What rare earths and chip controls predict about the Fable shutdown

We believe that excessive regulation of the AI sector could kill a transformative industry just as it’s taking off, and we’ll make every effort to encourage pro-growth AI policies.

— Vice President JD Vance, Paris AI Action Summit, February 11, 2025

The US administration’s decision to shut down Claude Mythos 5 and Fable 5 on Friday doesn’t just jeopardize Anthropic. It jeopardizes the future of the entire US AI industry. Because when people realize that it takes just the stroke of a pen to cut them off from an essential resource, they start planning for a future where they are no longer vulnerable.

In October 2022, the US instituted export controls on advanced chips, EUV lithography and chipmaking tools. These controls were expanded in October 2023 and again in March 2025 by the Trump administration, which also blacklisted dozens of additional Chinese entities.

China’s response has been a focused drive towards self-reliance, investing unprecedented amounts of capital to build indigenous chipmaking technology. In 2024, China launched its third “National Integrated Circuit Industry Investment Fund”, raising 344 billion yuan, approximately $47.5 billion, the largest single tranche of state capital ever directed at a technology sector. That same year, Huawei produced its Kirin 9000S: a 7nm chip built domestically by SMIC without EUV lithography, the kind of chip Western experts said China couldn’t make.

Huawei is now scaling its Ascend 910C AI accelerator (a direct competitor to Nvidia’s restricted H100s/H200s), targeting 600k units in 2026. In March 2025, eight Chinese government agencies jointly endorsed a national pivot to RISC-V, an open-source chip architecture designed to eliminate dependence on Western-controlled Arm and x86 instruction sets.

Chinese chipmaking still lags the West, but there is every reason to believe that their efforts for chip independence will ultimately prove successful.

China’s restrictions on rare earths provide another example. China restricted gallium and germanium exports in mid-2023, added antimony in September 2024, and on December 3, 2024 banned the export of gallium, germanium, antimony and superhard materials to the United States. In April 2025 China responded to US trade actions by adding added seven more rare earths to its control list.

It subsequently walked back some of its controls after the US made concessions on trade. But the US didn’t simply lose the trade war and then go back to business as usual.

Once a strategic vulnerability like this has been seen, it cannot be unseen.

The Pentagon took a stake in MP Materials, making it the company’s largest shareholder. The Big Beautiful Bill contains roughly $8.5 billion in new materials funding, and the US has signed agreements and MOUs for minerals with Australia, Japan, Malaysia and Thailand.

Now consider the Fable debacle. According to OpenAI CEO Sam Altman, artificial intelligence will be like a “utility”: “We see a future where intelligence is a utility like electricity or water and people buy it from us on a meter and use it for whatever they want to use it for”.

It will be, in other words, an essential resource for people and businesses alike. This belief - that AI labs are creating the next most important foundation of modern life - drives the soaring valuations of US AI companies.

What we saw on Friday is something we cannot unsee: access to this essential resource is intensely vulnerable to the whims of the American president, and short of an act of Congress, that vulnerability will persist.

It’s not just about losing access to your preferred coding or scientific research tool. These technologies are foundational in modern digital products. Tediware’s AI features use Claude Sonnet 4.6, with the option of Opus 4.7 for more complex tasks, but because of intermittent outages with Anthropic services, Tediware is designed to fall back to Codex if Claude is unavailable.

Now there’s a new source of instability: the risk of losing access to any of these models with virtually no warning (Anthropic was given just 90 minutes to shut down its newest and most powerful model).

Clearly, we’re all going to start designing non-US AI model fallbacks into our products. We’ll have to test them to make sure they work well, of course. And if we learn that they do, in fact, work quite well? Maybe the model selection dropdown starts to default to a non-US AI model. Maybe those models go to the bottom of the list…and then start to disappear from the list altogether.

The reason that’s possible is because there’s a major difference between AI models and chips or rare earths: the former is substitutable, the latter are not. It takes years of investment to build a new rare earths supply chain. It takes just a few minutes to route chat completions to a different model.

Vance warned that “excessive regulation” of the AI industry could kill it. What actually happened is worse: instead of regulation, we have retribution. The US AI industry is far from dead, but this self-inflicted wound drew blood. Regulation - an apolitical, scientific and evidence-based process of model review and certification - is now the only thing that will stop the bleeding.