Goldman Sachs picks its favorite Chinese AI models

Goldman Sachs initiated coverage on Hong Kong-listed Zhipu (Knowledge Atlas) with a price target of 1,880 HKD ($239.83), implying ~15% upside from that day’s close. The firm cited near-frontier performance from Zhipu’s open-sourced GLM-5.2 and expectations that coders’ usage can support frequent upgrades and sustained enterprise/coding momentum, while keeping the rating neutral. Zhipu stock has jumped ~70% over the prior 60 trading days, contrasting with sharp moves for Minimax (-70%+) and declines for Alibaba (~-10%) and Tencent (~-5%).

Analysis

The market is still pricing Chinese AI as if benchmark quality alone creates moat, but the economic center of gravity is shifting to compute access, inference efficiency, and distribution. That is structurally negative for pure model-race narratives and only conditionally positive for platform companies that can turn usage into recurring cloud or workflow revenue. In that framing, BABA and TCEHY are not obvious AI winners just because they have AI initiatives; they need proof that model traffic monetizes above the cost of serving it.

The second-order dynamic is competitive compression: once open-weight models reach near-frontier levels, differentiation decays faster and pricing power shifts away from model vendors toward whoever controls developer habits, enterprise procurement, and low-cost compute. That favors the largest balance sheets over smaller labs, but it also means the equity market can overreact to leaderboard rankings that may not persist for more than a few release cycles. The recent surge in one public proxy looks more like narrative momentum than durable earnings power.

Catalyst-wise, the next few weeks are about follow-through versus fade: more model releases, developer adoption chatter, and any cloud/AI revenue commentary from BABA or TCEHY. Over 1-3 months, the key falsifier is whether AI traffic actually lifts reported cloud growth or margins; if not, the sector will revert to treating these names as expensive R&D holders rather than monetization stories. Over 6-18 months, tighter US-China compute restrictions could paradoxically help the best-capitalized incumbents while crushing smaller model players that cannot secure enough inference capacity.

Key Decisions for Investors

• Avoid chasing the current AI momentum basket on the first gap-up; wait for consolidation or a 10-15% retracement before adding exposure, because the move is being driven by benchmark optics more than verified monetization.
• Relative-value: go long BABA / short TCEHY over 1-3 months if you want China AI exposure. Alibaba has the cleaner path to cloud/inference monetization, while Tencent’s AI story is still more dependent on consumer bundling and less on direct pricing power.
• Use rallies in BABA and TCEHY to trim or hedge unless the next earnings cycle shows quantifiable AI-driven revenue acceleration or margin expansion; absent that, the market is likely to keep compressing AI optionality multiples.
• Set a watch item on cloud revenue and capex efficiency for the next two reporting cycles. If AI-related spend rises without a visible lift in revenue per unit of compute, fade the sector; if margin improves, the thesis flips quickly.
• No direct long recommendation on the public model pure-play proxy here: the risk/reward is poor unless there is evidence of durable enterprise pricing power, not just model-ranking leadership.