The Intelligence Community Has an AI Problem — and It Costs $9 Billion

America's spy agencies are falling behind on artificial intelligence. Not because of bad intentions. Because they don't have enough chips to run the software.

According to the New York Times, the White House has approved a secret $9 billion request to acquire cutting-edge computer chips for U.S. intelligence agencies — specifically Nvidia's Grace Blackwell superchip. The CIA and NSA cannot fully deploy the latest AI models on their classified systems because their existing infrastructure can't handle the computational load.

The most powerful intelligence apparatus on earth is running on hardware that can't keep up with software that civilians are using for homework.

What $9 Billion Actually Buys

The Grace Blackwell superchip doesn't just need electricity. It needs specialized liquid cooling systems and data centers built to supply enormous amounts of electrical energy, according to the NYT. This isn't buying a laptop. It's building entirely new classified infrastructure from scratch.

Congress still must authorize the full $9 billion. But the administration isn't waiting — it's already reprogramming $800 million for a more rapid acquisition of computing capacity. That reprogramming doesn't require a congressional vote. It's happening now.

Experts told the NYT that even $9 billion probably won't be enough. Larger sums will likely be needed in future years as AI model requirements keep scaling up.

The Anthropic Problem

Because the chip shortage is so acute, White House Chief of Staff Susie Wiles authorized the NSA to keep using an advanced AI model built by Anthropic — even though the Pentagon has designated Anthropic a supply chain threat, according to current and former U.S. officials cited by the NYT.

The Pentagon flagged Anthropic as a supply chain risk. Susie Wiles overruled that designation to keep the NSA running.

Anthropic and the U.S. government are currently finalizing a classified contract allowing the NSA to maintain access to Anthropic's products. Earlier this year, the Defense Department demanded the authority to use Anthropic's technology for, quote, "any lawful use" — setting off a fight between the Pentagon and the company. The new contract reportedly drops that language.

The spy agencies are locked into a vendor the Pentagon considers a threat, because they don't have enough chips to use anything else. This reflects a core national security dependency with significant consequences.

The Turf War Inside the Administration

According to the Washington Post, there's a sharp internal split inside the Trump administration over who gets to call the shots on AI policy.

U.S. intelligence agencies are pushing for a bigger role in evaluating AI models — more power than the Commerce Department currently holds. This fight is playing out as President Trump prepares to travel to a summit in China, meaning the administration is debating internal AI governance while simultaneously trying to project AI dominance on the world stage.

Two people familiar with the matter told the Washington Post the proposal isn't yet public.

The Broader Federal AI Picture

This isn't happening in a vacuum. Brookings Institution researchers James Denford, Gregory Dawson, and Kevin Desouza published an analysis on May 18, 2026, documenting that federal AI spending is on a sharp upward trajectory, overwhelmingly concentrated at the Department of Defense.

In their July 2025 AI Action Plan, the Trump administration laid out three pillars: accelerating AI innovation, building U.S. AI infrastructure, and leading international AI diplomacy. The Brookings researchers confirmed that analysis of federal AI contracts shows the government is still largely in an experimental phase — though larger vendors are entering the market and longer multi-year contracts are replacing short pilots.

The government is spending more, but doesn't fully know what it's doing yet.

Coverage and Context

The NYT and Washington Post both covered pieces of this story. Neither connected the dots.

The Anthropic supply-chain-threat angle is buried in the NYT piece. The government is paying a company the Pentagon flagged as a risk because it has no other viable option. This represents a structural failure, not a bureaucratic footnote.

The Washington Post's framing leans hard into the turf battle narrative, treating this as an interesting political story. The real issue is whether the intelligence community or the Commerce Department sets the standards for which AI systems get classified access — a decision with consequences for decades.

Brookings provides the most sober read: federal AI spending is growing fast, but the government remains in an immature, chaotic phase where it's still figuring out what works.

The Bottom Line

Nine billion taxpayer dollars are about to flow — most of it to Nvidia — to fix a problem that should have been anticipated years ago. The chip crunch wasn't a secret. AI compute requirements doubling every few months wasn't a secret. The intelligence community watched it happen and didn't build the infrastructure to keep up.

Now you're paying for the catch-up. And while that bill gets sorted out, America's spy agencies are dependent on a vendor the Pentagon considers a supply chain threat. No one in charge seems particularly bothered by that contradiction.