ Browser.

The research report points out that Tencent's strategy of building an AI ecosystem is beneficial in the long run, but currently, as these AI features are provided to users for free, the huge computing power consumption and operational investment have a negative impact on the company's profit and loss statement in the short term:

"Application Developer" Tencent:

• Tencent is focused on the long term but sacrificing short-term profits, with the strategic focus on becoming a "generative AI application developer," prioritizing the needs of internal AI applications over external clients.

• It is expected that AI capital expenditures will cause about a 2 percentage point dilution in profit margins in 2025. The depreciation of its computers and other operational equipment will rise from 2.9% of total revenue in 2024 to 4.5% in 2025, which means without direct monetization methods to compensate, depreciation alone will bring a negative impact of 1.7 percentage points on profit margins.

• AI operating expenses (Opex), including talent recruitment (such as providing 28,000 internship positions over the next three years) and user acquisition costs (it was reported that the marketing expenses for Yuanbao App alone reached 281 million RMB in February), are expected to have an additional 1 percentage point impact on profit margins in 2025.

Capital Expenditure Outlook: Growth Slows, But Remains a Core Observation Indicator

Capital expenditure is a key measure of cloud vendors' commitment to AI investment.

JP Morgan predicts that after experiencing a peak in inventory accumulation in the second half of 2024, the overall capital expenditure growth in the industry will slow down in 2025:

• Alibaba: It is predicted that capital expenditure for the fiscal year 2026 (ending March 2026) will reach 109 billion RMB, a year-on-year increase of 25%. Alibaba reported 16 billion RMB in capital expenditure in the most recent quarter, and JP Morgan believes its annual plan is progressing steadily.

• Tencent: It is expected that its capital expenditure in 2025 will reach 94 billion RMB, a year-on-year increase of 22%.

Goldman Sachs' forecast provides a more detailed timeline, expecting that the capital expenditures of Alibaba and Tencent will bottom out in the second quarter of 2025 and begin to recover quarter-on-quarter in the second half of the year.

This does not contradict JP Morgan's judgment of annual growth slowing down, but rather jointly depicts a "first suppress then rise" annual expenditure scenario.

Limited Benefits: Chip Supply Recovery Is Not a "Game Changer"

JP Morgan clearly points out that the market should not view the recovery supply of Nvidia's H20 chip as a significant turning point for the development of generative AI in China.

Firstly, due to large cloud service providers having stockpiled AI chips in large quantities over the past few quarters, the H20 restrictions announced in early April did not cause widespread supply tightness in the industry. Secondly, since the release of the DeepSeek large language model in January 2025, the market demand for GPU computing power and the consumption of AI functions have shown a "moderate growth" trend, rather than the "exponential growth" that the public imagines. This moderate demand growth mainly comes from external customers of public clouds.

JP Morgan observed that the leading companies that extensively use GPUs for AI (such as ByteDance, Tencent, and Baidu) are using GPUs they purchased themselves, rather than renting GPUs from public clouds. Therefore, while this demand for GPUs does exist, it will not translate into "revenue" for cloud service providers like Alibaba Cloud and Tencent Cloud.

Most external customers of public clouds are small and medium-sized AI application developers. Although they have widely adopted generative AI, their consumption growth is relatively steady