This illustrated photo shows the Anthropic logo on the smartphone and the Claude Mythos logo in the background.
Sopa Images | Light Rocket | Getty Images
Anthropic’s suspension of its top AI model late last week was a harsh reminder for companies that relied on AI models that access could be cut off at any time.
This is a big theme this week, and one that Wall Street is watching closely as Anthropic and OpenAI prepare for big IPOs in the coming months.
microsoft CEO Satya Nadella has warned of the risks, even though the company is a major investor in OpenAI and backed Anthropic with billions of dollars last year. In a post on X on Monday, he wrote that companies need to “maintain control of their IP while building agent systems that improve over time.”
“No one wants a world where every company in every sector cedes value to a few models that eat up everything in sight,” Nadella wrote.
Investors are trading based on that theme. mini max and ChipuChina’s open-source AI lab, and Anthropic both soared on Monday as the battle spotlighted downloadable models that companies can run on their own.
The news of the human fable came late Friday at an awkward moment. About two hours ago space x The company closed its first day of trading following the largest IPO in history. SpaceX’s xAI division is a niche player in the artificial intelligence space, but CEO Elon Musk has been outspoken on the subject.
Anthropic announced that it has discontinued access to its Fable 5 and Mythos 5 models to comply with export control directives from the U.S. government citing “national security authorities.” Anthropic suddenly disabled the model for all customers to ensure compliance, but said all other models were unaffected.
There is another approach for developers who want complete control over access to their models. You can download open source models, run them on your own infrastructure, and customize them to fit your data and needs. If a model resides on a company’s own servers, it cannot be stopped by political battles.
Yash Patel, CEO of Applied Compute, which helps companies train and run custom models, said the battle for humanity “highlighted the importance of owning your own models.” He said this change has become much more mainstream these days.
“What we’re hearing more and more is the fact that they want a multimodal future, probably more in the last month than all of this year,” Patel said. “They don’t want to be tied to a single vendor.”
This could pose a problem for the United States, as the world’s two largest economies vie over control of the future of AI, as the open model that wins the most adoption is made in China.
DeepSeek model, tencentXiaomi, and MiniMax all rank among the most used this month on OpenRouter, even compared to private competitors. Zhipu framed the latest release as a kind of rebuttal to Washington. The company argued that cutting-edge AI should not belong to a few players or be freely withdrawn.
Cost may also speed up implementation. As the price of cutting-edge AI rises, companies have already begun to allocate routine tasks to cheaper models and save the most expensive models for the most difficult tasks.
Patel said customers are reacting to what he calls a “token apocalypse” as AI products move to pay-as-you-go pricing.
“The days of token maximization are over,” he said. Companies are now looking for “better, cheaper, faster models.”
This is forcing some companies to reconsider models they would have dismissed a few months ago, including open models made in China.
“Before, it was like I didn’t want to talk about it,” Patel said. “Now they’re like, ‘OK, how good could it be? And if it’s good, we’ll figure it out.'”
This is a reminder that the AI market is still in its infancy, as ChatGPT has been publicly available for less than four years. For investors, it reframes who is actually leading the AI race. Despite what current valuations suggest, the winners may not be just large, closed modeling labs.
WATCH: Defense Department’s Emil Michael talks about human action
