Much of what I read about developments in the new artificial intelligence technologies focuses on the capabilities of what the new models can do, like what problems they now seem ready to tackle. But a parallel and not-identical question is what models are actually being used in the global economy. Austin Horng-En Wang and Kyle Siler-Evans offer some evidence on this point in “U.S.-China Competition for Artificial Intelligence Markets: Analyzing Global Use Patterns of Large Language Models” (RAND Corporation, January 14, 2026).
The authors analyze “website traffic data across 135 countries from April 2024 through May 2025,” and in particular, they tracked “monthly website visits from each country to seven U.S.-based and thirteen China-based LLM [large language model] service websites.” The authors readily acknowledge that their numbers are imperfect. As one example, if some organization downloads an open-source AI tool and uses it, this will not be captured by web traffic. The timeframe is interesting in part because in January 2025, a Chinese company called DeepSeek launched an AI tool with capabilities that considerably exceeded expectations. But did the capabilities of DeepSeek translate into actual use patterns? The answer seems to be that it made a noticeable but not overwhelming difference.
The figure shows monthly visits to prominent LLM websites, with visits to US websites in blue and Chinese websites in red. Chinese websites were getting about 2-3% of the total visits in 2024. With the arrival of DeepSeek, the Chinese share rose as high as 13% in February 2025, but by August 2025 had sagged back to about 6%. As you can see, the growth in use of AI sites in summer 2025 mostly happened at US-based websites.
The authors explore whether the dominance of US AI providers might be explained by looking at factors like pricing, language support, and diplomatic ties, but without much success. For example, while paid subscriptions to US AI services cost more, many users are still relying on free access. Given the lack of other plausible explanations, they suggest that the capabilities of the US-based AI tools are currently better. But the DeepSeek experience shows that lots of users around the world do not view themselves as locked-in to AI providers from any given provider or country, and are quite willing to give something else a try.
