For investors, AI in China is not simply a high-growth technology theme, but a strategic capability tied to national competitiveness, industrial upgrading, and long-term productivity. As a result, policy support, capital allocation, and commercialization are unusually aligned, making AI one of the most consequential lenses through which to assess China’s next phase of growth.
China’s Cost-Led AI Strategy
China does not need to lead at the absolute frontier of model performance to succeed. Instead, its strategy emphasizes cost-efficient intelligence and rapid iteration. Open-source models significantly reduce deployment costs, accelerate developer adoption, and enable fast vertical fine-tuning. China now has the lowest application programming interface (API) costs globally, which is powerful because falling costs accelerate the scaling of enterprise and consumer AI use-cases. This cost-led approach helps explain the path China has chosen.
Two forces define China’s AI strategy. First, compute constraints. China’s AI capex is less than 20% of that in the United States, largely due to limited access to cutting-edge AI accelerators. Second, application-driven adoption. Even without leading-edge models, China’s growth in daily token consumption is comparable to that of the United States. Both enterprises and consumers are adopting AI at scale, with use-cases emerging across a wide range of industries.
To reconcile strong demand with constrained compute, China is pursuing three complementary approaches.
Architectural Efficiency and Model-Level Innovation
Developers are redesigning model architectures to reduce reliance on graphics processing units (GPUs) and high bandwidth memory (HBM) through techniques such as mixture-of-experts, conditional computation, and memory-efficient designs. In addressing compute constraints, DeepSeek provides a clear example of architectural innovation. The firm has advanced approaches such as mixture-of-experts and conditional memory to reduce GPU and HBM intensity across AI infrastructure. This is particularly important given that memory remains a key bottleneck and China has yet to achieve a breakthrough in HBM. By decoupling conditional memory from compute and offloading HBM requirements, these designs reduce the need for costly memory upgrades.
System-Level Engineering
This is led most visibly by Huawei. Rather than competing on single-chip specifications, Huawei’s strategy is to win at the system level. Through solutions such as Supernode, hundreds or even thousands of chips are interconnected to deliver competitive performance at the system scale. These architectures rely heavily on optical interconnects, which is why China’s AI infrastructure is notably more optical-intensive than that of the United States.
Domestic Substitution
Local AI chip vendors are expected to gain share, particularly in inference, supported by strong government-led substitution efforts. Inference workloads are more standardized, and economics matter more than peak training performance, allowing “good-enough” accelerators to become competitive more quickly.
From Constraints to Competitive Advantages
Taken together, this framework explains how China can continue to scale AI adoption and commercialization despite constraints at the frontier, turning efficiency, systems engineering, and demand into durable competitive advantages.
We therefore expect adoption to broaden across internet platforms, telecom operators, and data centers backed by state-owned enterprises—an area where government involvement differentiates China from the United States.
The Government’s Role in Scaling AI
China’s government plays a much more active role in shaping the AI and semiconductor ecosystem through three primary mechanisms. First, capital-market support, including green-channel IPO approvals for companies across AI chips, AI models, robotics, and memory. Second, direct capital support, exemplified by the National Integrated Circuit Industry Investment Funds. Third, policy support combined with increasing oversight, as illustrated by recent regulatory intervention in the Meta Manus acquisition.
Demand-Led Adoption and Platform Dynamics
Shifting from the supply side to the demand side, China benefits from a large base of digital-savvy consumers who trust AI and are willing to adopt it quickly. In chatbots, leading platforms—including ByteDance’s Doubao, DeepSeek, and Alibaba’s Qwen—each report monthly active users exceeding 100 million.
One visible consequence is the disruption of traditional search behavior: Baidu’s advertising revenue has declined by an estimated 15% to 20% as users migrate from keyword-based search to conversational interfaces.
We believe the next 12 months will be critical for AI agents, with multiple launches expected and Alibaba emerging as an early leader. However, integration remains the primary bottleneck. Cross-platform cooperation is not guaranteed, as illustrated by the recent Duobao AI phone, where attempts to execute tasks across ecosystems ran into limitations because access to third-party applications was restricted or blocked by app operators.
To understand why China’s internet leaders are still able to move quickly despite these frictions, it is helpful to examine their full-stack capabilities. Companies such as Alibaba, ByteDance, and Tencent are vertically integrated across infrastructure, models, and applications, which shortens the path from model development to deployment and monetization.
From an investor perspective, the key is distribution combined with deployment. We believe platforms that control user traffic, cloud infrastructure, and developer tools are best positioned to scale AI features rapidly and capture value—whether through cloud services, AI-driven advertising innovation, or higher e-commerce conversion rates.
It is also important not to underestimate ByteDance simply because it is privately held. ByteDance is currently the largest spender on AI-related capex in China and is aggressively gaining cloud share, positioning it as a central player—and potential winner—in the AI ecosystem.
Investment Implications Across the AI Stack
In terms of investment implications, increased availability of H200-class accelerators is effectively accelerating capex, as internet companies shift from renting compute to owning capacity. While power availability is not a binding constraint in China, thermal management is, which supports demand for cooling solutions, power management, and continued domestic substitution. We have already discussed memory and connectivity. Beyond this, physical AI is an important theme, as it represents the point at which AI becomes embedded in the real economy rather than remaining confined to digital applications.
In 2026, we expect China’s AI opportunity set to broaden and deepen across the full stack.
Model layer. Chinese large language models are expected to continue iterating rapidly, with ongoing multi-model and agentic releases from major platforms. IPO activity across the AI ecosystem should remain active, supported by a favorable domestic capital-markets backdrop.
Infrastructure layer. Earnings visibility is generally clearer at this level. Priority areas include localization of AI chips, memory, and SIM equipment; optics such as co-packaged optics and optical circuit switching; cooling and power solutions; testers; and server and switch ODMs. These themes are broadly consistent with infrastructure priorities in the U.S. market as well.
Application layer. This is where China has the greatest scope for structural differentiation beyond the major internet platforms, particularly through the rollout of AI agents and new applications. Attractive areas include vertical AI agents for enterprise use-cases, ADAS and robotaxi applications, and consumer-facing AI innovations.
Where else do we see growth in China? In parts two and three of this series, Andrey Glukhov and Pierre Horvilleur discuss biotech and advanced manufacturing, respectively.
Vivian Lin Thurston, CFA, partner, is a portfolio manager on William Blair’s global equity team.
China Growth Series
Part 1 | China’s AI Boom: From Compute Constraints to Commercial Momentum
Part 2 | China’s Next Phase in Healthcare Innovation
Part 3 | China Industrials: Growth Beneath Weak Headlines
