See capital expenditure.
Capital expenditure Spending on buildings and equipment, shortened to capex. Chapter 1.
From The AI Compute Race. Read the book
Where the record uses it
- Scorecard indicator Hyperscalers’ Q1 operating cash flow (OCF) / cash capex
- Scorecard indicator Microsoft Jun. quarter
- Scorecard indicator OpenAI + Anthropic annualized revenue
- Dated test Chapter 12, claim 5, the hyperscalers’ own borrowing, due 31 December 2027
Sources that use it
- BG2 Pod, “The SpaceX IPO, Fable 5, AI Capex Update & Market Check,” Brad Gerstner with Gavin Baker, Andrew Fox and Clark Tang, recorded 10 June 2026, published 11 June 2026 (Fox on the deflation narrative at 1:07:09–1:08:01, “Coming into this year … everyone expected token pricing, the price of compute, it’s all deflationary” and “this year what we’ve seen is the opposite”, an I think before each half on the tape and the ellipsis spanning “going back to this kind of what narratives were violated, you know, I think into this year”; Baker on $60 billion a gigawatt at 27:22–28:13, “it costs $60 billion to put a gigawatt on the ground today”, large-v3’s rendering, where medium.en reads 60 billion dollars, “we’ll call it 35 of that are the GPUs and the silicon”, “25 billion is the land, the shell, the power, and the cooling” and “I would hypothesize that those elements are probably going to be inflationary”; Fox on the monetization rates signed at 24:00; Tang on the $20–40 billion per gigawatt series at 1:08:37–1:09:20, “increasing from, call it, like 20 billion in the best of cases at the beginning of the year to now like 30 to even pushing 40”, its you know, a repeated 20 and the false start for before at the beginning dropped; Gerstner on 2027 capex against lab revenue at 1:04:00; the words settle Fox at the hand-off “What is it, Foxy?” and at Baker’s “Man, Foxy”, and a voice match against the tape’s settled turns on 23 September 2026, by Sourcing’s voice_match.py, WeSpeaker embeddings with the room’s shared voice removed, one seven-second reference excluded after the tool refused its first runs, not a human ear check, hears Fox through 1:07:13–1:08:05 with no change of speaker at “Right?”, so the deflation line and the “less than 0.2%” line are one voice, and Tang in six of seven pieces of the per-gigawatt series, the seventh settled Tang’s by the frames the next day; two decodings of the audio, medium.en and large-v3 on two slices, the frame sheets, the seat key, the mouth crops and the motion tables held in the set’s Quote audio checks folder of 24 September 2026 with their md5 in its manifest; a turn the Library note labels Gerstner’s and the show’s own seat key, the motion in one continuous three-shot and a voice match against nine settled turns give to Baker on 24 September 2026, the billion on both decoders where the caption drops it and the dollar sign the print’s; the speakers settled from the video frames at the name hand-offs on 23 September 2026 and Tang’s last span, 1:09:09–1:09:18, from the frames on 24 September, the caption track carrying no speaker labels). In the Sources
- David Cahn (partner, Sequoia Capital), interviewed by Alex Kantrowitz, Big Technology, “This Sequoia VC Has Tracked AI Capex Since Day One. Here’s the Math,” a clip published 9 August 2026 from the full episode published 5 August 2026, recording date not stated (“the only possible way to pay those dollars back is going to be AGI” at about 11:50–12:00, “those dollars” the 10 to 100 gigawatts of capex then being talked about, “far beyond where we are today, to be clear”; “people just don’t care about the math anymore at all. It’s all in, all the way, let’s see what happens” at about 10:50–11:00, the auto caption’s are between people and just on neither decoder; both read against two decodings of the audio, medium.en and large-v3 on 25 September 2026, four slices held in the set’s Quote audio checks folder of that day with their md5 in its manifest: gonna be AGI on both where the caption reads going to be, the print keeping the caption’s, the word unsettled; Cahn the speaker by the frames, the cold open 0:10–0:14 his). In the Sources
- Jensen Huang (NVIDIA) with Brad Gerstner and Clark Tang, “Does AI Revenue Justify Capex, 27 AI Bottlenecks,” BG2 Pod, YouTube, published 15 October 2026 (“our percentage of the world’s CapEx was in single digits” six years earlier, “with the world’s CSPs, we’re probably running about 40%” and “whether it’s on-prem or neoclouds, half of our business, our percentage of CapEx is probably, what, 80%?” ~54:38 to 55:03, no period named for either share; “it’s very likely that we’ll move to” a revenue share, “we’ll make an investment, we’ll get the benefit of the sale up front, but we’ll also have the benefit of” the share “going forward” ~1:28:58, the word for the share heard indistinctly at that moment; “how could you have rev share if there’s no offtake?” ~1:29:36, Huang’s own name for it; “NVIDIA’s architecture is the most productive. It generates the most tokens per watt,” and when the world’s power “reduces, it shifts their choices to the most productive version” because “they still want to make their revenues” ~52:24 to 52:45; audio-checked, and the question at ~1:29:36 placed with Huang’s settled turns against both hosts’ by a voice test on 16 October 2026) (the hosts’ firm, Altimeter, held NVIDIA as its largest position at 30 June 2026 by its Form 13F-HR, and its slides and the episode’s description carry its disclaimer that its funds are invested in the companies discussed) In the Sources
- Jim Covello (head of Global Equity Research, Goldman Sachs), interviewed by Sharmin Mossavar-Rahmani, “Understanding the AI Investment Landscape: Enterprise Adoption, Hyperscaler Capex, and Open-Source Models” (YouTube title: “How New Models Are Changing the AI Investment Landscape”), Goldman Sachs Wealth Management video, recorded 30 July 2026, its episode page dated 17 August 2026 and its YouTube upload 18 August 2026, Goldman’s own published video transcript (the two kinds of circular financing: “That’s just another form of discounting… That’s very different from if I don’t give this customer this money. They’re not going to be able to make those investments. That’s much more problematic. I don’t think that’s the sign of a very healthy supply chain”; on the buyer’s own routing layer, described by function and said to be missing: “I think you’re going to have a model optimization layer built into all enterprises that routes the high consequence queries to the big frontier models that are a little more expensive to use, and then it’ll route the low consequence queries to the open source model”; “That model optimization or model router really is yet to be developed in most organizations”; “right now, enterprises collectively are not making or saving money on their AI implementations”; read again from Goldman’s own transcript on 19 September 2026). In the Sources
- Jesper Nielsen, Rahul Khubchandani, James Tucker, Ketil Gjerstad and Gerry Hansell, Boston Consulting Group, “The AI Path Is Uncertain. Build a Portfolio of Bets,” 28 September 2026 (a consultancy’s advice to the companies it sells strategy and AI work to, its sum in the sidebar “The Revenue Hurdle Behind the AI Capex Boom”; “The five largest hyperscalers—Microsoft, Alphabet, Amazon, Meta, and Oracle—are on track to commit roughly $5 trillion of AI capital expenditure by 2031”; “To cover that depreciation and earn a 10% pretax return on a capital base of $3 trillion to $4 trillion, a company would need annual AI revenues of $1.6 trillion to $2.0 trillion, assuming cash operating costs equal to 20% to 30% of revenue”; “At a 15% hurdle rate, the revenue requirement rises to $2.1 trillion to $2.6 trillion”; “Today the combined AI revenue is roughly $100 billion, leaving a gap of 16 to 20 times. Labor substitution alone cannot close it.”; “At a take rate comparable to cloud or offshore IT services—20% to 35% of the cost displaced—$1.6 trillion to $2.0 trillion in AI revenue would imply replacing $5 trillion to $10 trillion in annual labor costs. That compares with a US white-collar wage bill of roughly $8 trillion to $10 trillion”; the wage bill given with no series or year; read whole on 20 October 2026). In the Sources
Related terms
From the glossary of The AI Compute Race.
How to cite this page
Dryburgh, Lee S. “Capex.” In The AI Compute Race. DRYBURGH. https://dryburgh.com/vol1/glossary/capex (accessed 2026-11-01).
BibTeX
@misc{dryburgh-capex-2026,
author = {Dryburgh, Lee S.},
title = {{Capex}},
howpublished = {In \emph{The AI Compute Race}, DRYBURGH},
year = {2026},
url = {https://dryburgh.com/vol1/glossary/capex},
note = {Accessed 2026-11-01}
}