Bernstein: Chinese large models will become the "king of cost-effectiveness" in the global market.

The AI large model market is moving toward a layered competitive landscape, and Chinese AI labs are expected to gain a pivotal position in the global market thanks to significant cost advantages.

According to Chase Wind Trading Desk, Bernstein analysts Robin Zhu and others pointed out in their latest report that even considering geopolitical constraints and assuming Chinese models have little penetration in the US market, Chinese AI labs can still access about 35% to 40% of the potential global AI market (TAM)—in absolute terms, approximately $320 to $350 billion.

The report argues that the high token pricing of Anthropic’s Claude Fable 5 has prompted developers to re-evaluate the cost of AI usage. This event marks an accelerated focus on AI investment returns in the market, and will hasten the shift of users toward more cost-effective models.

This view has profound implications for the global AI industry. In Bernstein’s baseline scenario, US frontier labs will continue to dominate highly specialized and high-premium frontier applications; in broader “tail” applications including consumer, SMEs, and emerging markets, Chinese AI labs are expected to systematically erode market share with lower token prices.

A New Framework for AI Commoditization: Perception Matters More Than Algorithms

Bernstein puts forward an AI commoditization analysis framework that differs from traditional perceptions. Traditionally, AI commoditization is thought to stem from convergence of underlying intelligent capabilities in models; Bernstein argues the real driver is human users' perception of model capability, and whether a model in a specific application scenario is “good enough” and can operate reliably at scale.

The report ranks AI application scenarios by speed of commoditization: consumer scenarios (such as ordering takeaway, booking hotels) will commoditize first; followed by enterprise workloads of high determinacy (such as Excel modeling); next, complex strategic planning and cybersecurity; finally, frontier scientific domains such as drug development, nuclear fusion, and space exploration—where users’ willingness to pay is nearly unlimited, supporting high premiums for frontier models over the long term.

Bernstein notes that Tencent WeChat’s AI agent announcement and Alibaba’s exploration in Qwen applications show that commercial deployment of AI agents in consumer scenarios—such as buying milk tea, tickets, or T-shirts—is imminent. Once a class of tasks is “solved,” the marginal return for further R&D investment in that area drops sharply, and AI labs’ resources naturally pivot to more complex and frontier tasks.

Market Segmentation: US Guards the Frontier, China Attacks the Heartland

Bernstein predicts the global AI market will gradually form a dual-layer structure. The first layer will be dominated by US frontier labs, continually unlocking new capabilities and serving clients with high willingness to pay and increasingly specialized needs; the second layer will be the “tail AI” market serving more ordinary enterprises and consumer needs, where competition will center around per-task costs, reliability, and developer trust.

Internationally, US acceptance of Chinese AI models is extremely low; Europe is less severe; in other regions, especially emerging markets like the Middle East and Southeast Asia, users’ acceptance of Chinese AI models is generally high.

The report also points out that the US frontier labs upgrading from Blackwell to Rubin and other next-generation chips could widen the China-US model capability gap in the short term; but historically, technology diffusion will push this gap to narrow again, and in the commercial world, a 6–12 month gap is not long relative to consumer habit stickiness and enterprise procurement inertia.

The Cost Advantage and Profit Outlook of Chinese AI Labs

Bernstein believes Chinese AI labs' cost advantage comes from several structural factors: lower developer manpower costs, “latecomer advantage” by following global SOTA as a research beacon, and flexible use of older generation chip clusters. These factors jointly ensure that Chinese labs’ absolute R&D spending will long be lower than US peers.

On profitability, the report is cautiously optimistic. Bernstein expects R&D expenditure at China’s top AI labs to continue rapid growth over the next five years—the report cites Zhipu and Minimax statements, noting a 50% annual compound growth rate “is not surprising.” But as more application scenarios are “solved,” the scope requiring continued exponential R&D investment will gradually narrow, the growth rate of R&D expenses may slow, and operating leverage for AI labs will be possible.

On inference, the report believes most AI labs’ inference profits are “moderate to strong”; operational costs other than R&D are relatively lean, and marketing expenses can stay low given "excellent inference abilities essentially sell themselves." Bernstein’s overall judgment is that the evolution path of AI commoditization is, in fact, “quite optimistic” for long-term economic models of AI labs.

Alibaba and Tencent May Benefit First

Bernstein maintains “outperform” ratings for Tencent and Alibaba in its report, with target prices of HKD 780 and USD 180/HKD 176 respectively.

The report believes both companies are key participants in the implementation of AI commercialization in China—Tencent WeChat’s AI agent layout and Alibaba’s Qwen ecosystem both point to the early commercialization of consumer-level AI scenarios.

For the overall AI investment theme, Bernstein states that the market response triggered by Claude Fable 5’s high token cost could accelerate the timetable for developers and users to re-examine AI investment returns.

The report believes AI users will choose between models to match marginal token cost and marginal task completion benefit; this logic will systematically favor Chinese AI labs that provide “good enough” inference capability at significantly lower pricing. Bernstein frankly admits its own experience of consuming about 2 billion tokens in the past month or two has further reinforced this view.

~~~~~~~~~~~~~~~~~~~~~~~~

The above highlights are from Chase Wind Trading Desk.

For more detailed interpretations, including real-time analysis and frontline research, please join【Chase Wind Trading Desk Annual Membership】

Risk Disclosure and DisclaimerMarkets involve risk. Investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users’ particular investment objectives, financial circumstances, or needs. Users should consider whether the opinions, views, or conclusions herein fit their individual situations. Investments made accordingly are at your own risk.