Goldman Sachs sharply raised Alibaba's capital expenditure forecast to 460 billion yuan: explosive growth in inference demand and improved AI efficiency drive stronger revenue.

```

Goldman Sachs believes that the explosive growth in demand will continue to drive up cloud service providers' capital expenditures (Capex). Chinese internet giants are increasingly pursuing differentiated strategic paths in the AI field: Alibaba is betting on the enterprise-grade AI cloud market with its full-stack capabilities, while ByteDance is making an all-out push in consumer-grade (To-C) applications.

In its report on the 23rd, Goldman Sachs raised its capital expenditure forecast for China’s leading cloud providers, predicting that Alibaba's total Capex for fiscal years 2026 to 2028 will reach 460 billion yuan, far exceeding the company’s previous target of 380 billion yuan. The bank believes that the surge in AI inference demand is the core logic supporting this judgment, and higher computing efficiency may actually increase the conversion rate of Capex to cloud revenue, thereby accelerating revenue growth.

Meanwhile, the strategic differences among the giants are becoming more apparent. Goldman Sachs pointed out that Alibaba leads in external AI cloud revenue scale and enterprise-grade (To-B) services, with a more clearly defined commercialization path. ByteDance, on the other hand, holds the largest share in the To-C field and in daily token consumption through its chatbot “Doubao,” demonstrating its determination to explore consumer-facing AI applications.

The report believes that the current valuation of China’s major tech stocks remains attractive. The market has not yet entered an AI bubble, and the valuations of Tencent and Alibaba still have room for a discount relative to their earnings growth prospects and global peers (such as Google and Amazon). On this basis, Goldman Sachs reiterates its “Buy” rating on Alibaba and Tencent.

Improvements in AI Efficiency Fail to Curb Capex Expansion

Recently, Chinese companies have made multiple breakthroughs in AI infrastructure and computing efficiency. For example, Alibaba Cloud launched a new GPU pooling system, Aegaeon, which reportedly saves 82% of GPU resources; DeepSeek’s new OCR model can reduce token consumption for text input by 90%.

However, Goldman Sachs believes that these efficiency gains do not necessarily mean a corresponding reduction in Capex. According to data cited in the report, demand for AI inference and token consumption is growing exponentially. ByteDance recently announced that its daily token consumption exceeded 30 trillion in September, doubling from April-May, and is approaching Google’s 43 trillion. Alibaba also revealed at the Yunqi Conference that its AI inference demand doubles every 2-3 months.

Goldman Sachs forecasts that China's cloud service providers (CSPs) will see a 50% year-over-year increase in Capex in Q3 2025. Strong inference demand is the key driver for continued investment by cloud providers, while improvements in technological efficiency help these investments translate into revenue growth more effectively, forming a positive cycle.

Diverging Paths: Alibaba To-B, ByteDance To-C

Goldman Sachs’ report clearly outlines the distinct strategic layouts of China’s two tech giants in the AI field.

Alibaba is focusing on the enterprise-grade AI market, leveraging its “unique full-stack AI capabilities” to lead in external AI cloud revenue scale. On October 23, Alibaba officially launched its new Quark AI chatbot assistant service, which utilizes its most advanced closed-source Qwen model and directly competes with ByteDance’s “Doubao” and Tencent’s “Yuanbao.” Furthermore, its Quark “Zaodian” application has also made progress in multimodal video and image editing capabilities with its closed-source video model Wan2.5.

In contrast, ByteDance is focusing more on consumer-facing AI applications. Its “Doubao” chatbot has become the leader in China’s To-C market and in daily token consumption. ByteDance is accelerating the commercialization of “Doubao”—for example, seamlessly integrating Douyin e-commerce services into chats to allow in-app transactions, and adding new features such as an AI keyboard.

Acceleration in Multimodality and Commercialization

The report points out that China’s multimodal large models are making progress in global markets and are building differentiated competitive advantages with strategies such as open-sourcing, low pricing, and high speed. For example, Tencent’s “Hunyuan Image 3.0” model ranks among the top in the LMArena text-to-image model leaderboard. In terms of pricing, the report cites comparison data showing that Alibaba’s Qwen3 Max model’s output price is 40% cheaper than GPT-5/Gemini 2.5 Pro.

The global application of Chinese AI models is also increasing. The report cites Airbnb's CEO, who said the company is making extensive use of Alibaba’s Qwen model to support its customer service agents, highlighting that Chinese open-source AI models are gaining recognition in global markets.

On the commercialization front, China’s To-C applications are moving along a path similar to ChatGPT. In addition to ByteDance’s “Doubao” accelerating integration of e-commerce functions, Alibaba’s Quark has also launched a one-stop AI image and video creation platform, “Zaodian.” Goldman Sachs believes that the commercialization path for Chinese To-C chatbots is still evolving, and may ultimately be driven more by advertising revenue.

In terms of valuation, Goldman Sachs believes there is not yet an AI bubble, and its US strategists expect the American AI Capex boom to continue until 2026. According to the report, Tencent and Alibaba’s expected GAAP P/E ratios for 2026 are 21x and 23x, respectively, which, compared to Google’s 24x and Amazon and Microsoft’s 28-30x, are still at “not demanding” levels.

Risk Disclosure and DisclaimerThe market has risks, and investment should be cautious. This article does not constitute personal investment advice, nor does it take into account individual users’ special investment objectives, financial situation or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this article is at your own risk. ```