Goldman Sachs comments on "The Battle of China's AI Giants": Alibaba vs Tencent vs ByteDance

If 2024 is the "parameter war" for models, then looking back from the end of 2025, the Chinese AI battlefield has evolved into a contest over capital efficiency, infrastructure hegemony, and traffic entry points.

On November 27, Goldman Sachs released a significant research report on the development of China's internet and artificial intelligence, analyzing the fierce competition in the current "big factory battle" of the Chinese AI industry, interpreting the different strategic choices of Alibaba, ByteDance, and Tencent:

• Alibaba has chosen the heaviest path, attempting to build a "full-stack" barrier similar to Google with an 80% year-on-year increase in capital expenditure and a full-stack layout, aiming to become the "full-stack hegemon" in the Chinese AI market;

• ByteDance leverages its terrifying traffic advantage, with a daily consumption of 30 trillion Tokens, dominating the application layer;

• Tencent has maintained its usual restraint, focusing more on "seamlessly embedding" AI capabilities into its vast social and payment ecosystem while reducing capital expenditure.

At the same time, Goldman Sachs believes that the competition in the AI field between China and the United States has entered a new normal of "dynamic alternation": as the U.S. model establishes new highs, the Chinese model will rapidly iterate and catch up in the following 3-6 months.

Alibaba's "full-stack advancement," Tencent's "restraint," and ByteDance's "traffic breakthrough"

1. Alibaba: A heavy bet on a "full-stack" approach comparable to Google

Alibaba is adopting a "full-stack" approach similar to Google.

Goldman Sachs emphasized Alibaba's capital expenditure in the report: Alibaba's capital expenditure in the September quarter surged 80% year-on-year, reaching 32 billion RMB.

Goldman Sachs analysts pointed out that despite facing fluctuations in chip supply, Alibaba remains committed to executing its AI infrastructure expansion plan, attempting to build a path similar to Google's "dedicated chips + full-stack services" through vertical integration of "basic models + multimodal capabilities" to consolidate its dominance in the B-end market.

This "heavy asset" strategy is translating into real profits.

The report shows that Alibaba Cloud's external revenue grew 29% year-on-year in the September quarter, with AI-related revenue achieving triple-digit growth for the ninth consecutive quarter. Goldman Sachs predicts that due to strong demand for AI training and inference, Alibaba Cloud's revenue growth rate will further accelerate to 38% in the December quarter.

On the C-end, Alibaba is also no longer low-key.

Its newly launched "Tongyi App" surpassed 10 million downloads in the first week of release, and Ant Group's "Lingguang" achieved 2 million downloads within 6 days, while the Quark app continues to iterate with AI.

2. ByteDance's "Traffic Breakthrough"

In the face of Alibaba's offensive, ByteDance's strategy has a distinct "traffic gene": leveraging its absolute dominance in C-end applications to feed back into the underlying infrastructure.

The report reveals that ByteDance's daily Token usage has surpassed 30 trillion. What scale is this? It is approaching the 43 trillion of global AI leader Google, and far exceeds other Chinese peers—by comparison, Baidu and DeepSeek currently have a daily Token volume of around 10 trillion.

Goldman Sachs mentioned in the report that ByteDance's "Doubao" consistently ranks first in domestic AI application activity, while its education application Gauth for overseas markets has seen a year-on-year revenue growth of 394%, setting a new historical high.

This enormous inference demand created through the application layer has also allowed ByteDance to complete a flanking maneuver against traditional cloud giants in the MaaS (Model as a Service) field.

Goldman Sachs cited IDC data indicating that in the critical future battleground of "public cloud market share for large models," ByteDance's Volcano Engine has captured 49.2% of the market share.

(3) Tencent: The Pragmatist's "Ecological Penetration"

Unlike the overt strategies of Alibaba and ByteDance, Tencent has maintained a consistent "restraint" on the AI battlefield.

Goldman Sachs noted that Tencent's capital expenditure decreased year-on-year in the same quarter, and it has lowered its annual budget target.

The report mentioned that, aside from the impact of chip supply fluctuations, this more reflects Tencent's restrained "ecological penetration" strategy, dedicated to seamlessly integrating AI capabilities into its vast WeChat ecosystem.

The report stated that Tencent has integrated the AI assistant "Yuanbao" into WeChat Pay, directly serving the operational efficiency of small and medium-sized businesses through AI copywriting tools and menu recognition functions. Although this approach lacks flashy capital expenditure figures, it possesses a high degree of implementation certainty.

Sino-U.S. Model Race: A Dynamic Catch-Up Cycle of 3-6 Months

Looking at the global dimension, Goldman Sachs also compared and analyzed the dynamics of AI technology between China and the U.S. in the report.

Goldman Sachs analyst Ronald Keung's team pointed out that although the market had previously worried about diminishing marginal returns of the AI Scaling Law, with Google's recent release of Gemini 3 Pro and the image generation model Nano Banana Pro, American foundational models have once again proven their breakthrough capabilities at the performance boundary, dispelling previous market concerns about "stagnation in AI technology."

However, this does not mean that Chinese players are being left behind. Goldman Sachs observed a clear and resilient "dynamic catch-up" cycle:

Whenever an American model achieves a significant version leap (Step-up), Chinese AI models typically follow up rapidly within the subsequent 3-6 months, narrowing the capability gap through technological iteration. This rhythm of "leading—catching up—leading again" has become the new norm in Sino-U.S. technological competition Goldman Sachs also mentioned that the resilience of Chinese manufacturers lies in their unique "China speed" and open-source ecosystem. Major companies like Xiaomi and Tencent are open-sourcing their high-performance models, and it is reported that 80% of AI startups in China are using these open-source models.

In addition, Chinese models are highly aggressive in cost control. For example, Kuaishou's "Kling" video generation model is significantly cheaper than similar products globally, building a moat at the application level through exceptional cost-performance ratio.

Value Traps?

Regarding the valuation issues that investors are concerned about, Goldman Sachs' Ronald Keung team analyzed:

"We are not yet in an AI bubble."

The team believes that there is currently no bubble in the Chinese AI sector. Tencent and Alibaba's expected price-to-earnings ratios (P/E) for 2026 are 21 times and 23 times, respectively, lower than Google's (24 times), and Amazon and Microsoft (28-30 times)