AI
The Acquisition Was the Sunset Notice
I thought I bought a device. What I really bought was a lease on a service, and the expiration date got set the day somebody bigger bought the company.
“Little to no human action other than remembering to turn it on and charge it.” That’s my whole spec for a memory device, and for a while I had one that met it.
I wear an AI pendant. It records my day, sends it to the cloud, transcribes it, and makes it searchable. When it works, it’s the closest thing I’ve found to a memory you don’t have to operate. I just have to remember to charge it and turn it on. I haven’t been to the app in months. I like it a lot.
It’s also going to stop working. Meta bought the company that makes it in December 2025. They stopped selling the pendant that same day and promised to keep supporting existing owners “throughout 2026,” and nobody has said anything about what happens after that. Owners in the EU, the UK, and a handful of other countries lost the service the day the deal was announced, and got two weeks to download their data before it was deleted. The hardware on my shirt will still be perfectly fine when the rest of us lose it. It’ll charge, and the light will come on, but on its own it won’t do much of anything, because the part that actually did the work lived on someone else’s servers.
When it sank in that the service was going away, my first reaction was about the replacement I’d tried, which I hated. It wanted me to tell it to process every recording. It should be automatic.
I wasn’t surprised by the shutdown itself, and that’s sort of the point. This has happened before. In February 2025 HP bought most of Humane, the company behind the AI Pin, for $116 million. HP took the software, the patents, and the engineers, and left the Pin itself behind. Ten days after the announcement, the Pins stopped connecting to Humane’s servers, and almost everything people bought them for stopped working with them. The pattern is consistent enough now that I think we should just call it what it is. In this category, the acquisition is the sunset notice. It isn’t printed on the box, but the countdown starts the day the deal closes.
Think about why. A big company buying a small hardware startup is almost never buying the product. It’s buying the team, the patents, maybe the know-how. The product is a cost center with customers attached. Keeping the servers running for a relatively small group of early buyers is pure expense. So there’s a grace period, sometimes a year and sometimes ten days, and then the lights go off.
What you bought is a lease on a service, and the device is the shell around it. Any hardware that needs a backend to function is basically a subscription, whether or not there’s a monthly fee. If there isn’t one, that’s worse, because nobody’s paying to keep it alive. That’s where mine ended up. The new owner made the paid plan free for existing customers, which sounds generous until you notice it also means nobody’s paying for the servers anymore.
Price it that way before you buy it. Take the device price and ask how many months of service you realistically expect to get. Not how long the hardware lasts, but how long the company lasts as an independent company that still cares about this product. For a startup in a hot category, that’s a short number. Divide one by the other and that’s your real monthly cost.
Then ask the questions that actually matter. Can I get my data out, in a format I can use, without asking permission? Does anything work offline, or locally, if the cloud goes away? Is there anyone else who supports this hardware if the maker disappears? In my case, I pulled the whole recording archive down to my own machine right away, so my data’s safe. Whether the device survives depends on somebody else’s app picking it up, so I’m testing an open-source alternative alongside the old one before I buy anything new. That’s the trade I should have priced on day one.
I’ll admit my spec has a hole in it, too. The gaps in my own recordings this year all came from me forgetting to charge it or wear it. The habit was failing about as often as the tech. There’s still a person in the loop, and the person is me.
None of this means don’t buy AI hardware. Some of it is genuinely great, and early adopters are how this stuff gets better. It just means you should stop thinking of it like a toaster. A toaster doesn’t need anyone’s permission to keep working.
If the gadget only works because a company keeps paying its server bill, you’re really betting on the cap table.