Steve Jobs popularised the tech company product reveal presentation, and even 14 years after his passing, nobody can rival his stage presence.

The iPhone was bound to change the future of mobile phones, but his incredible presentation — that it’s an iPod, a phone, and an internet communicator, in one device — was what sold it. His second most iconic presentation was for the MacBook Air, a laptop that challenged conceptions of how thin and light a computer could be. To show it, Jobs pulled it out of a plain manilla envelope that had been innocently sitting on his podium the whole time.

I thought back to this as I watched Nvidia’s 2025 “Consumer Electronics Show” presentation, as Jensen Huang, the company’s billionaire founder and CEO, unveiled Project Digits.

Most of his presentation has focused on the power of AI, and what it can achieve, running on massive AI supercomputers, running in enormous server farms. But wouldn’t it be nice if a supercomputer was small enough to run on your desk?

And so, he pulled out Project Digits, a $3,000 AI supercomputer, releasing in May, about the size of a Mac Mini, that can run Nvidia’s full AI stack locally. And the “supercomputer” label isn’t marketing.

It’s powered by Nvidia’s GB10 Grace Blackwell Superchip, an unspecified Blackwell GPU — related to the consumer versions unveiled at the same event — and comes with 128GB of RAM and 4TB of NVMe storage; and all of this with a standard computer power plug, in a tiny puck-like chassis. Given its focus on AI and how little these specifications relate to ordinary computers, saying it’s “1000 times more powerful than a MacBook” doesn’t mean a lot, but that’s not far off the mark. According to Nvidia, you can run their entire AI stack locally on this device, along with large language models with up to 200 billion parameters. They can also be linked together to run up to 405-billion-parameter models.

There is a lot of excitement about this computer online — and insufficient concern about how hot or loud this will run. But, whereas the MacBook Air was for the every man, this mini powerhouse is for a specialized use case.

If you develop AI tools, this is a fantastic development kit, letting you run and test it entirely locally before deploying it to a server; and for researchers working with large language models and neural networks, having this immense computing power, locally, will allow faster iteration on datasets and complex calculations, without ever sending the work to the cloud. When this hits the market, expect many more interesting AI applications from small teams and solo developers.

But, if you are anyone else, you shouldn’t buy this computer. If you want top graphics rendering performance — for 3D modeling or gaming — you should buy Nvidia’s new $2,000 RTX 5090 graphics card, which will be far faster for both uses and still be more than capable of handling local AI applications. And you’re not meant to use this as your only computer. Tinkerers will install Windows on it and try using it daily, but it will default to Linux, and the intention is that you use this, like a server, in conjunction with a main computer running an operating system.

What’s exciting here though isn’t necessarily what this product is, but what it suggests for the future. The first iPhone was flawed, with a high price, poor network compatibility, and no app store. But it opened the door to the modern smartphone ecosystem we enjoy today.

The first MacBook Air was weak and wasn’t practical for many people, but it showed how thin laptops could be, which we’ve enjoyed ever since.

Project Digits is the first consumer computer of its kind, and it’s a limited device. But it previews a potential future where we all have AI pucks on our desk, running complex systems on-device to help manage our lives and work, and improve the performance of our software.