Because there's no return on the copper you are digging up tens of thousands of miles of cable for?

Do you have any information showing what the ROI would be on recovering all this unused cable, negative or positive? Neither does anyone else. So stop assuming.

Because we don't already have that?

Apple is a corporation with literally hundreds of billions of dollars in their warchest. That makes them independent enough to tell shareholders that we're not bending to illegal bullshit from a president who refuses to stay in his legal lane.

Nobody is buying your excuses, or reading anything past that first sentence where you seem to not understand that Apple has far more resources to bring to bear than Harvard, but none of the actual courage or fortitude to stand up for what's right.

I know I would love to get a Mac Studio for doing LLM work. I'm shoehorning a model into my M4 MacBook Pro and getting tons done, but I have to reboot it every day in order to clear memory fragmentation because I'm running right at the edge - 92% of GPU memory in use. After a while, things start to creep into swap.

It would be awesome to have 128GB or more available unified RAM so I can use a bigger model, but the 35B mix-of-experts I'm running now is quite capable, it's just slow because my employer didn't understand the value of M4 Max versus M4 Pro.

I would appreciate a re-ify of Xserve with M-series silicon. Give me the redundant power supplies and BMC control back, and modern 10GbE (or qSFP+ for even better). Everything else can essentially be a Mac Studio in a 1U or 2U box with a fuckton of RAM.

Seems like they are hitting the sweet spot for local LLM use in a way that their competition is stumbling around.

Despite all the jingoist bullshit, you are right now using China-produced hardware to post this nonsense.

If you don't trust it, you control the network, don't you?

AI can't do shit about any of the following:
1. you restricting what tools it has available to use
2. you restricting what commands it can run
3. you airgapping the box it's running on.
4. you egress filtering at your router
5. you setting up DNS on the box it's running on

STOP WITH THE FUD ALREADY.

How is selling hardware capable of running really fucking big models locally turning into "Microsoft paying Apple to run unprofitable AI workloads"?

Try knowing what the fuck you are talking about.

Executive orders only address the Executive Branch.

I'm not in the Executive Branch, so the president's EO doesn't have the authority to direct me to the men's room.

Google should have told them to shove it up their ass, as they are a private enterprise disconnected from the government, and the president's EO's are unenforceable bullshit.

Google did not do this, because they are too busy capitulating.

I'm in favor of this approach. These people are incapable of shame, but they are not incapable of being taunted and mocked mercilessly until the few people left in the Republican party that ARE capable of feeling shame have had enough or leave it to go the way of the Whigs.

We're WAY past disagreement over policy. This party needs either a severe reckoning, or a freshly dug grave.

At some point we need to compile a database.

I submit the following:
Temu Putin
Shitler
Orange Caligula
Dementia Donnie
President Von Shitzenpants

And how's that worked out for him?

I guess we find out in ~60 days.

If he's in the Mac ecosystem and talking about a Mac Studio he doesn't even need that. Available with 128GB of unified RAM and enough GPU units to use it effectively, and this thing can run 70B models on your desk. Doesn't even need to be a current one, a used M3 studio can do this.

See: Trump Tariffs.

self-inflicted wounds are easy for stupid fucking people with way too much power.

If the Chinese government says "no more shipping things to Apple" then the entire country is a single source.

This isn't hard to understand. And it's absolutely in China's power to do so.