A new Goldman Sachs report identifies five critical debates in China's AI sector, from a 1,000-fold surge in processing demand to an accelerated shift toward domestic chips.
Back
Goldman Sachs Outlines 5 Key Debates Shaping China's AI Race
Goldman Sachs sees China's artificial intelligence sector entering a new, more contentious phase, with a new report highlighting a 1,000-fold increase in daily processing demand since the start of 2024 that is fueling a boom for cloud providers like Alibaba Group Holding Ltd.
"An ultra-tight constraint on computing power is pushing Chinese AI companies down a unique path—focusing on training and inference efficiency, data quality, and post-training optimization to build efficient architectures with fewer chips and less memory," the report said.
The analysis, which prompted Goldman to upgrade its rating for AI startup MiniMax to "Buy," points to China's daily token consumption—a measure of AI processing—surging to over 140 trillion. This growth comes as new models like DeepSeek V4 and Tencent's Hy3.0 intensify competition for established players.
The debates outlined by the bank carry significant weight for investors, framing a landscape where OS-level AI agents from firms like Bytedance could challenge the dominance of super-apps like WeChat, while the entire industry accelerates a strategic pivot to domestic silicon from providers such as Huawei.
Token Boom Fuels Cloud Giants as Competition Heats Up
The explosive growth in token consumption is the primary engine for AI cloud services, according to Goldman. Bytedance's Doubao model alone now accounts for 120 trillion tokens in daily usage, a figure that doubled in just three months. This demand directly benefits infrastructure providers, with Goldman analysts forecasting Alibaba Cloud's revenue growth to hit 40 percent in the March quarter, up from 36 percent previously. According to Gartner, Alibaba's IaaS market share in the Asia-Pacific region grew to 22.5 percent in 2025.
However, the market is increasingly fragmented. The recent launch of models like DeepSeek V4, which supports a 1-million-token context window with significantly lower memory requirements, has intensified the competitive pressure in the 20 billion to 300 billion parameter model range—a key battleground for AI agent applications. While Goldman notes the risk of a pricing war, it believes the tight global supply of computing power will ultimately support improving pricing for Chinese AI models.
The Agent Wars and a Pivot to Domestic Silicon
A core debate now centers on whether OS-level agents, such as Bytedance's Doubao phone assistant, will usurp app-based agents like the one reportedly planned for Tencent's WeChat. Goldman views this as a "profound paradigm shift" that could relegate today's independent super-apps to mere backend tool providers by capturing the main entry point for user traffic.
This software battle is unfolding alongside a hardware pivot. Driven by external pressures, Goldman sees an accelerating migration to domestic chips, primarily Huawei's Ascend 910C and 950 series, between 2026 and 2028. This trend toward technological self-reliance mirrors Goldman's observations in other sectors, such as the electric vehicle market, where domestic Chinese brands are rapidly gaining share against foreign competitors like Tesla, according to a separate Goldman report. The bank's bullish thesis on Chinese autonomous driving firm WeRide Inc., which it rates a "Buy," further reflects this theme of rising domestic tech champions.
This article is for informational purposes only and does not constitute investment advice.
[1] Goldman Sachs In-Depth: China's Five Key AI Debates[2] Goldman Sachs reiterates Apple stock Buy rating amid App Store slowdown By...[3] Goldman Sachs raises Apple stock price target to $340 on demand By Investing.com[4] What’s Behind Goldman Sachs’ Bullish Stance on WeRide Inc. (WRD)[5] Goldman sees Tesla stock orders decline as Chinese EV brands surge By...
