Goldman Sachs research published around July 12 named its three preferred Chinese AI models -- Zhipu, also known as "Knowledge Atlas" and the only publicly-traded name of the three on the Hong Kong exchange, alongside privately-held DeepSeek and ByteDance. Goldman initiated coverage on Zhipu with a HK$1,880 price target, roughly $239.83, and evaluated all three companies across four criteria: time-to-market, arena score (head-to-head model performance benchmarks), valuation and pricing, and AI video-generation capability.
The specifics of the ranking matter as much as the headline names. Zhipu's GLM model family and DeepSeek's models both ranked ahead of larger, better-known Chinese tech names -- Alibaba, Tencent and Minimax -- specifically on speed-to-market and arena score, suggesting Goldman's analysts see faster iteration cycles and stronger head-to-head benchmark performance as more predictive of investment value than sheer company size or existing market position.
“The specifics of the ranking matter as much as the headline names.”
Goldman didn't stop at picking individual models. The firm advised clients to rotate out of South Korean AI-related stocks into Chinese counterparts entirely, recommending a long "China AI basket" spanning the full stack -- power generation, semiconductors, AI infrastructure, models and applications -- rather than a narrow bet on any single layer. That's a notably broad and directional call from one of the largest sell-side research shops, effectively arguing that China's AI ecosystem as a whole offers better risk-adjusted exposure right now than Korea's.
For US investors, Goldman's China AI basket call is a useful signal that sophisticated institutional research increasingly treats the AI investment landscape as genuinely global and multi-polar, not a US-versus-everyone-else story -- a framing that matters for anyone allocating across geographies in AI-adjacent public and private markets. For founders and operators competing globally, Goldman naming DeepSeek and Zhipu ahead of Alibaba and Tencent on benchmark performance is a reminder that Chinese AI competitive dynamics are shifting fast, with newer, faster-moving labs outpacing established tech giants on model quality.
The bear case: sell-side research calls to rotate entire country baskets carry real execution and geopolitical risk that a model-benchmark comparison alone doesn't capture, including export-control uncertainty and access constraints that could affect Chinese AI companies regardless of underlying model quality. What to watch next: whether Zhipu's stock performance validates Goldman's price target over the coming months, and whether other major banks publish comparable China AI coverage that either corroborates or challenges Goldman's specific model rankings.