Chinese AI models, particularly Qwen, DeepSeek, and Kimi, are significantly impacting the market by being up to 50 times cheaper than their American counterparts, as highlighted in a recent J.P. Morgan report. This pricing pressure is evident as Chinese firms accounted for over 45% of traffic on the AI platform OpenRouter by April 2026, a sharp increase from under 2% in late 2024. Companies are now adopting strategies to optimize their AI spending, moving towards cheaper models and implementing model routing systems that allocate simpler tasks to these lower-cost options while reserving more complex tasks for premium models.

UBS: UBS is a global financial services firm that produces research on technology trends. The news draws on UBS analysis showing widespread corporate moves toward cheaper and open-source models in response to high AI spending.
Kimi: Kimi is an AI model from Moonshot AI known for strong long-context performance. According to the report covered in the news, Kimi exemplifies the Chinese models gaining rapid adoption on aggregation platforms and pressuring established US AI companies on pricing.
Meta: Meta designs custom AI hardware for its data centers and open-source model ecosystem. The report notes Meta’s chips as contributors to the trend of hyperscalers building specialized silicon to manage AI expenses more efficiently.
Qwen: Qwen is a family of large language models developed by Alibaba. In this news, Qwen is highlighted as one of the leading Chinese models offering significantly lower costs per token, contributing to pricing pressure on US providers and increased traffic share on platforms like OpenRouter.
Amazon: Amazon offers custom AI chips through its AWS infrastructure services. In the context of the news, Amazon’s silicon initiatives are cited as helping enterprises lower total costs compared with reliance on general-purpose accelerators.
Google: Google develops custom AI accelerators including TPUs for its cloud and internal workloads. The report highlights Google’s custom silicon as part of a broader industry shift toward in-house chips that reduce overall AI infrastructure expenses.
NVIDIA: NVIDIA supplies the dominant GPUs and accelerators used for AI training and inference. The news states that while NVIDIA maintains leadership in AI hardware, custom chips from other firms are eroding some of its cost advantages in the ecosystem.
OpenAI: OpenAI is a leading developer of frontier AI systems including the GPT series. The news notes that OpenAI faces direct pricing pressure from lower-cost Chinese models that are capturing substantial usage on third-party platforms.
DeepSeek: DeepSeek develops open-source AI models focused on efficient reasoning and coding capabilities. The news positions DeepSeek as a key Chinese contender that is helping shift enterprise workloads toward more affordable alternatives amid rising AI expenses.
Anthropic: Anthropic builds constitutional AI models such as the Claude family. The report indicates Anthropic is among the US providers whose pricing is being challenged by the rise of capable and inexpensive Chinese alternatives.
Microsoft: Microsoft integrates custom AI accelerators into its Azure cloud and works closely with OpenAI. The news references Microsoft’s custom chips among those gaining traction by delivering meaningful cost reductions for large-scale AI deployments.

Adoption: Chinese open-source models are increasingly integrated into enterprise workflows through local deployment or cloud catalogs.
Hardware: Major technology firms are accelerating development of custom AI chips to achieve meaningful reductions in total infrastructure costs.
Strategy: Companies are implementing model routing systems to direct routine tasks to lower-cost models while reserving premium models for complex work.