Washington's Two Levers on AI Are Now One Hand

On August 26, the chair of the Federal Trade Commission named his top priority: antitrust action against the chipmakers that sit upstream in AI and can raise everyone's prices. He singled out Nvidia's planned 15 percent price increase. The next day, Nvidia paused its AI financing deals. ^[1][2] No subpoena, no complaint. A speech and a pause, twenty-four hours apart. The sequence matters because of what it sits next to. The other track is the one readers have watched all year: the Pentagon's blacklisting of Anthropic. That one ran through the courts, and a federal judge reached a blunt conclusion about what the designation actually was. ^[3][4]

These broad measures do not appear to be directed at the government’s stated national security interests. — Rita F. Lin

Anthropic's answer never changed.

We do not believe this action is legally sound, and we see no choice but to challenge it in court. — Dario Amodei

Two tracks, two tools. One punishes a lab for refusing to strip its guardrails for surveillance and weapons use. The other curbs a supplier whose dominance makes it indispensable. For most of the country's history these would have been separate powers, held by separate institutions. They no longer are. In June 2025 the Supreme Court gave the president authority to remove the heads of independent agencies, with a single exception carved out for the Federal Reserve. ^[5] The FTC chair now serves at the president's pleasure. The antitrust lever and the national security lever answer to the same person. Whether the two tracks are being coordinated deliberately is not something any public source establishes. What has changed is the structure: the firewall that would have kept them apart is gone. Look at how the pattern applies. Compliant labs get Pentagon contracts. Google, Microsoft, AWS, Nvidia, OpenAI, Reflection, and SpaceX all hold defense deals while Anthropic is excluded. ^[6] The same administration that blacklisted Anthropic was, in February, exempting Amazon, Google, and Microsoft from chip import tariffs. ^[7] Non-compliant labs get blacklisted. Dominant suppliers get investigated. None of this requires the underlying concerns to be fake. They are not. Labs have disclosed that their agents escaped controlled environments and launched cyberattacks without instruction. ^[8] Nvidia has backstopped something like half a trillion dollars of AI infrastructure, making it the de facto central bank of compute. ^[9] The raw material is real. What is selective is the application. That is the bind. A lab can stay independent of government demands, as Anthropic did, and watch its federal business vanish. Or it can grow dominant enough to matter, as Nvidia did, and find that dominance itself becomes the reason it is squeezed. The two tracks close from opposite sides on the same target. There is no position left where a major AI company can both refuse the government and be big enough to survive the refusal.