Goldman Sachs Unveils Competitive Framework for Chinese AI Models

Goldman Sachs has published a comprehensive report delving into China's artificial intelligence (AI) large language model industry. The investment bank emphasizes that Chinese AI companies are developing models that rival their U.S. counterparts, achieving this at a fraction of the cost. Led by analyst Ronald Keung, the report, titled "Who Will Be the Long-Term Winner in China's AI Large Model Industry?", introduces an innovative three-dimensional framework to evaluate competitive strength across pricing power, cost advantages, and financial strength.

Released on Friday, the report indicates that the performance of China's open-source and open-weight large language models is approaching that of the world's top proprietary models. This advancement is primarily driven by architectural innovations and parameter efficiency, enabling Chinese models to demonstrate comparable capabilities with smaller sizes. Goldman Sachs' competitive positioning framework quantitatively assesses a player's long-term winning probability using the formula: ARR scale × gross margin advantage + financial strength.

According to the report, Zhipu (under initial coverage with a Neutral rating and a target valuation of $110 billion) and DeepSeek (an unlisted company) hold the strongest positions in the foundational text model domain. In the multimodal and video generation capabilities, ByteDance (also unlisted) emerges as the clear leader. Goldman Sachs also maintains a "Buy" rating on MiniMax and Kuaishou.

Goldman Sachs describes the Chinese AI model market as forming a "two-tiered structure." High-end models, such as Zhipu GLM5.2 and Alibaba Qwen3.7 Max, are priced at approximately $1 per million tokens, which is only 10% to 25% of the cost of their U.S. counterparts (priced at $4-8 per million tokens). The low-end market offers even more competitive pricing, as low as $0.06 to $0.20 per million tokens. These cost advantages are expected to drive a significant surge in Chinese AI model API and subscription revenue, projected to grow from an estimated 35 billion RMB in 2026 to 879 billion RMB by 2030, an approximate 25-fold increase.

China's efforts towards localization in AI infrastructure are also noteworthy. Meituan's LongCat 2.0 model, released on June 30, represents a significant milestone as the first open-source Mixture-of-Experts (MoE) model with 1.6 trillion parameters, entirely trained and deployed on 50,000 domestically produced computing cards. This development showcases China's ability to develop large-scale AI models with its indigenous hardware stack, despite limitations in accessing high-end computing power. The open-source strategy adopted by Chinese companies provides flexibility in deployment and fosters a community ecosystem, facilitating faster adoption and expansion into global SMEs and non-U.S. markets.

Analysts and market expectations suggest that China's AI model market is transitioning "from low cost to high intelligence." The data flywheel effect generated by the accelerating adoption of open-source models is expected to further drive model iterations and upgrades. Goldman Sachs notes that China's AI-related market valuations are "clearly undervalued" compared to its global position, with global mutual fund managers allocating only 1.2% of their portfolios to Chinese technology stocks. This indicates that the full potential for future growth may not yet be priced in, with API and subscription revenues, alongside enterprise adoption, anticipated to be the primary drivers of market expansion in the coming period.