Goldman Sachs: The gap between China and the U.S. in AI has narrowed to "3-6 months," and Doubao phones indicate a "change in the application traffic pattern."

Author: Bao Yilong

Source: Hard AI

As we approach the end of 2025, on December 19th, the Goldman Sachs research team published a report summarizing the key progress and achievements China has made in the field of artificial intelligence, and looking ahead to potential development trends in the coming year.

Goldman Sachs believes that the gap between China's AI models and the top models in the United States has narrowed to 3-6 months. Additionally, the report suggests that the launch of AI assistants like ByteDance's Doubao mobile assistant may signal a fundamental change in the mobile application traffic landscape.

Goldman Sachs is optimistic about the cloud computing and data center sectors, as well as AI full-stack beneficiaries. It is expected that the capital expenditure of China's leading cloud service providers will further rise to 500 billion RMB in 2026, a 20% increase from the estimated over 400 billion RMB in 2025.

Among these, the proportion of domestic chips and computing power is expected to significantly increase from 20-30% in 2025 to 40% in 2026.

At the same time, the analysis points out that the global AI video generation model market size is expected to grow from 1 billion USD in 2025 to 39 billion USD in 2033, with an 8-year compound annual growth rate of 56%.

Continuous Breakthroughs in Cutting-edge AI Models and Agent Capabilities

Goldman Sachs believes that the technological gap between China and the U.S. in AI is rapidly narrowing.

Although U.S. models such as Google Gemini 3 and OpenAI's GPT-5.2 continue to lead with each update, Chinese AI models typically catch up and narrow the gap within the following 3-6 months.

After Google released Gemini 3, OpenAI launched GPT-5.2 on December 11, which further improved upon GPT-5.1 in general intelligence, coding, and long-context understanding. This model performs better in creating spreadsheets, building presentations, image perception, tool usage, and handling complex multi-step projects.

Chinese companies are closely following. On December 16, Xiaomi released the open-source MiMo-V2-Flash model, ranking among the top two open-source models globally in several agent benchmark tests. On December 18, ByteDance released the latest Doubao-Seed-1.8 large language model, further enhancing agent and multimodal capabilities.

Goldman Sachs observes that U.S. foundational text and multimodal models (such as GPT-5, Sora2, Gemini 3 Pro) continue to lead with each update, while Chinese AI models typically catch up and narrow the gap within the following 3-6 months, followed by another phase of leap forward.

New Era of AI Assistants — Doubao Mobile Assistant

On December 1, the ByteDance Doubao team released a technical preview of the Doubao mobile assistant (the first device being ZTE Nubia), which is an AI assistant integrated at the operating system level through collaboration with smartphone manufacturers.

Zhipu AI has also open-sourced AutoGLM, an AI agent model capable of executing multi-step tasks such as food delivery orders and flight bookings on over 50 high-frequency Chinese applications by interpreting screen content and simulating input Xiaomi revealed that its smartphone AI assistant, Super Xiao Ai, has 120 million monthly active users and 65 million daily usage conversations, capable of achieving over 3,000 proxy skills across 100 commonly used internet applications through various protocols.

The research report points out that the smoother capabilities of operating system-level AI assistants may bring users closer to the era of voice commands/universal assistants, allowing users to focus on entertainment content consumption on mobile devices while the assistant helps reply to messages in the background through voice activation.

Goldman Sachs believes that operating system-level intelligent assistants indeed have the potential to disrupt the existing application market landscape, and they may pose a threat to existing application traffic and advertising businesses. However, at the same time, these new technologies may also bring new business opportunities.

However, the report emphasizes that challenges also arise due to "walled ecosystems" or security issues.

Goldman Sachs continues to monitor the challenges faced by ByteDance in fiercely competitive areas (such as social media, music, trading, instant messaging, etc.) in utilizing artificial intelligence technology, as well as the potential impact of these challenges on its business development.

In fact, in these areas, ByteDance's applications have recently ranked among the top in the free domestic application download list on the iOS system.

Explosion in AI Inference Demand: Daily Token Processing Volume Exceeds 50 Trillion

On the 18th, ByteDance announced that the average daily token usage of its Doubao large model has exceeded 50 trillion (compared to 30 trillion in October), ranking first in China and third globally, highlighting the continued strong momentum in AI adoption and token usage.

ByteDance's Volcano Engine previously announced that its MaaS currently serves 80% of top fast-moving consumer goods brands, 90% of major automotive OEMs, 80% of leading brokerages, 70% of China's top 985 universities, and 9 out of the top 10 smartphone manufacturers by global shipments.

The company expects its revenue for the fiscal year 2025 to exceed 20 billion yuan, doubling from last year.

Goldman Sachs expects that capital expenditure by Chinese cloud service providers will further increase by 20% to about 500 billion yuan by 2026, with a significant increase in the proportion allocated to domestic chips/computing power (from an estimated 20-30% in fiscal year 2025 to 40% in fiscal year 2026).

China's Multimodal Models Accelerate Global Penetration

Chinese AI models are accelerating global market penetration through cost, open-source, and speed advantages.

In mid-December, BBAT (Baidu, ByteDance, Alibaba, Tencent) respectively released new multimodal models:

Alibaba (December 16): Released the new version of the video generation model Wan2.6, supporting multi-camera narratives and stable multi-character dialogues, capable of generating 15-second 1080p HD videos. The Tongyi App officially integrated with Amap, marking another milestone in its evolution into a life/work partner.

Tencent (December 17): Released the Hunyuan WorldPlay 1.5, a streaming video diffusion model capable of achieving real-time interactive world modeling with long-term geometric consistency

ByteDance (December 18): Released Doubao-Seed-1.8 and video generation model Seedance 1.5 Pro, which is a multimodal model for jointly generating audio and video from text or images.

As the gap in multimodal capabilities with global enterprises narrows, Goldman Sachs believes that Chinese AI models can differentiate themselves through open-source methods as well as competitive pricing and speed.

According to data comparisons in the research report, Kuaishou's Kexi 2.5 Turbo is much cheaper than Google's Veo 3/OpenAI's Sora 2, while also being competitively performant.

Goldman Sachs expects the global AI video generation model total addressable market (TAM) to grow from USD 1 billion in 2025 to USD 39 billion in 2033, with an 8-year compound annual growth rate (CAGR) of 56%. Specifically:

• Professional User Market: Expected to be around USD 700 million in 2025, accounting for 64% of the total TAM, expanding to USD 17 billion by 2033, with an 8-year CAGR of 49%, primarily driven by an increase in the proportion of paying users and ARPU growth.
• Enterprise User Market: Expected to expand from approximately USD 400 million in 2025 to USD 22 billion in 2033, with an 8-year CAGR of 66%, accounting for 57% of the total TAM, mainly driven by the increased penetration of AI in digital video advertising production and film and entertainment video production.

In the global foundational model market, Goldman Sachs believes that the revenue share of Chinese manufacturers will be around 4% in 2025, and is expected to steadily grow to 7% by 2029 as model capabilities improve and pricing (token pricing discounts narrow) improves.

Evolution of Chip Supply Landscape and Capital Expenditure Outlook

Goldman Sachs believes that large Chinese cloud computing companies adopting a multi-chip combination approach undoubtedly brings new opportunities for the development of China's AI cloud industry—these companies are no longer completely reliant on overseas chip supplies.

Goldman Sachs estimates that the total capital expenditure of BBAT (Baidu, ByteDance, Alibaba, Tencent) will exceed RMB 400 billion in 2025 (a year-on-year increase of 62%).

For 2026, the research report expects the capital expenditure scale of large Chinese internet companies to grow by another 20%, with a significant portion of the funds allocated for purchasing domestic chips or for the construction of related computing facilities.

Goldman Sachs believes that Alibaba's capital expenditure this year is significantly higher than Tencent's, a phenomenon mainly attributed to Alibaba's strong capabilities in AI infrastructure and full-stack technology, similar to Google's full-stack technology advantage formed by dedicated TPUs.

Analysts continue to be optimistic about the capital expenditure outlook for large Chinese internet companies. According to forecasts, Alibaba's capital expenditure scale will be at a relatively high level in the industry from 2026 to 2028, expected to reach RMB 460 billion.

The research report points out that higher computing efficiency may enhance the conversion rate of AI capital expenditure to revenue, accelerating cloud revenue growth under the support of strong training or inference demand Based on the judgment of AI full-stack capabilities, the explosive demand for cloud computing, and the expansion of data center computing power, Goldman Sachs has provided a clear industry preference ranking: the first choice is cloud computing/data centers, the second is the gaming industry, the third is mobile transportation, and the last is the e-commerce industry.