Bernstein: Chinese LLMs to Capture 35%-40% of Global Market on Cost Advantage
0xBroomberg
Bernstein estimates Chinese AI labs can address 35%-40% of the global AI market — roughly $320-350 billion — even with near-zero U.S. penetration, driven not by closing the capability gap but by users deciding a cheaper model is "good enough."
Where does the 35%-40% figure come from?
The report assumes Chinese models have almost no penetration in the U.S., limited acceptance in Europe, but broadly high acceptance in the Middle East, Southeast Asia, and other emerging markets.
Add those regions up and Chinese AI labs' total addressable market (TAM) reaches roughly 35%-40% of the global total — about $320-350 billion in absolute terms.
This means → even if geopolitics locks the U.S. market almost entirely, demand from the rest of the world still leaves Chinese AI companies with a very high ceiling.
Why would users switch to cheaper models?
Bernstein treats Anthropic's Claude Fable 5 and its steep token pricing as a market inflection signal — developers are reassessing costs and accelerating migration toward better value-for-money alternatives.
The core logic: AI commoditization is driven not by convergence in underlying model intelligence but by user perception — the point at which a model hits the "good enough" threshold for a given use case.
In plain terms = think smartphone chips: a flagship and a mid-range chip score very differently in benchmarks, but most people cannot tell the difference in daily use. AI models are heading down the same path.
How will the global AI market split?
Tier one: dominated by U.S. frontier labs, continuously unlocking new capabilities for high-willingness-to-pay, highly specialized clients — drug discovery, nuclear fusion, space exploration — where users will pay near-unlimited premiums.
Tier two: consumer, SME, and emerging-market "long-tail AI," where competition shifts to cost per task, reliability, and developer trust.
This means → Chinese labs' main battlefield is tier two — winning not with the strongest model but with the most cost-effective one, capturing massive long-tail demand.
Where does the Chinese cost advantage come from?
Three structural factors: lower developer labor costs + a fast-follower advantage from using global state-of-the-art models as a development beacon + flexible use of prior-generation chip clusters.
The report cites Zhipu and Minimax, noting a 50% CAGR in R&D spending over five years for top Chinese AI labs is "not surprising."
But as more use cases get "solved," the frontier requiring exponential R&D investment narrows → R&D growth slows → operating leverage — revenue growing faster than costs — begins to emerge.
In plain terms = the catch-up phase burns cash, but once mainstream use cases are "good enough," the burn rate slows and profit margins open up.
Why are Tencent and Alibaba singled out?
Bernstein maintains an outperform rating on Tencent with a target price of HK$780, and on Alibaba with targets of US$180 / HK$176.
Tencent's WeChat AI-agent push + Alibaba's Qwen ecosystem both point to consumer-level AI commercializing first — the report says AI-agent deployment for ordering bubble tea, booking flights, or buying a T-shirt is "just around the corner."
This reflects Bernstein's core bet: whoever is closest to the consumer and owns the most everyday use cases will be first to monetize the AI commoditization wave.
Could the U.S.-China model gap widen again?
The report flags a risk: U.S. frontier labs upgrading from Blackwell to Rubin and other next-gen chips could widen the capability gap in the short term.
But historically, technology diffusion narrows such gaps again — Bernstein argues a 6-to-12-month lead window is "not long" relative to consumer-habit stickiness and enterprise-procurement inertia.
This means → the generational hardware lead is real, but in the commercial world users switch far more slowly than chips iterate — time favors the pursuer.
Content is for reference only, not financial advice.