Lily JamaliNorth America Technology correspondent
Getty Images
Every month, tons of of thousands and thousands of customers flock to Pinterest searching for the newest kinds.
One web page titled “the most ridiculous things” is stuffed with loads of wacky concepts to encourage creatives. Crocs repurposed as flower pots. Cheeseburger-shaped eyeshadow. A gingerbread home manufactured from greens.
But what would-be patrons could not know is the tech behind this is not essentially US-made. Pinterest is experimenting with Chinese AI fashions to hone its advice engine.
“We’ve effectively made Pinterest an AI-powered shopping assistant,” the agency’s boss Bill Ready informed me.
Of course, the San Francisco-based tastemaker might use any variety of American AI labs to energy issues behind-the-scenes.
But since the launch of China’s DeepSeek R-1 mannequin in January 2025, Chinese AI tech has more and more been part of Pinterest.
Ready calls the so-called “DeepSeek moment” a breakthrough.
“They chose to open source it, and that sparked a wave of open source models,” he stated.
Chinese rivals embrace Alibaba’s Qwen and Moonshot’s Kimi, whereas TikTok proprietor ByteDance can also be engaged on comparable expertise.
Pinterest Chief Technology Officer Matt Madrigal stated the energy of those fashions is that they are often freely downloaded and customised by corporations like his – which isn’t the case with the majority of fashions provided by US rivals like OpenAI, which makes ChatGPT.
“Open source techniques that we use to train our own in-house models are 30% more accurate than the leading off-the-shelf models,” Madrigal stated.
And these improved suggestions come at a a lot decrease value, he stated, typically ninety % lower than utilizing the proprietary fashions favoured by US AI builders.
‘Fast and low-cost’
Pinterest is hardly the solely US enterprise relying on AI tech from China.
These fashions are gaining traction throughout an array of Fortune 500 corporations.
Airbnb boss Brian Chesky informed Bloomberg in October his firm relied “a lot” on Alibaba’s Qwen to energy its AI customer support agent.
He gave three easy causes – it is “very good”, “fast” and “cheap”.
Further proof could be discovered on Hugging Face, the place individuals go to obtain ready-made AI fashions – together with from main builders Meta and Alibaba.
Jeff Boudier, who builds merchandise at the platform, stated it’s the value issue that leads younger start-ups to take a look at Chinese fashions over their US counterparts.
“If you look at the top trending models on Hugging Face – the ones that are most downloaded and liked by the community – typically, Chinese models from Chinese labs occupy many of the top 10 spots,” he informed me.
“There are weeks where four out of five top training models on Hugging Face are from Chinese labs.”
In September, Qwen topped Meta’s Llama to change into the most downloaded household of enormous language fashions on the Hugging Face platform.
Meta launched its open-source Llama AI fashions in 2023. Up till the launch of DeepSeek and Alibaba’s fashions, they had been thought of the go-to alternative for builders engaged on bespoke functions.
But the launch of Llama 4 final yr left builders underwhelmed, and Meta has reportedly been utilizing open-source fashions with Alibaba, Google, and OpenAI to coach a brand new mannequin set for launch this spring.
Airbnb additionally makes use of a number of fashions, together with US-based ones, internet hosting them securely in the firm’s personal infrastructure. The information is rarely supplied to the builders of the AI fashions they use, in line with the firm.
Chinese success
Going into 2025, the consensus was regardless of billions of {dollars} being spent by US tech corporations, Chinese corporations had been threatening to tug forward.
“That’s not the story anymore,” Boudier stated. “Now, the best model is an open-source model.”
A report revealed final month by Stanford University discovered Chinese AI fashions “seem to have caught up or even pulled ahead” of their international counterparts – each when it comes to what they’re able to, and the way many individuals are utilizing them.
In a current interview with the BBC, former UK deputy prime minister Sir Nick Clegg stated he felt US corporations had been overly targeted on the pursuit of AI which can at some point surpass human intelligence.
Last yr, Sir Nick left his submit as head of world affairs at Meta, the developer of Llama. Boss Mark Zuckerberg has dedicated billions of {dollars} to attaining what he calls “superintelligence.”
Some specialists are actually calling these ambitions obscure and ill-defined – giving China a gap to dominate the open-source AI house.
“Here’s the irony,” Sir Nick stated. In the battle between “the world’s great autocracy” and “the world’s greatest democracy” – China and America – China is “doing more to democratise the technology they’re competing over”.
The Stanford report additionally recommended China’s success in creating open-source fashions may very well be partly defined by authorities assist.
On the different aspect of the world, US corporations like OpenAI are below intense stress to extend income and change into worthwhile – and is now turning to advertisements to assist get there.
The firm launched two open-source fashions final summer time – its first in years. But it has poured most of its assets into proprietary fashions to assist it become profitable.
OpenAI boss Sam Altman informed me in October it has invested aggressively into securing ever extra computing energy and infrastructure offers with companions.
“Revenue will grow super fast, but you should expect us to invest a ton in training, in the next model and the next and the next and the next,” he stated.
