AI is an arms race, and the US wants $9 billion in Nvidia superchips to keep up
The US has backed a secret $9 billion request for AI superchips to help its intelligence agencies keep pace with leading AI labs.
Intelligence analysis by GPT-5.4 Mini
The article says the government wants a large pile of Nvidia-based AI hardware because modern models need far more compute, memory, cooling, and power than older systems. Congress still has to approve the funding.
The story is about the government buying very powerful computer parts so its spy agencies can keep up with the fast-moving world of AI. These parts are like the engine in a race car: without them, the car cannot go very fast.
The article says the special chips need a lot of memory, storage, power, and cooling. That is why this is not just one computer, but a huge and expensive setup.
It is a bit like trying to build a giant factory just to stay in a race that keeps speeding up. The government wants to keep pace, but Congress still has to agree to pay for it.
Analysis
What the request is about
The article says the US government has given the green light to a secret $9 billion request for AI superchips, but Congress still has to approve the money. The goal is to help agencies such as the CIA and NSA keep pace with the biggest AI companies, including Anthropic and OpenAI.
Why the hardware matters
ZDNET describes Nvidia's Grace Blackwell chips as the core of this effort. The GB10 combines a 20-core Arm CPU from MediaTek with an Nvidia Blackwell GPU, plus 128GB of LPDDR5x memory and 4TB of NVMe storage. In the article's framing, that setup can deliver about 1 petaflop of FP4 AI performance while drawing only 140 watts. The point is that advanced AI now needs specialized hardware, not just more software optimization.
Scaling gets expensive fast
The article explains that the real cost comes when these chips are deployed at data-center scale. It cites Nvidia's GB300 NVL72 rack, which can pack up to 72 GPUs and 36 CPUs into a liquid-cooled unit. At that scale, the spending jumps sharply, with racks priced in the millions and full data centers requiring huge amounts of power and cooling. That is why the article frames the issue as both a technology race and an infrastructure race.
The bigger picture
The government is not only trying to use AI itself, but also to inspect models that the public uses. The article links that push to earlier efforts to create a testing process before public release, though that plan was dropped after industry pressure. The broader message is that AI has moved fast enough to outpace policy, procurement, and the government's own hardware base.
Key points
- The US has approved a secret $9 billion request for AI superchips, but Congress still needs to sign off.
- The chips are meant to help the CIA and NSA keep up with leading AI companies and models.
- The article says Nvidia's Grace Blackwell systems are central because modern AI needs massive compute, memory, cooling, and power.
- The story frames AI hardware spending as both a national-security issue and an infrastructure race.
- The piece also notes that past government efforts to set up pre-release AI testing ran into industry pushback.
