In this image, the Anthropic logo is seen on a smartphone with the Claude Mythos logo in the background.
Images of Sopa | Lightrocket | Getty Images
Anthropic’s suspension of its top AI models late last week hammered home a harsh reality for the companies that relied on them: access can be cut off at any time.
It’s been a big theme this week, and Wall Street will be watching it closely as Anthropic and OpenAI prepare for potentially huge IPOs in the coming months.
Microsoft CEO Satya Nadella warns of the risks, as his company is a major investor in OpenAI, and backed Anthropic last year in a multibillion-dollar round. He wrote Monday in a post on X that companies need to “build agent systems that evolve over time, while still maintaining control over their IP.”
“The last thing any of us wants is a world where every company in every sector assigns value to a few models that consume everything in sight,” Nadella wrote.
Investors have been trading on that theme. MiniMax again ZipChina’s open-source AI lab, both exploded on Monday as Anthropic’s battle put the spotlight on downloadable models that companies can run themselves.
The Anthropic-Fable news came at a bad time on Friday. About two hours earlier, SpaceX has completed its first day of trading following the largest IPO on record. SpaceX’s XAI unit is a niche player in artificial intelligence, but CEO Elon Musk is an open voice on the topic.
Anthropic announced that it has withdrawn access to its Fable 5 and Mythos 5 models to comply with an export control order from the US government that cited “national security authorities.” Anthropic has already disabled the models for all of its customers to ensure compliance, but said that all of its other models will not be affected.
For developers who want full control of model access, there is another way. The open source model can be downloaded, run with the company’s infrastructure, and customized for its data and needs. If the model lives on the company’s own servers, no political infighting can shut it down.
Yash Patel, CEO of Applied Compute, which helps companies train and deploy custom models, said Anthropic’s battle “highlighted the importance of having your own model.” He said flexibility has become more common of late.
“What we’ve been hearing more and more, probably more in the last month than in the entire year, is the fact that they want a multi-dimensional future,” Patel said. “They don’t want to be locked into one vendor.”
That could be a problem for the US, as open-source models are gaining traction from China, just as the world’s two largest economies are in a battle to control the future of AI.
Models from DeepSeek, TencentXiaomi, and MiniMax all ranked among the most used OpenRouter this month, even when competitors closed. Zipu has filed its latest release as a countermeasure to Washington. Cutting-edge AI, the company argued, should not be owned by a few players or withdrawn at will.
Costs can also speed up adoption. As the price of modern AI rises, companies are already doing routine work on cheap models and saving more expensive ones for more complex tasks.
Patel said customers are responding to what he calls the “token-pocalypse” as AI products move toward usage-based pricing.
“The time for tokenization is over,” he said. Companies are now looking for “better, cheaper, faster models.”
That’s pushing some businesses to rethink models they would have discarded months ago, including China’s open-source models.
Before it was like I don’t even want to talk about it,” said Patel. “Now they’re like, okay, how good can it be, and if it’s good, we’ll figure it out.”
It’s a reminder that the AI market is still young, as ChatGPT’s public release happened less than four years ago. For investors, that reshapes who is leading the AI race. The winners may not just be large closed model labs, despite what their current ratings suggest.
WATCH: Emil Michael of DOD on Anthropic Action
