'ZDNET Recommends': What exactly does it mean?
ZDNET's recommendations are based on many hours of testing, research, and comparison shopping. We gather data from the best available sources, including vendor and retailer listings as well as other relevant and independent reviews sites. And we pore over customer reviews to find out what matters to real people who already own and use the products and services we’re assessing.
When you click through from our site to a retailer and buy a product or service, we may earn affiliate commissions. This helps support our work, but does not affect what we cover or how, and it does not affect the price you pay. Neither ZDNET nor the author are compensated for these independent reviews. Indeed, we follow strict guidelines that ensure our editorial content is never influenced by advertisers.
ZDNET's editorial team writes on behalf of you, our reader. Our goal is to deliver the most accurate information and the most knowledgeable advice possible in order to help you make smarter buying decisions on tech gear and a wide array of products and services. Our editors thoroughly review and fact-check every article to ensure that our content meets the highest standards. If we have made an error or published misleading information, we will correct or clarify the article. If you see inaccuracies in our content, please report the mistake via this form.
[Close] Follow ZDNET: Add us as a preferred source on Google.
ZDNET's key takeaways
- Nvidia's GB10 chips power modern AI models.
- The US wants $9 billion for AI superchips.
- Congress still needs to approve the funding.
The sudden -- even by tech standards -- pivot to AI hasn't just left businesses scrambling to play catch-up; even America's spy agencies are struggling to keep up.
**Also: **AI isn't getting smarter, it's getting more power hungry - and expensive
This is why the government has given the thumbs-up to a secret $9 billion request for superchips that will allow the CIA and NSA to keep up with what big AI players like Anthropic and OpenAI are doing.
So what are these superchips? #
The latest batch of AI models needs a massive amount of computing power -- not to mention a huge supply of power and specialized cooling that comes as part of cutting-edge, modern-day data centers -- to run, and the silicon to deliver this is Nvidia's Grace Blackwell superchips, named after American mathematician David Blackwell and American computer scientist and Navy pioneer Grace Hopper.
Also: How to learn Claude Code for free with Anthropic's AI courses - one took me just 20 minutes
These superchips, called the GB10, feature a 20-core Arm CPU made by MediaTek, codenamed Grace, with an Nvidia GPU based on the Blackwell architecture. Take this chip and add 128GB LPDDR5x -- it's the demand for memory for AI that's skyrocketed the price of RAM and things like Raspberry Pi boards -- and 4TB of storage in the form of an NVMe M.2 SSD, and you have a chip that offers 1 petaflop of FP4 AI performance for a power draw of only 140 watts. That's just the chip.
This isn't a lot when you consider that most modern gaming PCs can swallow up to 1,000 watts of power.
This one chip has the power to fine-tune AI models with 70 billion parameters. Just in terms of storage alone, a model like this needs some 140GB of space.
Want a GB10 system? Best Buy sells a rack version starting at about $5,000!
**Also: **What Google's TurboQuant can and can't do for AI's spiraling cost
But the real power usage comes when you scale these up. The GB300 NVL72 is a rack of as many as 72 GPUs and 36 CPUs in a single, liquid-cooled unit. Now scale this up to data center proportions, and you can start to see why the power demand goes through the roof.
A rack can cost anywhere between $1.8 million and $4 million. And a data center can have as many as 100,000 racks.
But if you want to run big AI models such as Anthropic's Claude, OpenAI's GPT 5.5, or DeepSeek's V4, this is what you need.
Why does the government need this much power? #
AI is seen as both a next-generation tool and a national security threat, something that is again moving faster than governments can legislate for or put guardrails in place. Just the other day, a planned executive order that would have outlined a process where AI companies would "volunteer" their models for government testing for a period of up to 90 days ahead of public release was scrapped after pressure from industry leaders.
This order makes it clear that not only does the government want to leverage AI itself, but it also wants to be able to examine models used by the public.
**Also: **How Qualcomm's new wearables chipset could spell the end of smartphone dominance
This would need serious hardware horsepower.
There's also an element of playing catch-up from a lack of investment in computing hardware over the past years. Combine that with the current shortage of chips and other AI hardware, and all that means having to spend billions just to stay in the game.
The $9 billion, which still needs to be approved by Congress, would allow the government to acquire both the infrastructure and hardware it needs to stay relevant in the AI game.
But buying chips and expanding data centers takes time, so in the interim, some $800 million of the defense budget has reportedly been repurposed to acquire more cloud compute power. The intelligence services are also continuing to make use of an advanced AI model developed by Anthropic called Mythos, despite the company being labeled as a supply chain threat.
$9 billion is just the tip of the iceberg #
And that $9 billion, while it sounds like a lot, really isn't in the grand scheme of AI. Amazon Web Services is investing $50 billion to upgrade its government cloud computing services, a platform that the intelligence agencies use extensively.
Also: I quit ChatGPT for a free, private, and local AI called Ollama - here's why
And the successor to Grace Blackwell silicon is in the pipeline -- the Vera Rubin platform, named after an American astronomer. These chips combine a brand-new, custom-built Arm-based CPU called Vera and a high-performance GPU called Rubin, and are designed to offer up to 10 times more performance per watt compared to Grace Blackwell and make use of high-performance HBM4 memory.
AI is now very much a modern-day arms race, and governments wanting to even keep up are going to have to invest moonshot money.