# AI capex — X 热门讨论 (2026-10-08 13:32 UTC)
## @kokko_coco (kokko | 古賀興司) · 10-08 07:25 · ♥86 ↻20 💬1 BlackRockがQ4の株式見通しを出している。その中で興味深いのが、ハイパースケーラーやAIクラウド企業のAIインフラ投資について、投資回収期間が数カ月前の最大4年程度から3年未満に短縮しているとの指摘。クラウド売上の加速も確認できており、AI設備投資は過剰投資ではなく、実際の収益に結びつき始めているとの判断。したがって、「AIバブルだから過剰投資でcapexが止まってアウト」というより、「ROIが確認されることで次のAI投資が正当化される」という循環を想定している模様だ。 https://x.com/kokko_coco/status/2108096304563753151
## @DKThomp (Derek Thompson) · 10-08 12:36 · ♥77 ↻13 💬5 New newsletter: THE 26 MOST IMPORTANT FACTS ABOUT AI AND THE ECONOMY
After weeks of feeling a bit lost in a sea of decontextualized AI factoids, I decided that what I really wanted to read was a structure, a story. So I wrote one.
Today’s newsletter is a fact-by-fact, chart-by-chart story about what AI is actually doing to the economy, with answers to questions like:
1. I keep hearing about trillions in "AI capex," but where exactly is all that money going? (Chart 1 has got you)
2. If AI is powering an otherwise moribund economy, how is it also starving other sectors of capital and resources? (Chart 2 has got you)
3. Do Americans find AI helpful, or do they think it's going to destroy society? (Definitely yes, and ... uh, also yes)
4. How many of the best-performing stocks in 2026 are tied to the AI buildout? (arguably, 19 out of 25)
https://t.co/ssQ4uFcQMM https://x.com/DKThomp/status/2108174694411026544
## @zerohedge (zerohedge) · 10-08 13:18 · ♥48 ↻9 💬10 The hyperscaler and AI pipeline for IG/HY debt is done: no more can fit. That's why everyone is scrambling to use SPVs/Project financing/chip-collateralized structures to fund capex before that last door slams shut too. > 引用 @zerohedge: Credit remains the big story: here everything is falling apart. 1. Oracle bond yields, CDS record high 2. SpaceX bonds new all time low/YTW high 3. Bond behind Meta's Hyperion data center (proj Beignet) hit record low 4. Paramount CDS trading at financial crisis levels https://t.co/bxEqe8IQFw https://x.com/zerohedge/status/2108185347637096922
## @IREN_Bull (IREN Bull) · 10-08 06:42 · ♥30 ↻1 💬3 I respect the detailed analysis, but I think you’re making a fundamental mistake: you’re focusing almost entirely on the cost of IREN’s growth while dramatically underestimating the economic value that growth creates.
And your central financing argument has already been challenged by IREN’s latest results.
1. The financing thesis is increasingly outdated.
Kent Draper just confirmed that IREN’s latest contracts with SUB-INVESTMENT-GRADE customers secured debt financing covering 90% of GPU capex, PLUS customer prepayments covering another 50%.
That’s 140% of GPU capex financed without issuing equity for the hardware.
Your premise is that attractive financing depends on Microsoft-quality customers. Yet IREN is demonstrating that even lower-rated customers can support substantial financing. The lending market is evolving rapidly, and IREN is benefiting directly.
2. You’re ignoring the economics behind the borrowing.
You emphasize Mackenzie’s 9% financing versus Microsoft’s roughly 6%. But IREN’s recent contracts generate over $20M in annual revenue per MW versus approximately $12M/MW for Microsoft.
On 100MW, that’s potentially $800M MORE in annual revenue. A three-point interest premium on $2B of debt costs just $60M annually.
Revenue isn’t profit, obviously. But dismissing the financing as expensive without accounting for substantially greater revenue density misses the entire point.
Management estimates approximately TWO-YEAR GPU paybacks. Compare that with contracts extending several years. That’s the potential for substantial cash generation after initial capital recovery.
3. Your dilution argument is equally incomplete.
IREN has raised approximately $19B in financing, with only around $3B coming from equity. And Kent Draper’s latest comments directly undermine your dilution thesis.
Draper confirmed that even IREN’s SUB-INVESTMENT-GRADE customers are now supporting debt financing covering 90% of GPU capex, PLUS customer prepayments covering another 50%.
That’s 140% of GPU capex funded without issuing equity for the hardware. Think about the significance of that. IREN is increasingly financing its expansion through lenders and customers rather than shareholders.
Meanwhile, NBIS is also raising billions through convertibles and other financing instruments. Why is dilution an existential threat at IREN but an acceptable cost of growth at NBIS?
The relevant metric isn’t the number of shares issued. It’s the economic value created PER SHARE.
And Draper’s comments suggest IREN’s financing model is becoming more efficient as it scales, not less.
4. You’re confusing aggressive expansion with poor underlying profitability.
Depreciation matters. GPU replacement matters. But accounting depreciation isn’t an immediate cash payment, and negative free cash flow during massive infrastructure expansion doesn’t establish that individual deployments are unprofitable.
The real questions are how quickly invested capital is recovered and how much cash the assets generate over their useful lives.
IREN’s improving revenue density, customer prepayments and GPU financing arrangements make that calculation considerably more attractive than your earnings projections suggest.
And applying a theoretical $250B cost to the entire 5GW pipeline ignores phased development, future contracts, refinancing, cash generation and IREN’s unencumbered infrastructure assets.
5. Your software argument is backward-looking.
SemiAnalysis’ ranking is a snapshot, not a permanent verdict. IREN acquired Mirantis specifically to expand its software and cloud capabilities.
More importantly, owning scarce, energized, AI-ready infrastructure is itself an enormous competitive advantage. Software can be developed or acquired. Securing gigawatts of power and constructing operational AI facilities is considerably harder to replicate.
(Cont.) https://x.com/IREN_Bull/status/2108085513089343575