Multiple foreign enterprises, including those from the US and the EU, are increasingly turning to Chinese artificial intelligence (AI)-powered large-language models (LLMs), with the share of tokens used by US firms on Chinese AI models reportedly hitting as high as 46 percent in the first half of 2026.

Attributing the trend to a combination of factors including cost-effectiveness, enhanced performance, open-source autonomy, and supply chain diversification, Chinese analysts said that they expect more foreign companies to adopt Chinese AI models.

DoorDash co-founder Andy Fang said last week that the US food delivery group now delegates "lower-level work" to Kimi K2.6, a model from Chinese start-up Moonshot AI, and reserves US AI tech firm Anthropic's Fable for only "the hardest work," the Financial Times reported on Monday.

The new combination "vastly outperform[ed]… at a cheaper cost" than a previous setup that used only US frontier models from Anthropic, he said on X.

In addition, cryptocurrency exchange Coinbase's CEO Brian Armstrong, in an X post on June 28, outlined five ways in which the crypto exchange was keeping AI costs low, including experimenting with Chinese LLMs as default options.

"We're experimenting with defaulting to open-weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task," Armstrong wrote.

GLM 5.2 is a model developed by Chinese AI firm Z.ai.

In June, US AI start-up Lindy moved 100 percent of its traffic from Anthropic's Claude models to DeepSeek V4 - an open-source model released by Chinese AI firm DeepSeek in April 2026.

Switching to DeepSeek V4 saves millions of dollars and increases performance on many core use cases for Lindy, Lindy CEO Flo Crivello wrote in an X post in June, saying that the move is "Transformative for the business."

The share of tokens used by US companies on Chinese AI models via OpenRouter - one of the major overseas multi-model aggregation and routing platforms, serving as an important window for observing global LLM usage trends - has sat above 30 percent every week since February 8, with that figure rising as high as 46 percent. The average across the previous 12 months was just 11 percent, falling to 4.5 percent in the first half of 2025, CNBC reported on July 7.

A token is the smallest unit of text that an AI LLM reads and processes. It can be a whole word, part of a word, a punctuation mark, or even a single character.

Chen Jing, vice president of the Technology and Strategy Research Institute, told the Global Times on Monday that the first half of 2026 marked a turning point. Foreign enterprises' adoption of Chinese LLMs is no longer sporadic but systemic, and while multinational subsidiaries in China prioritize localization, international developers and corporations are actively pursuing cost effectiveness, Chen said.

"A key shift has recently emerged: the technical capabilities of Chinese open-source LLMs have advanced rapidly, effectively narrowing the generational gap and pushing practical utility past the critical threshold. Previously, overseas developers shunned Chinese models due to notable performance gaps. Today, however, with some domestic models like DeepSeek V4, GLM 5.2, and Qwen 3.7 narrowing the performance gap to an acceptable range, their overwhelming cost advantage has become impossible to ignore," Chen said.

"Over the past year, the coding and agent capabilities of domestic models have been comprehensively enhanced, and open-source text models have dramatically closed the gap with international top-tier standards in global competitiveness, while multimodal generation has achieved parity or even taken the lead in certain niche areas," Yan Yijun, vice president of Chinese AI start-up MiniMax, told the Global Times on Monday.

Several Shanghai-based AI enterprises, including MiniMax, have contributed widely welcomed and extensively adopted models to the open-source community, Yan said, noting that "this proves one thing: great models can be built not by stacking computing power, but by relying on efficiency and innovation."

Liu Gang, chief economist at the Chinese Institute of New Generation Artificial Intelligence Development Strategies, told the Global Times on Monday that the trend is inevitable as Chinese LLMs deliver frontier-level performance in critical areas like coding while driving input and output costs down to a fraction of previous rates, creating a massive competitive moat.

DeepSeek V4 Flash, on the cheapest endpoint, costs $0.09 input / $0.18 output per million tokens. For comparison, GPT-5.5 is currently priced at $5 input / $30 output per million tokens, according to an article on OpenRouter in June.

"In addition to the strength of Chinese LLMs in cost effectiveness, another key advantage is their open-source strategy, which breaks down deployment barriers. Global developers can freely download, deploy locally, and customize these models for secondary development. For data-sensitive enterprises, open-source models allow private on-premise deployment, ensuring that data never leaves the local domain and thus ensuring data sovereignty," Chen said.

"More foreign companies from other countries and regions including Africa and India are expected to adopt Chinese LLMs," Liu said.