So CES wrapped up in Vegas this week and my inbox is still recovering. Every year I tell myself I'm not going to read the liveblogs and every year I end up with four browser tabs open at 11pm reading about rollable screens nobody will ever buy. This year the thing that actually stuck with me wasn't a TV or a robot vacuum, it was Jensen Huang standing on stage in that leather jacket announcing something called Project DIGITS.
Quick version if you missed it: Nvidia is selling a little box, roughly the size of a Mac mini, that they're calling a "personal AI supercomputer." It's got their new GB10 Grace Blackwell superchip in it, starts at $3,000, and the pitch is you can run genuinely large AI models on your desk instead of renting cloud GPU time by the hour. Huang held the thing up in his hand during the keynote like it was a trophy. Supposedly it ships in May.
I'll be honest, my first reaction was not "wow, democratizing AI compute," it was "okay but can I actually buy one." Because here's the thing about Nvidia hardware announcements at this point: the announcement and the ability to purchase the item are two completely different timelines that happen to share a company name. Remember the RTX 4090 launch? I stood in a virtual queue for a Best Buy restock in 2022 for forty five minutes and watched it sell out before I got to the cart page. Forty five minutes! For a graphics card! I ended up paying scalper prices on a marketplace listing I'm still a little ashamed of, and I'm not going to say how much over MSRP it was, but it was a lot.
Which brings me to the other CES headline from this same keynote: the RTX 5090 at $1,999 and the RTX 5080 at $999, both announced this week with a launch date later this month. Two thousand dollars for a graphics card used to be an insane sentence and now it's just Tuesday. I don't doubt the performance numbers Nvidia is throwing around, the new DLSS stuff with multi frame generation looked legitimately wild in the demo clips, but I've stopped getting excited about launch day pricing because launch day pricing is fictional. It's the MSRP that exists for exactly the first ninety seconds before every card gets bought by a bot and relisted for 40% more.
I don't say this to be a downer about it. The tech itself is genuinely interesting, and Project DIGITS in particular is a real shift if it ships the way they're describing it and doesn't quietly become vaporware like some CES darlings do (looking at you, every "rollable laptop" concept from the last five years that never made it to a shelf). Having actual local inference hardware that isn't a $40,000 datacenter card is a real unlock for anyone who wants to run models without a subscription meter running in the background. I just want to see it exist in someone's actual house before I get invested.
Separately, and I'll keep this short because it deserves its own post honestly, the Supreme Court heard oral arguments yesterday on the TikTok divest-or-ban law, with the deadline sitting right around the corner on the 19th. I watched about twenty minutes of the coverage on it during lunch and the justices sounded pretty skeptical of TikTok's First Amendment argument, for what that's worth from a non-lawyer reading tea leaves. My personal, wildly unqualified read is that this doesn't end with TikTok surviving in its current form, but I've been wrong about bigger things than this so take it with a grain of salt. Feels strange that something so many people use every single day might just switch off, or get handed to a new owner, based on a ruling that could land any day now.
Anyway. CES always leaves me a little exhausted and a little hopeful at the same time, which I think is the correct way to feel about an industry that promises you a supercomputer on your desk in the same week as folding phone concepts that still creak when you open them. I'll believe the $3,000 AI box when one shows up on my actual desk in May. Until then it's just a very nice press photo.