Goldman Sachs Names Zhipu, DeepSeek and ByteDance Its Top Chinese AI Picks

Wall Street Picks Its Chinese AI Winners

Goldman Sachs on Friday initiated coverage on Zhipu, also known as Knowledge Atlas, setting a price target of 1,880 Hong Kong dollars, or about $239.83, according to CNBC. That target implies nearly 15% upside from where the stock closed that day, even though Goldman rated the stock neutral, not a buy.

Zhipu isn't Goldman's only favorite. In a separate report released the same day, the bank named DeepSeek and ByteDance as its other top picks in Chinese AI, according to CNBC. Neither of those companies is publicly traded, so investors can't buy shares directly.

Why Zhipu Is Getting Attention

Zhipu has skyrocketed since its listing in Hong Kong in January, according to CNBC. Over the last 60 trading days, the stock has jumped 70%.

The company's open-source GLM-5.2 model is considered to rival Anthropic's Fable 5 on several metrics, CNBC reported. Goldman's analysts wrote that GLM-5.2 is reaching "near-frontier performance" with rising adoption from both domestic enterprises and global small-to-medium businesses.

Goldman's analysts said that adoption by coders should let Zhipu keep pushing out model upgrades fast enough to hold its lead in enterprise and coding applications inside China, according to the CNBC report.

The Rest of the Field

Goldman ranked Zhipu's GLM models and DeepSeek's models ahead of offerings from Alibaba, Tencent, and Minimax, particularly on time-to-market and Arena score, a competitive benchmark used to grade AI models head-to-head, according to CNBC. ByteDance came out on top in AI video generation.

The divergence in stock performance among Chinese AI names has been stark. While Zhipu shares climbed 70% over the past 60 trading days, Minimax has cratered, dropping more than 70% in the same window, according to CNBC. Alibaba is down nearly 10%, and Tencent has slipped roughly 5%. A handful of winners and many losers reflect the stock-specific dynamics that Goldman's report highlights: coverage of the whole basket, with conviction concentrated in three names.

The China AI Story Washington Doesn't Want to Hear

Goldman's analysts framed the moment bluntly. China's open-source and open-weight AI models are reaching a critical point of intelligence performance compared to global proprietary models, according to CNBC. This is happening despite years of U.S. export controls designed to choke off China's access to advanced chips.

Goldman's analysts said agentic AI is driving explosive demand for these value-for-money models at the lower end, according to CNBC.

At the same time, Goldman's analysts flagged that computing access remains the real swing factor going forward, citing U.S. and Chinese regulations, company balance sheets, and inference efficiency as the variables that will determine who keeps winning, according to CNBC. Chip access still matters, and Washington's export rules are still a live constraint on how far these firms can scale, even if they've found workarounds so far.

If chip controls are actually slowing DeepSeek, ByteDance, and Zhipu at the margins, easing them would hand Beijing's AI champions more raw compute at exactly the moment they're already closing the performance gap. Critics of loosening export rules aren't wrong to worry about that.

Yet Goldman's own numbers show these companies are producing near-frontier models anyway, chip constraints and all. That undercuts the assumption that export controls alone will keep China's AI industry permanently second-tier.

What's Unresolved

Goldman's neutral rating on Zhipu, despite the 15% upside price target, signals the bank isn't fully convinced the stock's run has staying power. A neutral rating with a bullish target is an odd combination, and it likely reflects volatility risk in a market where Minimax lost 70% in the same stretch Zhipu gained 70%.

The bigger open question is whether U.S. policymakers will tighten or loosen chip export rules in response to reports like this one. Goldman's analysts didn't make a policy recommendation. They just laid out the competitive picture and let the numbers speak: China's AI models are getting good, fast, and the compute fight isn't over.