The upper end for AI-pilled companies is paying upwards of $7,500 / seat a month. Which would get you some engineer or managers in most parts of the country, or multiple lower level employees. I have no idea what Kroger pays for their AI, but if they think that AI eliminates thousands of jobs then they are using frontier AI at the aforementioned rate. Or they have made a terrible blunder and are going to be short handed.
Most likely they are not using AI for much of anything but don't want to publicly take responsibility for cutting jobs to trim their expenses. Pure accounting trick with a little AI-flavored PR.

I didn't mean to overstate. GlobalFoundries aren't the primary for Apple or Qualcomm. They make some chips with them. I'm not certain if any of those projects are AI related, but for Qualcomm almost certainly yes.

Apple, AMD and Qualcomm are on GlobalFoundries.
I suspect SpaceX/xAI will have an inference chip soon, if they haven't announced it already. I think they are going through TSMC. But I'm not aware if that's firm.

I have had some luck using Claude to build Typst templates (typesetting system). It is especially useful if I hate the way a document looks and I want to rip it apart and redo it. I usually just put up with it, but the AI will mindless work on a few variations that I can preview.

At my job, AI is used by my boss to second guess the work engineering is doing. He burns a lot of our time having us check things and explain things. Almost entirely false positives, but he sure is smug when AI does catch something.

It's better as a solo game than multiplayer. The main draw is exploring and building bases. There are lots of other little mechanics but it's not a deep game. The story is interestingband unique, but confusing and short. You're fed story about two or three sentences at a time over several hours. They whole plot likely doesn't fill a page of text.

Multiplayer co-op was a big disappointment for my wife and I. We can hang out together, but it is hard to sync up main story or side missions together.

The space combat is bland. The first person shooter mini game in derelic spaceships is extremely boring.

Keep the Maia 200 in service, because they have 100's of GB of RAM per node. And are still very effective for inference.

Fabless IP is the norm in the industry. Nvidia and Apple are no different in that respect. And really, these are more like chips that compete with what Amazon, Meta, and Google are doing. Again, most of them also go to TSMC or Samsung for their foundry.

If you want to ditch Nvidia, you should probably use AI to make a new CUDA competitor instead of wasting time designing silicon.

In what world would the staff stocking shelves on the night shift have access to the HVAC controls of a supermarket?

That would mean having your equipment managed and monitored locally instead of by the vendor through a service contract.
Putting everything online is CHEAP.

A lot of commercial "security" companies just sit in a central office with staff that monitors hundreds of different customer locations. All those cameras are online somewhere.

I can get it to iterate and improve. So I believe it is possible with a better model and better prompts.

For writing "business logic", AI is already capable enough. AI is cranking out satisfactory code for the garbage requirements that coders have to deal with in the corporate sphere.
For elegant art that coding as a hobby demands, the AI doesn't satisfy. It isn't satisfying to use, and the results are not satisfying to read. That said, I made a bunch of shitty little games with AI.

What actually impressed me is that Claude is clearly a coding model, but I squeeze stuff out of it that isn't code at all.
Like a minimal table-top RPG rules set inspired by original AD&D (1st edition) and mashed up with the aesthetics of 24XX by Jason Tocci.
I even used Claude to run some solo playtests, forced to read the rules and play both sides of a short adventure. Then it gives me feedback about what it found confusing, what it had to make up, what worked, etc. Certainly it gets things wrong a LOT. Often trying to slip into 5e rules instead of the rules in the new system or the AD&D 1e rules that are my baseline.

I wouldn't bother testing the submitter. We need to be more willing to reject submissions without feeling the need to explain ourselves in depth.

If they're using AI, then it should be trivial for them to rewrite it. So nothing of value is lost by dropping bloated PRs.

I usually have to convince Claude to stop writing 5+ line comments in the code. Too much exposition, too much boiler plate, and it often writes a complex generic solution instead of a concise specific solution.