Elon Musk said Dario is right after Anthropic chief Dario Amodei called on frontier AI companies to slow down together. That should concern anyone who still believes competition matters. Musk presented Grok as an open-source alternative while attacking OpenAI for becoming closed-source and profit-driven. Amodei built Anthropic around safety, but the company later dropped its flagship pledge not to train more powerful systems without adequate safeguards already in place. Musk has endorsed Amodei’s call that would let the largest AI companies help decide how fast the field can move and who gets to move with them. When the companies with the most money, computing power, and government access write the limits, safety can turn into market control very quickly.
Anthropic is not merely offering advice here on X (formerly Twitter). Federal disclosures show it spent $3.53 million on lobbying during the first half of 2026. The company also gave Public First Action another $20 million, bringing its total support to $40 million. Anthropic says that money supports public education and sensible safeguards. It still gives one company an enormous voice in deciding what sensible means. Amodei also wants companies to coordinate on common standards with government mediation or antitrust waivers. That should concern any free market analysis, which is not small request, It would place firms that benefit from high barriers inside the process that creates those barriers.
Musk is not outside this concentration of power either. In May 2026, Anthropic agreed to pay SpaceX $1.25 billion per month for access to approximately 325,000 Nvidia GPUs across its Colossus facilities. The filing says the rate was reduced during May and June and that either company could terminate the agreements after the initial three months with 90 days’ notice. Musk is now supporting Amodei’s proposed slowdown while his company supplies Anthropic with computing infrastructure it uses to remain at the frontier. SpaceX’s own filing describes a vertically integrated structure combining Grok, proprietary access to real-time data through X, physical computing infrastructure, Starlink and launch capabilities. It says this structure gives the company a substantial cost and speed advantage. Grok also has a federal market through an agreement making it available to government agencies for 42 cents per organization through March 2027. Musk can criticize closed systems in public while building one of the most vertically integrated systems in the industry. This is what market control looks like in deployment.
The international damage could be worse than the domestic damage if this choking the United State industry goes undisputed. I have already warned in FPIF that Washington’s AI export controls could not hold and that the Pentagon’s AI rush was creating a security weakness. Washington keeps confusing control with strength. Stanford's 2026 AI Index found that the gap between the best American and Chinese models had narrowed to 2.7 percent by March 2026. China already leads in AI publications, citations, and patent grants. The United States has more capital and more leading models, but China is building something just as important. And the United States is now watching two giants in the AI industry openly supporting coordination that could produce market control.
Chinese firms are pushing cheap open-weight models that developers can download, change, and run locally. The U.S.-China Economic and Security Review Commission found that Alibaba’s Qwen ecosystem included more than 100,000 derivative models. Wider adoption produces more testing, more applications, and more opportunities for further adaptation. If American rules raise costs and lock smaller firms out, developers abroad will not wait for Washington’s permission, the United States can’t maintain being a superpower with this arrogant display for market control. Governments and businesses in Africa, Asia, Latin America, and the Middle East will choose models based on price, language support, local control, and whether they can keep their own data. A cheaper Chinese system they can adapt may look better than an American system they can only rent.
This is how the United States regulates itself into second place. Safety matters, but not when it becomes theater or lets companies act as private governments over their own market. Evaluators should answer to public institutions, costs should follow demonstrated risk, and vendors should not write the standards used to choose vendors. American strength came from new firms challenging old ones. If Washington replaces that with a protected council of frontier companies, Anthropic and xAI could gain long-term control of the American market while China gains developers, markets, and a path toward global AI dominance.
