What does China’s growing AI adoption mean for investors?

The global artificial intelligence (AI) race is entering a new phase. Alongside the competition to build more powerful models, businesses are looking for affordable ways to put AI to work. Chinese developers are gaining ground as their models improve, raising questions about what this means for U.S. leaders. For investors, the opportunity depends on both the expansion of the market and how companies translate adoption into earnings.

Why are Chinese models gaining ground globally?

Chinese AI developers are making their mark beyond the domestic market. Although their models still lag leading U.S. systems, smaller performance gaps, particularly in areas such as coding, are making them more attractive to businesses looking for affordable ways to use AI.

Exhibit 1 shows this expansion among users in both the U.S. and the Global South, a group of emerging and developing economies. Chinese models now account for a substantially larger share of activity on OpenRouter than at the start of the chart. The trend extends beyond one platform: their share of usage on Vercel¹ rose from 11% in January to 55% in August 2026.²

Affordability is an important driver. As businesses use AI more frequently, small savings on each task can add up. This becomes particularly relevant as AI agents undertake several steps to complete a task, increasing the amount of information processed. Beyond Chinese models, U.S. companies are also recognizing this demand, with firms such as Anthropic rolling out smaller, more cost-efficient models that offer strong performance at a fraction of the cost of their most advanced systems.

Many Chinese developers also release open-weight models³, allowing businesses to download and adapt them for their own systems. This gives businesses greater flexibility over how they use AI, although running and maintaining the models still carries costs. Together, improving capabilities and easier access are lowering the barriers to adoption. 

How are businesses choosing which models to use?

For businesses, choosing an AI model involves balancing cost with the value it delivers. An affordable model may be sufficient for routine tasks, such as summarizing documents or answering simple customer enquiries. For more demanding work, especially in large enterprises, trust, reliability, security and proven performance can matter more than price. These factors give leading U.S. models a meaningful advantage where companies are deploying AI into sensitive, regulated or business-critical workflows.

The price of using a model is also only part of the decision. Companies need to consider how easily it fits into existing systems, how it handles sensitive information and how much human oversight its answers require. A lower-cost model offers less of a saving if employees spend more time checking and correcting its output.

This suggests businesses may increasingly use several models, matching each to the task at hand. Chinese providers have an opportunity to win more everyday workloads, particularly among startups that are more willing to experiment with lower-cost alternatives, while advanced U.S. models are likely to remain attractive for large enterprises and higher-value use cases where trust, reliability and integration with existing systems are central to the buying decision. The models handling the most activity may therefore differ from those receiving the largest share of customers’ budgets.

Where is AI spending going?

The spending picture remains more favorable to U.S. providers than usage figures alone might suggest. Exhibit 2 shows Chinese models’ share of OpenRouter spending rising to around 36% in the Global South and 24% in the U.S., still below their respective usage shares. In August, Anthropic accounted for 64% of estimated spending through Vercel’s AI Gateway.⁴ Separately, downloadable or “open-weight”, models handled 56% of tokens but represented just 14% of estimated spending. The open-weight category is not exclusively Chinese, but the comparison illustrates the gap between usage and revenue. These estimates reflect activity on one platform and use published prices rather than actual customer bills. 

Chinese providers are nevertheless beginning to attract more spending. Vercel reported that Moonshot and Z.ai drove more than 90% of the increase in open-weight models’ spending share in July.5 This suggests that better capabilities can help Chinese developers compete beyond the cheapest end of the market.

For investors, both trends matter. U.S. providers’ ability to attract spending points to the commercial value of their models. Chinese providers’ growing adoption offers a foundation for future revenue growth. Lower prices could also encourage customers to use AI more widely, expanding the overall market. Whether that translates into stronger profits will depend on how quickly demand grows and how efficiently companies deliver their services.

Investment implications

China’s progress broadens the AI investment opportunity. More affordable models could make AI worthwhile for businesses and tasks where it previously cost too much, supporting demand for applications and cloud services. This may be especially relevant for startups and cost-sensitive users that are more willing to experiment with lower-cost Chinese models. U.S. leaders can continue to benefit from advanced capabilities, greater trust and reliability, and established relationships with business customers. Growth in one market need not come entirely at the expense of the other.

The opportunities also extend beyond model developers. Cloud providers supply the computing power, software companies bring AI into everyday work, and businesses using these tools effectively may improve productivity and reduce costs. This favors diversification across both markets and different parts of the AI industry.

A positive outlook still calls for selectivity. Price competition can squeeze profits, while heavy investment raises the pressure to deliver results. Investors should focus on companies demonstrating growth in paying customers, improving profitability and disciplined spending, while remaining mindful of valuations. As AI becomes more accessible, the investment opportunity will increasingly depend on who can turn wider usage into lasting earnings.

¹ Vercel is a cloud platform for deploying and hosting modern web applications, while OpenRouter is an API routing layer that provides access to multiple AI models. In short, Vercel supports application delivery, whereas OpenRouter supports model connectivity. 

² OpenRouter and Vercel’s AI Gateway are services that allow developers to access models from different providers. Usage here is measured in tokens, the small units of information models process. These platforms represent a portion of the market and should not be treated as a measure of total global adoption. OpenRouter’s Global South grouping covers 82 countries across Latin America, the Caribbean, Africa, Central Asia, South and Southeast Asia and the Middle East. The Vercel adoption figures were reported by CNBC on 26 September 2026.

³ “Open-weight” models’ learned settings are available to download, subject to license terms. This does not necessarily mean their training data are public or that they are free to operate.

⁴ Vercel’s September 2026 report covers August activity. Spending is estimated using published model prices; actual customer bills may differ. These figures are not audited revenues or a measure of the whole AI market.

⁵ Moonshot and Z.ai accounted for more than 90% of the increase in open-weight models’ spending share on Vercel in July. Open-weight models are not exclusively Chinese. The short observation period does not establish customer retention or long-term profitability. 

f7cd556a-d11a-11f1-8ae3-b1764270d1f9

IMPORTANT DISCLAIMER on Artificial Intelligence (AI): This media contains AI generated or enhanced contents, including but not limited to the image, voice, and spoken contents of JPMorgan Asset Management representatives. These elements have been created or modified using AI technology and do not represent an actual recorded appearance/presence or live presentation by the individual. The representations, translations, and avatars in this media may not accurately reflect real individuals, events, or languages. This media is intended for informational purposes only and should not be relied upon as a sole source for decision-making.