Ask Onix
Pinterest adopts Chinese AI for smarter shopping suggestions
Pinterest is leveraging open-source AI models from China to refine its recommendation engine, a shift that reflects broader industry trends favoring cost-effective and customizable alternatives to U.S. proprietary systems.
Why Pinterest turned to Chinese AI
CEO Bill Ready revealed the platform now integrates models like DeepSeek R-1, released in January 2025, to enhance its AI-driven shopping assistant. The move follows what Ready termed the "DeepSeek moment"-a breakthrough sparked by the model's open-source release, which triggered a wave of similar tools from Chinese developers.
Competitors like Alibaba's Qwen and Moonshot's Kimi have also gained traction, alongside ByteDance's efforts in the same space. Pinterest's Chief Technology Officer, Matt Madrigal, highlighted the appeal of these models: they can be freely downloaded and tailored, unlike most U.S. offerings, such as OpenAI's ChatGPT.
"Open-source techniques we use to train our in-house models are 30% more accurate than leading off-the-shelf models,"
Matt Madrigal, Pinterest CTO
Cost and performance drive adoption
Madrigal noted the financial advantage: Chinese models can cut costs by up to 90% compared to proprietary U.S. systems. This affordability has attracted other major U.S. companies, including Airbnb, which relies on Alibaba's Qwen for its AI customer service agent. Airbnb CEO Brian Chesky cited three reasons for the choice: "very good, fast, and cheap."
Hugging Face, a hub for AI model downloads, has seen Chinese models dominate its trending lists. Jeff Boudier, a product lead at the platform, observed that Chinese labs often occupy four of the top five spots in weekly rankings. In September 2025, Qwen surpassed Meta's Llama as the most downloaded large language model family on the site.
U.S. firms play catch-up
Meta's Llama models, once the go-to for developers, lost ground after the release of Llama 4 in 2025, which underwhelmed users. Reports suggest Meta is now collaborating with Alibaba, Google, and OpenAI to train a new model set for release this spring.
A Stanford University report from December 2025 found Chinese AI models have either caught up to or surpassed global counterparts in capability and adoption. Former UK Deputy Prime Minister Sir Nick Clegg, who previously led global affairs at Meta, criticized U.S. firms for fixating on "superintelligence"-a vague goal that has ceded ground to China's open-source dominance.
"Here's the irony: In the battle between 'the world's great autocracy' and 'the world's greatest democracy,' China is doing more to democratize the technology they're competing over."
Sir Nick Clegg, former Meta executive
Profit pressures reshape U.S. strategy
U.S. companies like OpenAI face mounting pressure to monetize. After releasing two open-source models in summer 2025-its first in years-OpenAI has prioritized proprietary systems to boost revenue, including a push into advertising. CEO Sam Altman acknowledged aggressive investments in computing power and infrastructure, framing it as a long-term bet.
"Revenue will grow super fast, but expect us to invest heavily in training the next model-and the next."
Sam Altman, OpenAI CEO
Data security remains a priority
Despite reliance on Chinese models, companies like Airbnb emphasize data security. The firm hosts AI models on its own infrastructure, ensuring user data never reaches third-party developers. This hybrid approach-combining U.S. and Chinese models-reflects a pragmatic balance between cost, performance, and control.
