How DeepStack settles a question · Part 04
Does AI capex earn its return?
Is the capex spending defense, or is the return not there?
From Part 04: AI Spending Has Met Its First Real Bill. 9 voices on 2 sides, attributed and dated; the DeepStack Call with its history; 5 dated tests that settle it; the results as they come in.
The mechanism
Hyperscaler capital budgets are still climbing, about $800 billion a year across the four, $165 billion in the latest quarter, and spending has crossed operating cash flow: Alphabet's second-quarter free cash flow was negative $6 billion, its first negative quarter since the 2004 IPO. Once capex exceeds internal cash, the buildout needs outside money, and creditors now demand a premium on hyperscaler paper. That is the mechanism that turns scale from a moat into a test: the market rewards cash-backed returns and charges for faith, so every new AI dollar has to out-earn its financing cost. The bullish case regains force only if Alphabet's larger 2027 budget lifts the stock, cash generation catches up with investment and the two leading labs fund growth on durable, non-circular revenue.
Approved by Marco Cauduro on October 10, 2026.
The DeepStack callSkeptical
Capital intensity is not a moat. Every new AI dollar now has to out-earn its financing cost.
Our Call takes the side “The return is not there”.
Call history
- v2Prove the returnConviction High (4/5)Side: The return is not thereSide named by the owner.
- v1Prove the returnConviction High (4/5)First call, as published with the story. The conviction, the actions and the flips were set down for the site on October 7, 2026.
6 voices
The return is not there
Warden Capital argues the AI investment cycle cannot earn a reasonable return absent AGI within a few years: at roughly $700 billion per year of investment, model providers would need about $1 trillion in annual end AI revenue by 2029, ramping…
Warden Capital says they are relatively confident the AI boom ends in a bust but do not know when, are not certain it causes a US recession, and cannot judge how deep one would be; they add that the debt-heavy neocloud and datacenter financing…
He argues the consensus revenue path is stretched, since the ~$863 billion of incremental annual AI revenue by 2030 would require roughly $100 per month of extra AI spending from every person in the US and EU including children, and equals 19.1x…
He argues the hyperscalers' AI capital spending may destroy shareholder value: consensus 2026E-'30E capex of about $3.5 trillion against ~$863 billion of incremental revenue at a 40% incremental margin and 20% tax rate implies roughly an 8%…
Global AI spending is projected to reach 2.59 trillion dollars in 2026, and to achieve even a relatively low unlevered return on investment would require 207 billion dollars in after-tax income per year starting immediately, which the portfolio…
AI-related stocks have accounted for 50% to more than 75% of the S&P 500's total gains since late 2022, and AI companies now constitute approximately 50% of the S&P 500, exceeding even the Information Technology sector's 35% weight at the S&P 500's…
5 voices
The spending is defense
The Mag7 capex boom should be read the same way: the risk of not investing heavily is much higher than the risk of overinvesting, which is why Alphabet announced more investment and others will follow, since their existence is at stake.
Hyperscalers have poured free cashflow and debt into computational infrastructure to meet AI's demands, creating a generational run in semiconductors where demand has outpaced supply.
AI usage is highly concentrated among power users and top-decile enterprises, which have increased token output by more than 17 times since April 2025, creating an approximately 8-fold gap between typical and top-decile enterprises, and a gap…
Unlike dot-com era suppliers, today’s hyperscalers fund AI investment from profitable core businesses such as advertising, cloud, software, subscriptions, and e-commerce, so the question is not whether they can afford the spending.
The most important signal for AI-era investors is the relationship between capital spending, free cash flow, and the growth those investments eventually produce, because in the dot-com boom stock prices turned before earnings and cash flow collapsed.
6 voices
Voices that cross sides
If AI becomes cheaper and commoditized, spending hundreds of billions of dollars becomes much harder to justify economically.
The bullish case compares current AI infrastructure buildout to building the Las Vegas Strip in the 1950s, arguing that enormous upfront investment creates an eventually valuable economic ecosystem.
The current hyperscaler capex boom, with Alphabet, Amazon, Meta, and Microsoft together spending close to $800 billion in 2026, requires external capital funding for the first time because capex now exceeds internal cash flow.
AI chip prices are currently inflated by speculative demand from companies ordering more chips than they need to secure allocation, and reselling them at premiums—a dynamic similar to Rolex watch secondary market pricing.
The widening in AI-related credit is largely a supply-and-duration absorption story rather than fundamental deterioration, and saying so is true but the least informative thing one can say about it.
AI-related issuance across investment grade, high yield and loans stands at $489bn year to date against $322bn for all of 2025, is 23% of USD IG gross supply this year, and for the first time the six largest tech names carry more risk in the US…
Where they cross
Who answers whom.
The specific points where one voice meets another: the same evidence read two ways, or the same mechanism with a different sign.
Quanta 72 qualifies Steve Eisman, on a condition
Warden Capital contradicts Steve Eisman, on the evidence
Steve Eisman qualifies Quanta 72, on a condition
JunkBondInvestor qualifies Moses Sternstein, on the mechanism
Warden Capital qualifies JunkBondInvestor, on the mechanism
Alternative explanations
Three other readings.
What would have to be true for the Call to be wrong, as the other side argues it.
- The spending is defense, so the return test is the wrong test: the risk of not investing is higher than the risk of overinvesting when a company's existence is at stake (the case Kris makes).
- The hyperscalers fund the buildout from profitable core businesses in advertising, cloud and software, unlike the dot-com suppliers, so affordability was never the question (the case Quanta 72 makes).
- The return is visible where the contracts are: Nebius, with multibillion-dollar deals with Microsoft and Meta, has a credible path to more than 4 GW of capacity by 2031 (the case Orbis Investments makes).
What settles it
The dated tests.
The tests the story set, each with its specification where the owner has written one and the piece of the argument it verifies; on the Docket, resolved on the Results page.
Alphabet discloses its higher 2027 budget; a rise in the stock revives the defense case, another fall validates the return test.
Part 04: AI Spending Has Met Its First Real Bill Pendingp04-t01Pre-specifiedChecks the defense claim
- Question
- Did Alphabet shares close higher on the first full trading day after Alphabet disclosed a 2027 capital-spending budget above its 2026 budget than at the close before the disclosure?
- Source
- Alphabet investor relations, quarterly earnings release and call
- Rule
- Read the 2027 capital-expenditure figure or range Alphabet gives in the release or on the call, and the Nasdaq close of GOOGL on the first full trading day after it against the close before. The answer is yes if the 2027 budget is above 2026's and the close is higher; no if the close is lower or unchanged.
- If the source is late, revised or silent
- If Alphabet gives no 2027 figure by February 15, 2027, or guides 2027 at or below 2026, the test is inconclusive, and the Result says what was read and what was missing.
The Call, before the dateExpects no. Gives the yes answer 45%.
0 · no100 · yes
- Next lab financing
OpenAI and Anthropic revenue and funding terms emerge; deceleration, not decline, is the trigger for a reflexive break.
Part 04: AI Spending Has Met Its First Real Bill Pendingp04-t02Open questionChecks the lab-revenue claim
Specification pending
Hyperscaler free cash flow shows whether operating cash flow catches capex growth; a catch-up makes external funding temporary.
Part 04: AI Spending Has Met Its First Real Bill Pendingp04-t03Open questionChecks the cash-catch-up claim
Specification pending
$50B–60B of queued paper prices; narrower concessions support absorption, wider spreads signal borrower scrutiny.
Part 04: AI Spending Has Met Its First Real Bill Call heldAnswer: nop04-t04Checks the absorption claim
- Question
- Did the investment-grade corporate bond spread (ICE BofA US Corporate Index option-adjusted spread, FRED series BAMLC0A0CM) read tighter on October 2, 2026, the end of the four weeks after Labor Day, than its August 2026 average?
- Source
- FRED, ICE BofA US Corporate Index option-adjusted spread (BAMLC0A0CM)
- Rule
- The reading on October 2, 2026 (the last observation on or before that day) against the mean of the daily observations from August 1 to August 31, 2026: confirmed if it is 3 basis points or more below the August mean, refuted if it is 3 basis points or more above it. The split between AI-related issuers and the rest of the market, as the press reported it on September 22, 2026, is read into the Result as evidence and does not decide the test.
- If the source is late, revised or silent
- A series that is late, revised after the reading or silent for the window is inconclusive, reviewed by the owner, and the Result says what was read, what was missing and the next step. A difference inside the 3 basis point band is inconclusive and closes the test.
- Token and productivity data
Prices, volumes and productivity move together; sustained volume growth over price and a visible productivity surge support the bulls.
Part 04: AI Spending Has Met Its First Real Bill Pendingp04-t05Open questionChecks the productivity claim
Specification pending
What changed
The record of Part 04.
The Part’s publication, every change to the Call, every result that came in and the next dated test, oldest first.
- PublishedPart 04 published; the Call: Prove the return.
- Call updatedThe Call: Prove the return. Capital intensity is not a moat. Every new AI dollar now has to out-earn its financing cost. Our Call takes the side “The return is not there”. by Marco Cauduro
- Next testNext: Jan 1, 2027, Alphabet discloses its higher 2027 budget; a rise in the stock revives the defense case, another fall validates the return test. Pre-specified
Sources (8)
- Potential Multibaggers - July 26 spending case
- a16z - July 31 equipment backlog
- a16z - August 21 usage data
- Warden Capital - second-quarter letter
- First Pacific Advisors - Crescent Fund letter
- Prof G Markets - August 4 capex tally
- Voss Capital - second-quarter letter
- JunkBondInvestor - July 26 credit weekly
Market data are as of the dates cited. Voices are paraphrased from public writing and attributed by name.
Analysis and opinion, not investment advice. Voices are paraphrased from public writing and attributed by name; DeepStack shows no score, ranking or accuracy for any person. Every question, one page each.