As 2025 draws to a close, Goldman Sachs Research published a report on December 19, summarizing China's key advancements and achievements in artificial intelligence (AI) while forecasting trends for the coming year.
Goldman Sachs notes that the gap between China's AI models and top-tier U.S. models has narrowed to just 3-6 months. The report also highlights that the launch of AI assistants like ByteDance’s Doubao Mobile Assistant could signal a fundamental shift in mobile app traffic dynamics.
The firm remains bullish on cloud computing, data centers, and full-stack AI beneficiaries. It projects that China’s leading cloud service providers will increase capital expenditures to RMB 500 billion in 2026, up 20% from the estimated RMB 400 billion in 2025.
Domestic chips and computing power are expected to account for 40% of this spending in 2026, a significant rise from 20-30% in 2025. Additionally, the global AI video generation market is forecast to grow from $1 billion in 2025 to $39 billion by 2033, reflecting a 56% CAGR over eight years.
**Frontier AI Models and Agent Capabilities Continue to Advance** Goldman Sachs observes that the technological gap between China and the U.S. in AI is rapidly shrinking. While U.S. models like Google’s Gemini 3 and OpenAI’s GPT-5.2 maintain leadership with each update, Chinese AI models typically catch up within 3-6 months.
Following Google’s Gemini 3 release, OpenAI launched GPT-5.2 on December 11, enhancing general intelligence, coding, and long-context understanding. The model excels in tasks like spreadsheet creation, presentations, image perception, tool usage, and multi-step project execution.
Chinese firms are keeping pace. On December 16, Xiaomi unveiled its open-source MiMo-V2-Flash model, ranking among the top two open-source models in multiple agent benchmarks. ByteDance followed on December 18 with its latest Doubao-Seed-1.8 large language model, improving agent and multimodal capabilities.
Goldman Sachs notes that while U.S. foundational text and multimodal models (e.g., GPT-5, Sora2, Gemini 3 Pro) lead with each update, Chinese models consistently close the gap within months before achieving another leap.
**The New Era of AI Assistants: Doubao Mobile** On December 1, ByteDance’s Doubao team released a technical preview of Doubao Mobile Assistant (first integrated with ZTE’s Nubia smartphones), an OS-level AI assistant developed in collaboration with smartphone manufacturers.
Zhipu AI also open-sourced AutoGLM, an AI agent capable of executing multi-step tasks like food delivery orders and flight bookings across 50+ high-frequency Chinese apps by interpreting screen content and simulating inputs.
Xiaomi revealed that its smartphone AI agent, Super XiaoAI, boasts 120 million monthly active users and 65 million daily sessions, supporting over 3,000 agent skills across 100 popular internet apps.
The report suggests that OS-level AI assistants could usher in an era of voice commands and all-purpose helpers, allowing users to focus on entertainment while the assistant handles background tasks like messaging.
Goldman Sachs believes these assistants have the potential to disrupt the app market, threatening existing traffic and ad revenue models while creating new opportunities. However, challenges like "walled ecosystems" and security concerns remain.
The firm also highlights ByteDance’s competitive challenges in sectors like social media, music, e-commerce, and messaging, where its apps dominate iOS download charts in China.
**AI Inference Demand Surges: Daily Token Processing Exceeds 50 Trillion** ByteDance announced on December 18 that Doubao’s daily token usage has surpassed 50 trillion (up from 30 trillion in October), ranking first in China and third globally, reflecting strong AI adoption.
Its Volcano Engine MaaS platform now serves 80% of top FMCG brands, 90% of major automakers, 80% of leading brokerages, 70% of China’s elite 985 universities, and 9 of the top 10 global smartphone manufacturers by shipments. The company expects 2025 revenue to double year-on-year to over RMB 20 billion.
Goldman Sachs projects China’s cloud service providers will increase 2026 capital expenditures by 20% to ~RMB 500 billion, with domestic chip/computing power share rising to 40% (from 20-30% in 2025).
**China’s Multimodal Models Gain Global Traction** Chinese AI models are expanding globally through cost, open-source, and speed advantages. In mid-December, BBAT (Baidu, ByteDance, Alibaba, Tencent) released new multimodal models:
- **Alibaba** (Dec. 16): Launched Wan2.6, a video generation model supporting multi-scene narratives and stable multi-character dialogues, producing 15-second 1080p videos. Its Tongyi App integrated Amap, marking another step toward becoming a life/work assistant. - **Tencent** (Dec. 17): Released Hunyuan WorldPlay 1.5, a streaming video diffusion model enabling real-time interactive world modeling with geometric consistency. - **ByteDance** (Dec. 18): Unveiled Doubao-Seed-1.8 and Seedance 1.5 Pro, a multimodal model for joint audio-video generation from text/images.
As China narrows the gap in multimodal capabilities, Goldman Sachs notes differentiation via open-source models, competitive pricing, and speed. For instance, Kuaishou’s Kling 2.5 Turbo is significantly cheaper than Google’s Veo 3 or OpenAI’s Sora 2 while offering comparable performance.
The global AI video generation market is projected to grow from $1 billion in 2025 to $39 billion by 2033 (56% CAGR), driven by: - **Pro users**: $700 million in 2025 (64% of TAM), expanding to $17 billion by 2033 (49% CAGR), fueled by rising paid users and ARPU. - **Enterprise users**: $400 million in 2025, surging to $22 billion by 2033 (66% CAGR), led by AI adoption in digital ad and film production.
China’s share of the global foundational model market is estimated at 4% in 2025, rising to 7% by 2029 as model capabilities and pricing improve.
**Chip Supply and Capex Outlook** Goldman Sachs highlights that China’s cloud giants adopting multi-chip strategies reduce reliance on foreign suppliers, creating opportunities for domestic AI cloud growth.
BBAT’s 2025 capex is estimated at over RMB 400 billion (+62% YoY), with 2026 spending expected to rise another 20%, partly allocated to domestic chips and computing infrastructure.
Alibaba’s capex notably exceeds Tencent’s, attributed to its strong AI infrastructure and full-stack capabilities, akin to Google’s TPU advantage. Analysts remain optimistic about Chinese internet firms’ capex, forecasting Alibaba’s 2026-2028 spending to reach RMB 460 billion.
Higher computing efficiency may boost AI capex-to-revenue conversion, accelerating cloud revenue growth amid strong training/inference demand.
Based on full-stack AI capabilities, cloud demand, and data center expansion, Goldman Sachs ranks sector preferences as: 1. Cloud computing/data centers 2. Gaming 3. Mobility 4. E-commerce
Comments