NEED TO KNOW

  • White House approved a secret $9 billion request for intelligence agency AI chips
  • Pentagon designated Anthropic a supply chain threat; chief of staff authorized NSA use anyway
  • Congress has not appropriated the $9 billion; $800 million is being reprogrammed now

WASHINGTON, DC (TDR) — The White House has approved a secret $9 billion request to buy the chips US spy agencies need to run frontier AI models, per current and former officials cited by The New York Times.

The big picture: Spy agencies underbuilt the data centers needed for frontier AI on classified networks, and the $9 billion is meant to close the gap.

Freedom-Loving Beachwear by Red Beach Nation - Save 10% With Code RVM10

MORE NEWS: France Bars Ben-Gvir as Netanyahu Distances From His Own Minister

  • Funding targets Nvidia Grace Blackwell superchip infrastructure, which needs specialized power and liquid cooling
  • $800 million is being reprogrammed without waiting for Congress
  • Most classified AI runs on AWS, which committed $50 billion to government cloud last year

Why it matters: A chief-of-staff authorization is routing the intelligence community around a Pentagon security designation, with the public learning through leaks rather than appropriations debate.

  • Pentagon formally designated Anthropic a supply chain risk in March, a label historically reserved for foreign adversaries
  • WH chief of staff Susie Wiles authorized NSA to keep using Anthropic's model
  • Anthropic and the government are finalizing a classified agreement
  • DOD AI contracting hit $90.7B in potential value in 2026, a 1,605% jump from 2024

Driving the news: The standoff predates the $9 billion. Anthropic sued the Pentagon after refusing to drop policy against mass surveillance and autonomous weapons; a judge granted a preliminary injunction on a related ban in April.

CLICK HERE TO READ MORE FROM THE THE DUPREE REPORT

Following ongoing debates over border security and immigration policy in 2026, do you support stricter enforcement measures?

By completing the poll, you agree to receive emails from The Dupree Report, occasional offers from our partners and that you've read and agree to our privacy policy and legal statement.
  • The $800 million reprogramming is administrative; no authorizing legislation needed
  • Wiles's NSA authorization predates any congressional briefing
  • OpenAI and xAI struck their own Pentagon classified-network deals after Anthropic's designation
  • House Intel ranking Democrat Jim Himes publicly backed early intel-agency AI access

What they're saying:

  • Jim Himes, House Intel ranking Democrat — "It would be insane for U.S. intelligence agencies to not have early access to advanced AI models"
  • Dario Amodei, Anthropic CEO, on Pentagon designation — "We do not believe this action is legally sound, and we see no choice but to challenge it in court"
  • Dan Meyer, national security law partner at Tully Rinckey — "You can't just ban a company from doing business unless there's some reason to do it"

Yes, but: The accountability problem is the secrecy, and the framing that fits one tribe collapses on the other.

Close

MORE NEWS: Pardon, Purge, Pay: DOJ Moves to Erase Jan. 6 in Three Acts

  • The "win the AI race" frame doesn't explain why the Pentagon's threat designation is being routed around without public process
  • The "rein in surveillance AI" frame doesn't explain why congressional appropriation is deferred while the workaround proceeds
  • A classified agreement plus a chief-of-staff carve-out forecloses normal review
  • Pentagon and NSA hold opposite security calls on the same vendor

Between the lines: Two parts of the same executive branch reached opposite conclusions on a vendor's security risk, and neither resolved the conflict before the spending decision. The public learned of a classified deal through a leak. The $9 billion has not faced an appropriations vote. The review that would normally test which side is right has been deferred.

What's next:

  • Congress must appropriate the $9 billion before the chip buildout proceeds at scale
  • The classified Anthropic-government agreement is being finalized, with no public timeline
  • Anthropic's lawsuit against the Pentagon remains active
  • The White House is weighing an executive order on whether Commerce or the IC evaluates AI models

If two parts of the same executive branch reach opposite security calls on the same vendor, and the disagreement is settled by a chief-of-staff override behind classification, what's left for Congress to do?

Sources

This report was compiled using reporting from the Arkansas Democrat-Gazette syndicating The New York Times, GV Wire, NBC News, CNBC, Brookings, and Nextgov/FCW.

Editor's note: TDR's research and drafting workflow uses Anthropic's Claude AI. See AI Policy.

Freedom-Loving Beachwear by Red Beach Nation - Save 10% With Code RVM10