# AI capex — X 热门讨论 (2026-09-14 04:02 UTC)

## @The_AI_Investor (The AI Investor) · 09-14 02:15 · ♥43 ↻5 💬5 The AI bear case has been pretty entertaining to watch.

Every time one argument dies, a new one appears. I guess they came from

“AI is useless.”

to

“AI is so useful and powerful that we may need to deliberately stop it.”

That is quite a journey for the bear case.

1/ “LLMs are toys. Nobody actually needs this.”

Then people started using them for real work.

2/ “Fine, but scaling is over. We’re running out of data. No need for all that compute.”

Scaling kept going.

3/ DeepSeek: “See? AI is getting too efficient. Compute demand is about to collapse.”

Instead, cheaper intelligence made people use more of it.

4/ “Okay, but AI is basically just for coding.”

Then agents started spreading into research, customer service, sales, finance, security, science, and pretty much every knowledge workflow.

5/ “There are only so many tokens one human can consume.”

Then reasoning models and agents arrived.

One human request can now trigger dozens or hundreds of model calls behind the scenes.

6/ “Fine, demand is real. But Big Tech can’t keep spending this much.”

Then AI revenue started showing up, cloud growth accelerated, and capex plans kept getting bigger.

7/ “Okay, but there isn’t enough power.”

So hyperscalers started signing PPAs, backing new generation and planning multi-GW AI campuses.

8/ “Sure, but who is going to finance all this?”

Banks, private credit, infrastructure funds and hyperscalers themselves: hello.

9/ “Okay okay… it’s circular financing.”

Not circular demand, it's massive.

And now the newest bear case:

10/ “AI is becoming too powerful. We need to slow it down.”

We somehow went from:

“AI is useless.”

to

“AI is so useful and powerful that we may need to deliberately stop it.”

That is quite a journey for the bear case. https://x.com/The_AI_Investor/status/2099321089059168364

## @onechancefreedm (EndGame Macro) · 09-13 23:26 · ♥31 ↻2 💬9 I would love to know whether any AI CEOs are already being warned behind closed doors by political forces about what could happen if the balance of power changes in the midterms. Control of Congress changes who chairs the committees overseeing technology, who controls the investigative agenda, who can authorize subpoenas and hearings, and who writes the legislation that could eventually impose mandatory audits, liability rules, compute restrictions, disclosure requirements or even limits on certain frontier models.

That alone could put enormous pressure on executives. A hostile congressional majority could demand internal safety reports, board communications, model release discussions, partnership agreements and records dealing with children, employment displacement, market concentration and alleged safety failures. Congress would not suddenly control the FTC or DOJ just because the midterms changed hands, but it could intensify oversight, control appropriations, refer findings, build the public record for future legislation and make executives spend months defending themselves under oath.

Now imagine those CEOs are already being told privately that the political environment is likely to change. The message would not even need to sound like an explicit threat. It could simply be made clear that if they resist the emerging regulatory framework today, they may face subpoenas, investigations and legislation tomorrow, while companies that cooperate now may have far more influence over how the eventual rules are written.

That is what could make the current push to slow AI much more politically significant. If CEOs already expect valuations to reset and capex to slow, and they also believe future political power could become far more hostile, cooperating with a slowdown may suddenly look less like surrender and more like self preservation. The question I would love answered is whether anyone is already making that calculation clear to them behind closed doors. > 引用 @MarioNawfal: 🇺🇸 Obama is urging Democrats to STOP treating A.I. as a side issue, and start putting it front-and-center...

At a closed House Democratic fundraiser, the fmr. POTUS warned House Minority Leader Jeffries that the technology is moving FAST in private hands, and that somebody needs to govern it.

If Democrats take the House in November, he wants a public framework ready to go...

He wants A.I.-related matters to be at the TOP of the list of 2028 priorities, re: safety, kids, and what happens when people are made redundant.

Is that a winning strategy, you think?

Source: NYT / Writer: Samuel https://x.com/onechancefreedm/status/2099278451974795629

## @SureshKBN (Suresh K) · 09-14 03:29 · ♥38 ↻0 💬4 probability of AI Capex slowdown is as easy as U.S. could surrender its AI leadership to china and other https://x.com/SureshKBN/status/2099339607653659002