The White House is asking for $9 billion to buy AI chips for spies.

The White House has approved a $9 billion request to buy cutting-edge AI chips for the CIA and NSA, according to reporting by The New York Times. The request, which now requires congressional approval, would fund Nvidia Grace Blackwell superchip infrastructure for the intelligence community — a clear signal that American spy agencies believe they are dangerously behind in the AI arms race.

The ask is staggering in its scale and blunt in its implication: the nation’s most powerful intelligence agencies lack the computing capacity to run modern AI models, and they need $9 billion to catch up.

But the $9 billion figure is just the entry point to a much larger conversation about surveillance, civil liberties, and the speed at which government AI capabilities are being built without corresponding oversight.

## Why the Intelligence Community Needs This Now

The CIA and NSA sit atop an ocean of data — intercepted communications, satellite imagery, financial transactions, social media posts, and signals intelligence from across the globe. For decades, the bottleneck was collection. Agencies could gather more data than they could process.

AI models, particularly large language models and computer vision systems, can turn that data ocean into actionable intelligence in real time. They can identify patterns in communications that human analysts would take months to find. They can flag suspicious financial flows across millions of transactions simultaneously. They can track individuals across surveillance cameras using gait analysis and facial recognition.

The problem: running these models requires enormous computing power. Without the hardware, the intelligence agencies have the data but cannot extract its value at speed. China’s intelligence apparatus, by contrast, has been integrating AI into surveillance and analysis at scale for years — unconstrained by the privacy considerations that at least nominally limit Western agencies.

The $9 billion request is essentially an acknowledgment that the gap has become a vulnerability.

## What $9 Billion Actually Buys

Nvidia’s Grace Blackwell superchip is the most advanced AI processor currently available. It is designed for the kind of massive parallel computation that trains and runs large language models, computer vision systems, and real-time analysis pipelines.

At roughly $30,000 to $40,000 per unit for the B200 GPU alone, $9 billion would purchase somewhere in the range of 200,000 to 300,000 chips, plus the infrastructure to house, cool, and power them. This is not a lab upgrade — it is building an entirely new computational layer for the intelligence community.

The request also includes infrastructure costs: data center construction, power supply upgrades, security hardening, and the personnel to operate it all. Intelligence officials have reportedly told Congress that current computing resources cannot handle the models needed for modern threat detection, and that each month of delay compounds the disadvantage.

## The Oversight Gap

Here is what makes this $9 billion request different from a normal defense procurement: the agencies asking for it operate with minimal public oversight.

The CIA’s budget is classified. The NSA’s budget is classified. The legal frameworks governing what AI systems can and cannot do with Americans’ data are still being written — and they are being written slowly while the technology is being deployed quickly.

This timing matters. Pope Leo XIV’s statement this weekend calling for a “slower pace” of AI adoption and warning that AI makes war “more feasible” was aimed partly at exactly this kind of development. The Vatican argued that “a more moral AI is not enough if that morality is determined by a few.” The $9 billion chip request is a case study in that concern: a few classified agencies, making decisions about surveillance capabilities that affect hundreds of millions of people, with virtually no public input.

Meanwhile, President Trump abruptly postponed a planned executive order on AI regulation just days ago, saying he “didn’t like certain aspects” and feared it could slow the US in its race against China. The message from the executive branch is consistent: speed is the priority, oversight can wait.

## The Precedent Problem

Intelligence agencies have a history of expanding surveillance capabilities beyond their original mandates. The NSA’s bulk data collection, revealed by Edward Snowden in 2013, was justified by the same logic being used now: national security requires capabilities that the public cannot be told about.

AI amplifies the surveillance capacity of those programs by orders of magnitude. A system that can process millions of communications in real time, identify patterns across disparate data sources, and generate predictive models of individual behavior is not just a faster version of existing tools. It is a fundamentally different capability.

The $9 billion request does not appear to include any provisions for oversight, auditability, or limits on domestic surveillance — though such details may exist in classified portions of the request.

## What This Means For You

A $9 billion intelligence spending request may seem remote from daily life, but the technology it funds will shape how the government interacts with your data for decades.

**Your digital footprint is about to become more legible to the government.** AI-powered surveillance does not just track terrorists. It processes patterns across populations. The same infrastructure that identifies foreign threats can be pointed inward — and historically, it has been.

**The oversight conversation is happening right now, without you.** Congress will vote on this funding. The frameworks governing how intelligence agencies can use AI are being drafted. If you care about the balance between security and privacy, this is the moment to pay attention and contact your representatives.

**AI regulation is not just about consumer tech.** The debate around AI safety often centers on chatbots and deepfakes. But the biggest AI deployments — the ones with the most power and the least oversight — are happening in the intelligence and defense communities. $9 billion is the starting bid.

**China is the justification, but the capability does not go away.** Every dollar spent on AI surveillance infrastructure creates a permanent capability. The China threat may justify the build-out today, but the systems will outlast any specific adversary. Ask yourself: who do you trust with that power in 10 years?