re <<data>>
bloomberg.com
China’s Embrace of Open-Source AI Will Tip the Scales
Beijing’s push for accessible AI is self-serving. But it has some key lessons for the US.
7 March 2025 at 04:00 GMT+8
By Catherine Thorbecke
Catherine Thorbecke is a Bloomberg Opinion columnist covering Asia tech. Previously she was a tech reporter at CNN and ABC News.
China's open-source AI revolution is here.
Photographer: Sheldon Cooper/SOPA Images/LightRocket/Getty Images
More than anything, China’s embrace of open-source artificial intelligence systems will be what tips the scales in its favor in the heated tech race with the US.
Washington and Silicon Valley now have a limited window to adapt and respond, or risk a future where Chinese AI overwhelmingly powers the products and applications used by the global industries of tomorrow.
The DeepSeek shock earlier this year provided the first wake-up call, after the Hangzhou-based startup released powerful open-source models built at a fraction of the cost of US tools and allowed developers to freely build off of them. But the second came this week, in the closely read policy report released at the National People’s Congress in Beijing.
The National Development and Reform Commission’s annual work statement mentioned AI nine times — but perhaps more significantly, went on to declare that China will develop a system of open-source models. It’s as clear a signal as it gets that Beijing is throwing its entire might behind the push to make the building blocks of AI available for the public to use and modify. Making the source code available allows more people to access and cheaply implement the technology, rather than leaving it to a handful of tech companies. The position stands in contrast with the direction that many US tech leaders, spearheaded by OpenAI, are going with their expensive and opaque proprietary models.
It may seem ironic that China, with its global reputation for tight restrictions on speech and content, may ultimately expand access to this technology. It’s true that the nation’s AI models still face tight restrictions; specifically that outputs must adhere to Chinese Communist Party views and censorship policies.But it’s a highly strategic choice. State-backed media has had a field day promoting this position, especially when the Trump administration’s foreign policy chaos is giving Beijing a unique moment to assert leadership on the global stage. Given Beijing has had to navigate sweeping restrictions on access to advanced chips and other US measures aimed at curtailing its AI ambitions, this approach is its best bet to stay relevant.
Officials have also made clear a broader goal. The spokesman for the nation’s top legislature earlier this week pledged that China will “prevent technological innovation from becoming a game for rich countries and the wealthy,” and praised DeepSeek for showcasing “inclusivity” in development. Beijing is betting its vastly cheaper and more accessible AI services will end up being what the rest of the world chooses. If Washington doesn’t pivot to policies that promote American leadership in accessible and open-source products, it could quickly see China close the technology gap, while losing swaths of global customers to firms under the sway of Beijing.
Some of the most prominent Chinese tech companies have recently announced that they will allow developers access to their top models as momentum swelled after DeepSeek burst the floodgates. Earlier this week, influential startup Zhipu AI said it planned to release multiple new open-source models. Fellow startups MiniMax and Moonshot have also made similar announcements. And Big Tech companies are jumping on the bandwagon: Baidu Inc. earlier this month said it would open up its latest Ernie model. Alibaba Group Holding Ltd. has long been doing so with its products, most recently with the market-moving unwrapping of its latest reasoning model this week and a pledge to make the code available for its latest video-generation models. The list goes on.
Still, some in the global tech community have criticized DeepSeek for not being fully open-source, but rather “open-weight.” But it’s offering far more transparency and less restrictions to use than Meta Platforms Inc.’s tools. This is all paving the way for AI development to emerge from tinkerers and curious coders around the globe, rather than gatekeeping it within the firm control of corporate interests.
The backbone of China’s innovation ecosystem has always been people, and this approach allows it to leverage its immense talent pool to advance broader AI ambitions. Promoting open systems increases the number of researchers from the country and beyond who can simultaneously experiment and publish new breakthroughs to improve these tools.
Critics of open-source AI have argued that it carries unique risks, heightening the potential for bad actors to exploit it. But the reality in Silicon Valley is that these threats haven’t stopped companies from speeding ahead with its rollout. Addressing the dangers requires global cooperation and thoughtful regulation; and should be done in a way that doesn’t monopolize the power of tech giants.
In an unusual admission, OpenAI CEO Sam Altman disclosed in a January Reddit post that, “I personally think we have been on the wrong side of history here and need to figure out a different open source strategy.”
It’s not too late, Sam.
If the US wants to maintain its global position in AI leadership, Washington and Silicon Valley must work together to promote American open-source AI to the globe. Inaction risks being caught flat-footed while the world embraces Chinese AI, and all the ideologies that accompany it. |
