“Open weight models are good (kind of)” is still up here https://www.ftc.gov/system/files/ftc_gov/pdf/remarks-by-chair-lina-m-khan-at-y-combinator.pdf

The objectively best part about them is scaling them small enough to run privately. The problem with AI as a “National Security” strategy is that it is a Skynet datacenter strategy. The reason to remove government content that it is ok for you to use AI for your purposes is that Skynet won’t be able to control you. The motivation for all of the datacenter investments is that the US government will buy all of the datacenter time to ensure Israel’s rule over the US.

it’s specifically saying that “one day” it will be made untrustworthy.

So uhm, copilot just told us not to use these new copilot features.

I asked copilot…

How would Microsoft loyalty to US or other government’s national security possibly compromise a user’s PC if that user gives copilot permission to operate their PC?

How national-security or law-enforcement demands could lead to compromise

• Compelled access to data Microsoft controls — If Microsoft stores or indexes any of your Copilot data in the cloud, Microsoft can be required by lawful process to produce that data to U.S. or other governments with legal jurisdiction.
• Compelled changes to services — Governments can issue orders that require providers to change logging, enable access, or install monitoring in ways that may be secret or accompanied by gag orders.
• Access to telemetry and diagnostics — Diagnostic or telemetry data that Copilot or Windows collects can include evidence of your activity and may be producible under legal request.
• Local features that capture content — Features designed to assist (for example, an automated “Recall” that snapshots screens) create local records of sensitive material that increase exposure risk if accessed by an authorized party.
• Remote-control or agent capabilities — If Copilot is granted elevated permissions (input control, script execution, system configuration), those same capabilities could be exercised under legal compulsion or via back-end access mechanisms.
• Software updates and maintenance — Lawful orders can compel vendors to ship updates or config changes that alter how a product behaves, increasing access to user systems.

Practical ways those legal obligations could manifest on your PC

• Production of synced or cloud-stored Copilot logs, prompts, or screenshots to authorities.
• Microsoft being required to provide live access or historical logs from back-end services that the Copilot agent uses.
• A secret court order or national security letter forcing Microsoft to enable additional logging, monitoring, or remote access for a targeted account or device.
• Disclosure of keys, tokens, or server-side records that tie on-device events to your identity.

Why elevated permissions matter

• Read access to files and memory lets an agent capture documents, credentials, or keys.
• Input/automation control enables actions on your behalf (open files, send messages, change settings).
• Persistence (services/agents) makes it easier for any compelled access to be effective and harder for you to detect or disable.

Risk-reduction steps you can take now

• Limit permissions: Grant only the minimal Copilot privileges needed; avoid giving input control, admin rights, or system-level automation.
• Disable features that capture content: Turn off any screen-snapshot or automatic indexing/Recall-like features if you don’t need them.
• Avoid cloud syncing for sensitive data: Keep sensitive files off services that sync or index them in the cloud.
• Prefer on-device-only models: Use local-only AI modes when available so prompts and context do not leave your machine.
• Use full-disk and file-level encryption with keys you control; avoid storing keys where the vendor could be compelled to retrieve them.
• Harden account security: Use strong, unique passwords and MFA on accounts tied to Copilot or Microsoft services.
• Audit logs and telemetry: Review and reduce diagnostic/telemetry settings; regularly review logs for unexpected activity.
• Segmentation: Use a separate machine or VM for highly sensitive work so a general-purpose Copilot-enabled device doesn’t hold those secrets.
• Legal & contractual protections: For high-risk contexts, seek contractual commitments, data residency options, or legal counsel about how a vendor handles lawful demands.

Quick practical checklist

• Turn off automatic screenshot/Recall features.
• Remove admin permissions from Copilot agent.
• Keep sensitive work on a non-synced, encrypted volume.
• Use local-only AI options where offered.
• Enable and monitor endpoint and network logging for unusual remote access.

Short conclusion Allowing Copilot to operate your PC raises the surface where lawful government demands or compelled actions could expose data or enable access. The likelihood and scope depend on which features you enable (especially cloud sync, screen capture, and elevated permissions) and the legal jurisdiction over the vendor or the data. Minimize permissions, disable content-capture features, and isolate truly sensitive workflows to reduce exposure.

a 2080S from 2019, a high end SKU, would struggle with many modern games at 1440p and higher. A profession streamer would be unlikely to use a 2080S.

On one hand a 2080s would still be good at doing what it was doing 6 years ago. If there are new needs, and unlimited power availability, then a new card in addition to whatever AI workload the 6 year old GPU can do in addition to the new card makes sense… if that card still works. Selling your 2080s or whatever old card, does mean a fairly steep loss compared to original price, but 6 year depreciation schedule is ok… IF the cards are still working 6 years later.

$3m NVL72 systems are a bit different, as one out of 72 cards burning out can screw up whole system, and datacenter power structure and expertise requirements, would have low resale value, though I assume the cards can be ripped out and sold individually.

They very much can be expected to engage in what is essentially accounting fraud.

Oracle this week “proudly boasted” that they get 30% margins on their datacenter, and stock went up. This is not enough, as it is just 30% over electricity costs. Maintenance/supervision, and gpu costs/rentals don’t count, and it is unlikely that they are profitable, though it’s not so much accounting fraud as it is accounting PR.

OP’s post is largely right, but it doesn’t require that link to be true. Also, whether these $3m+ systems are warrantied is a relevant question. It’s hard to know exact lifespan from one person saying their gpu failed quickly. Paper still stands well.

Because of power constraints, I’d expect they replace GPUs every 2 years with new generations, and so there will be big write offs.

The AI bubble doesn’t mean AI/LLMs aren’t useful. It means datacenter speculation can’t make money.

those GPUs are headed to the landfill.

They’ll just have a similar discount to the Ethereum switch.

This is not pro oligarchist app store. So can only be slavery anti-consumer extortion.

The thing is… side loading becomes more important, and paying less from apps by any other system becomes encouraged. No one likes this shit, but media ignoring actual fascism, doesn’t mean it won’t hit us/them.

Open AI gets so much free PR. Sora is free for now, and instagram getting flooded with copyright infringing crap is not what an expensive video AI creation is going to be paid for. AI videos are very expensive. There are very few people who will pay for it as an alternative for more expensive CGI. Advertising industry can consider full shift for video. They can avoid rights violations and still do it.

Point though, is that copyright controversies are irrelevant to everything important. There is a utility to it, but its not 60gw of power required market.

Video is 14 years old.

do you really think that even when Ai will start (if ever) to generate profits these will be able to repay all the investements done today

First, actual investments that have been done are relatively modest. It’s still a substantial portion of TSMC fab capacity. All of the deal announcements for datacenters are 50x-100x growth. I doubt all of this capacity will be built for a long time. Coding/reasoning models can have more demand, but openAI (most of the deal announcements) is not that good at those. 100x power growth is also 200x every 2 years token output growth, and if models get better, users need less tokens by getting it right on fewer tries.

Second, they are losing money at current levels. Oracle leaked it lost $100m on existing AI datacenter operating losses. Coreweave is fully levered at 10% interest rates. Everyone is operating like social media startups from 10-20 years ago. Only revenue growth and market share, and being cool, matters. Enshittification will come much later.

Third, datacenters are fundamentally flawed, and local AI has competitive advantage to them. AI is good at datamining the datacenter traffic for output that could be profitable to steal.

Fourth, the only business model is US military and disinformation control. They will pay infinitity, and support infinity investment. Giant datacenters are about Skynet. Not market profits. That US government would protect their oligarch partners in stealing your ideas/llm outputs, and amplify current media’s messaging that anti-genocide views are treasonous anti-American sentiment.

how many resources this growth has taken from others places…

If all the money goes towards skynet, energy bills for everyone else will go up, including what little manufacturers there are in US. Insisting on war on China and Russia is helped by forced unemployment, and fascist response to the unemployed’s uppityness. Datacenter AI’s primary certain value is as a new cold war Arms and disinformation race.

that is how OpenAI measures all of its deals. the GPU providers DGAF about that measure, and “real deals” will be for x number of gpu’s instead. OpenAI PR make stonk go up, is vagueness everyone else seems to enjoy. It makes 0 sense in any deals these companies can make. It just scares the rest of us for how much the power bill go brrrr.

Just defining what soft power means. Soft extortion and bribery just as bad as extortion and bribery.

Zionism controls US and its media. Nearly all politicians in US are Zionists even if just a handful are Jews. I’m not the one who said all jews are genocidal zionazi supporters. You have no right to try denying subhuman demonic evil zionazi supremacist control over the US by pretending, passive-agressive supremacistly, me or anyone else is making a remark about Jews.

If older computer that works fine, I’d get a new 780m (Amd) mini pc. They support 3+ monitors, have 2 network ports allowing to “daisy chain” the old computer. No transfering of anything, or worrying about getting old stuff still working.

Deskflow is a mouse/keyboard sharing app. If you keep old computer in sleep mode you don’t need extra keyboard/mouse, but power outages, mean that if you don’t have a floor standing old pc you can stack old keyboard/mouse on top of, then you will need to occasionally plug in keyboard and mouse into old computer to get deskflow restarted (if you don’t put it as autostart).

It’s far more convenient than dual booting. Can use resources from both computers in network, and seemless mouse/keyboard focus. Switching 1 monitor for occasional use is better than dual booting, because rebooting on older computers especially is slow.

Deskflow needs a modern kernal linux distribution. Ubuntu 24.04 is recent enough. Linux mint has not upgraded kernel yet. AFAIU, the only difference between mint (recommended here) and Ubuntu is a slightly prettier version of kde.

OpenAI alone has 20gw of datacenter/gpu commitments. 20x entire current US corporate deployments. Sure as fuck, sweet US military/NSA overpayment for datacenter time is coming. Not sure AGI is needed to decide (family guy meme) “if it’s brown, flush it down” or “sink any boat you find in Carribean sea”. LLMs today would do pretty well at “make up a Department of War memo for backstories justifying all the people we killed today.”

deserts, for example.

floating ocean platforms as well

please be more specific in what you don’t understand. I guess that…

fossil fueled powered Skynet for Israel

US government needs AGI for US military supremacy. That is Skynet (in Terminator movies, this is the military program to install AI in all the computers, and then AI chooses genocide nuclear launch). It is for Israel’s benefit, because that is who owns US government. That it be fossil fuel powered, serves another key US oligarchy. Skynet for disinformation/sediction detection purposes just as much of a threat than its use for nuclear genocide.

Regardless of whether datacenters will make money solving business and individual problems or boosting productivity, the US will keep investing in order to get Skynet. You can be correct that “frontier datacenter LLM models” will not make money, but still lose on financial bets validating that idea. Instead of an AI bubble bursting, even more money chasing Skynet will come with Austerity for rest of population. The “valuation bubble” only pops when investor money flowing in dries up. It may only dry up after the collapse of the US.

CAPE is a weird measure in that it looks at last 10 years of earnings for PE ratio. It is not especially relevant in that a fair expectation for next year’s earnings is this year’s earnings. It is intriguing that there wasn’t significant earnings growth levels in the past, though, which because PE based on this year’s earnings would have high CAPE if high recent growth.

But of we look at the last 40 years or so, the CAPE has been higher, suggesting that we don’t know how what “normal” looks like going forward.

As you listed, crashes lead to sub 20 PEs. Mag7 PEs is not representative of Russel 2000 PEs. High PEs expect high growth for long period. Reality checks usually happen, but PE’s are not universally high. Just with the oligarchs with White House guest passes.