Goldman's Keung: Chinese AI May Cut Global Prices-Only Zhipu Is Public

Cost and adoption are becoming the real test

Goldman's Ronald Keung argues that the more important shift is not simply that Chinese AI is improving. It is that the race is moving toward a cost-and-adoption contest. lowering the cost of AI models will drive much higher adoption, and that changes how investors should think about pricing power, market share, and valuations.

Goldman says Chinese models are reaching near parity with US rivals while costing far less. Chinese high-end models are priced at roughly $1 per million tokens, versus $4 to $8 for US equivalents. That is not a small discount; it is a meaningful shift in how enterprise buyers and developers may compare offerings.

Goldman's map is horizontal rather than centered on one winner. Zhipu and DeepSeek lead in foundational text models, while ByteDance has gained ground in multimodal and video generation. Goldman's three preferred names are only one of which is publicly traded, which helps explain why Zhipu remains the main listed proxy for this theme.

Goldman initiated coverage on Zhipu with a price target of HK$1,880 and highlighted significant ramp-up in domestic enterprise & global SME adoption tied to GLM-5.2. The more important point is the potential mechanism: extensive usage by coders could support faster model iteration and stronger user retention.

The real debate is monetization, not just quality

Bears have the cleaner boundary condition: adoption of cheap AI doesn't necessarily translate into profitability, and Zhipu has said it would not pursue short-term monetization from AI applications. That makes the core split less about quality versus cheap and more about whether usage becomes sticky enough to support durable revenue.

What to watch next

• Monetization: Whether enterprise and SME usage turns into revenue or remains secondary to growth.
• Ecosystem lock-in: Whether coder habits and upgrade frequency create switching costs.
• Execution constraints: Whether cost leadership can be maintained through efficient architectures, compute access, and financial strength.