Goldman Sachs: The narrative of AI infrastructure in China is reignited, with a surge in token consumption for enterprise-level large models, and Alibaba is the core beneficiary

With the explosion of enterprise-level application demand, the investment narrative for AI infrastructure in China is being reignited.

Goldman Sachs pointed out in a recent report that, benefiting from the accelerated adoption of large models by enterprises and the continued strong demand for computing, full-stack cloud vendors represented by Alibaba are ushering in new growth opportunities.

Data shows that the actual deployment and application of AI in the business sector is significantly accelerating. The daily Token consumption of enterprise-level large models in China reached 10.2 trillion in the first half of 2025, a staggering increase of 363% compared to the second half of 2024.

Goldman Sachs believes that Alibaba, with its leading model capabilities, 47% share of the Chinese public cloud market, and diversified chip supply, is in a favorable position and has room for international expansion.

Based on an optimistic outlook for the industry, Goldman Sachs raised Alibaba's target price from $163 to $179, maintaining a "Buy" rating. Goldman Sachs specifically increased its valuation of Alibaba Cloud from $36 per ADS to $43 and raised its growth expectations for Alibaba Cloud in the second to fourth quarters of fiscal year 2026.

Goldman Sachs analysts Ronald Keung and others emphasized that Chinese cloud vendors have made progress in self-developed inference chips and adopted a "multi-chip strategy," which means that the growth of the Chinese AI cloud industry "is no longer solely reliant on overseas chip supply." This shift, coupled with a strong capital expenditure outlook, brings compound growth potential to the industry.

Surge in Enterprise Adoption Rate, Token Consumption Soars

Chinese enterprises are embracing generative AI at an unprecedented pace, which is the core driving force supporting the infrastructure narrative. According to Frost & Sullivan data, the total daily Token consumption of enterprise-level large models in China reached 10.2 trillion in the first half of 2025, a staggering increase of 363% quarter-on-quarter.

Among numerous model providers, Alibaba, ByteDance, and DeepSeek have become the top three popular choices for Chinese enterprises when selecting general large models.

Goldman Sachs cited a report from Omdia, indicating that Alibaba has successfully penetrated most of the Chinese Fortune 500 companies deploying generative AI by 2025, occupying the leading position in this market segment. This business model, which charges based on API calls and Token usage, places large cloud service providers with leading AI models and agency capabilities in the most advantageous position.

Infrastructure Narrative Reignited, Multi-Chip Strategy Breaks Bottlenecks

Goldman Sachs believes that the investment narrative for AI infrastructure in China has been reignited after Alibaba announced better-than-expected cloud business revenue and capital expenditure. The firm predicts that capital expenditure for Chinese cloud service providers (CSPs) will grow by 39% year-on-year in the third quarter of 2025, providing the hardware foundation for the continued growth of AI cloud revenue.

At the same time, the evolution of the chip supply landscape has also enhanced market confidence. Goldman Sachs believes that this diversified chip supply strategy is reshaping the development prospects of the Chinese AI cloud industry.

**Based on the above positive trends, Goldman Sachs reiterated its "Buy" rating on Alibaba and raised its 12-month SOTP (Sum of the Parts) target price to $179. Goldman Sachs raised its growth expectations for Alibaba Cloud in the second to fourth quarters of fiscal year 2026 from 28%-30% to 30%-32%, reflecting its latest AI full-stack products, robust computing demand, and capital expenditure outlook **

The report compares the current market phase in China with the U.S. market two years ago (at the end of 2022 when ChatGPT was released), when the market rewarded companies that proposed "building good infrastructure will naturally attract customers" and increased capital expenditures.

Goldman Sachs analysts believe that Alibaba, with its leading model capabilities, a 47% share of the Chinese public cloud market (according to IDC reports), and a diversified chip supply, is in a favorable position and has room for international expansion. Investors are closely watching the upcoming Cloud Habitat Conference in Hangzhou, looking forward to updates on Alibaba Cloud and Alibaba's AI progress.

Giants Spark an Arms Race, but Commercial Monetization Still Has a Long Way to Go

Beyond the infrastructure level, China's AI models and application layers are also developing rapidly, with the report listing several key advancements:

Alibaba: Released the next-generation model architecture Qwen3-Next on September 12, claiming its performance is 10 times that of the previous generation, while the construction cost is only 1/10. Among them, the Qwen3-Next-80B-A3B model with 80 billion parameters runs 10 times faster than the 32B model released in April.

Baidu: The Wenxin (ERNIE) X1.1 released on September 9 has significant improvements in authenticity, instruction adherence, and agent capabilities, with performance comparable to GPT-5 and Gemini 2.5 Pro.

Tencent: Released the leading 3D world generation model HunyuanWorld-Voyager, as well as the high-resolution text-to-image model HunyuanImage 2.1.

At the same time, AI is beginning to be embedded as "agents" in mainstream applications. For example, Meituan launched an AI assistant "Xiao Mei" based on its self-developed large model, supporting voice ordering and restaurant reservations; Alibaba's Amap launched the AI-native guide "Xiao Gao Teacher," aiming to transform map services into personalized travel companions.

Goldman Sachs stated that in the fields of text-to-video, text-to-image, and other multimodal areas, Chinese models are rapidly narrowing the gap with global counterparts. However, in terms of commercial monetization, Chinese AI applications still have a long way to go.

As of August 2025, the total annual recurring revenue (ARR) of global AI applications is approximately $30 billion, while China's AI applications have an ARR of only $1.5 billion, accounting for 5% of the global share