How DeepStack settles a question · Part 08

Who funds the buildout?

Who writes the checks for the buildout?

From Part 08: AI Lenders Are Financing the End of Scarcity. 13 voices on 6 sides, attributed and dated; the DeepStack Call with its history; 5 dated tests that settle it; the results as they come in.

The mechanism

Lenders are financing the capacity that will end the shortage they lend against. The spending has raced ahead of demonstrated revenue: the cash-flow forecasts behind the hyperscaler credit story need operating cash flow to roughly triple, from about $600 billion to $2 trillion, while off-balance-sheet commitments across five hyperscalers sit near $2.7 trillion and Oracle's free cash flow runs at negative $28.7 billion. Scarcity supports the collateral today, but every dollar lent adds the supply that erodes it, so a shortage of power, chips and capacity is a poor long-lived security against a fixed contract. The Call would change if September-quarter cash flow moved visibly toward $2 trillion, customer concentration fell sharply and the first Blackwell renewal held its price without new guarantees.

Approved by Marco Cauduro on October 10, 2026.

The DeepStack callSkeptical

Lenders are right to finance capacity and wrong to assume it stays scarce. Scarcity is a season; debt is a contract.

Lenders are mispricing scarcity · Conviction High (4/5)

Our Call takes the side “The cash has to triple”.

Read the full call

Call history2 versions · last changed
  1. v2Lenders are mispricing scarcityConviction High (4/5)Side: The cash has to tripleSide named by the owner.
  2. v1Lenders are mispricing scarcityConviction 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 cash has to triple

  1. Torsten Slok

    The tech silo is betting on a future in which demand for AI and tech services explodes, while the silos covering the companies that would pay for those services see a much more modest outlook, and both cannot be right at the same time.

    Source

  2. Torsten Slok

    In S&P 500 Q2 earnings season data, AI adoption keeps broadening: 69% of companies now point to a live deployment, up from 64% last quarter.

    Source

  3. Michael Burry

    Alibaba's shift to financing an AI data center buildout through share issuance is a demerit against the investment thesis, as the buildout may be 'practically interminable' without a clear productive path.

  4. Michael Burry

    Most analysts covering the former software/social-media/AdTech companies now recast as hyperscalers have never had to model maintenance capital expenditure, and most still are not doing it, or not doing it well.

    Source

  5. Daniel Romero

    Using a model assuming equipment costs rise to $100 billion per GW by 2030, five-year equipment cost recovery, 25-year site cost recovery, a 15% annual return on invested capital, and a 65% cloud operating margin, AI labs would need to reach $6.8…

    Source

  6. Daniel Romero

    Daniel Romero states they are not arguing the buildout is unsustainable, but that one-year compute paybacks will probably become the exception rather than a baseline assumption, possibly dependent on specific timing and shortages, and that 2026…

    Source

5 voices

The spending pays

  1. Market Sentiment

    Based on their capex sustainability model, Market Sentiment estimates that next year's ~$1.2T+ of AI capex should drive ~$2T of incremental revenue, which they say will only be a fraction of the total incremental revenue AI generates as a business…

    Source

  2. Market Sentiment

    Aggressive capital spending tends to be suboptimal because after an initial euphoric rise in stock prices tied to rising capex, investors realize they overpaid for the underlying inputs, from land to memory to fiber-optic cables.

    Source

  3. Aurelion Research

    The customer base is broadening beyond a concentrated set of hyperscalers to enterprises, AI-native companies and cloud providers, with AI-related infrastructure commitments expected to exceed $3.1 trillion, making the next phase of demand less…

    Source

  4. Aurelion Research

    Hyperscaler capex is durable rather than at risk of a sudden cut, because the largest spenders keep reaffirming the necessity of AI investment and their EBITDA and EBIT margins are still rising, giving them capacity to keep funding it.

    Source

  5. Kris

    AWS has had more demand than supply, with customers wanting more AWS capacity for AI than could be served despite large Nvidia orders and Amazon's own Trainium chips, and Kris considers this kind of capacity-constrained spending among the best…

    Source

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.

Alternative explanations

Three other readings.

What would have to be true for the Call to be wrong, as the other side argues it.

  1. The checks are already written by customers: AWS has had more demand than it can serve despite large Nvidia orders and its own Trainium chips, the best kind of spending there is (the case Kris makes).
  2. The spending pays for itself: next year's $1.2 trillion or more of AI capex should drive about $2 trillion of incremental revenue, a fraction of what AI generates as a business (the case Market Sentiment makes).
  3. Capex is durable, not fragile: the largest spenders keep reaffirming it while their margins still rise, and the customer base is broadening beyond the hyperscalers (the case Aurelion Research 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.

  1. September-quarter results test cash flow toward $2 trillion and whether maintenance costs enter accounting.

    Part 08: AI Lenders Are Financing the End of Scarcity Pendingp08-t01Pre-specifiedChecks the cash-flow tripling claim

    Question
    Did the combined net cash from operating activities of Microsoft, Alphabet, Amazon and Meta for the quarter ended September 30, 2026 exceed their combined figure for the quarter ended June 30, 2026?
    Source
    SEC EDGAR, the four companies' earnings releases (Form 8-K) and quarterly reports (Form 10-Q)
    Rule
    Sum the net cash from operating activities each of the four states for the September quarter and compare it with the same sum for the June quarter. The answer is yes if the September sum is higher; no if it is equal or lower. Whether maintenance capital spending or shorter server lives entered the accounting is DeepStack's reading from the same filings and goes in the Result, not in the answer.
    If the source is late, revised or silent
    If any of the four has not reported by November 7, 2026, or restates its June quarter, the test is inconclusive, and the Result says what was read and what was missing.

    The Call, before the dateEither answer fits the Call. Gives the yes answer 80%.

    80%

    0 · no100 · yes

  2. New Mexico rules on the fuel-cell permit; approval supports schedule, denial turns Oracle’s notice into a warning.

    Part 08: AI Lenders Are Financing the End of Scarcity Pendingp08-t02Pre-specifiedChecks the permit-schedule claim

    Question
    Did the New Mexico Environment Department approve the air-quality permit for the fuel cells at the Project Jupiter campus in Doña Ana County by November 23, 2026?
    Source
    New Mexico Environment Department, Air Quality Bureau permit decisions
    Rule
    The answer is yes if the department publishes a final permit approval, with or without conditions, on or before November 23, 2026; no if it publishes a denial by that date.
    If the source is late, revised or silent
    A decision deferred past November 23, a permit withdrawn by the applicant, or an approval under appeal on that date is inconclusive, and the Result says what was read and what was missing.

    The Call, before the dateExpects yes. Gives the yes answer 55%.

    55%

    0 · no100 · yes

  3. Pipeline in-service date tests whether Project Jupiter can secure fuel for its planned campus.

    Part 08: AI Lenders Are Financing the End of Scarcity Pendingp08-t03Open questionChecks the fuel-supply claim

    Specification pending

  4. First Blackwell renewal

    Contract repricing shows whether scarce compute retains value or lenders face faster obsolescence.

    Part 08: AI Lenders Are Financing the End of Scarcity Pendingp08-t04Open questionChecks the scarce-compute value claim

    Specification pending

  5. Fluidstack terms

    Government-loan terms reveal the price Washington assigns to financing bottlenecks it wants to remove.

    Part 08: AI Lenders Are Financing the End of Scarcity Pendingp08-t05Open questionChecks the public-financing claim

    Specification pending

What changed

The record of Part 08.

The Part’s publication, every change to the Call, every result that came in and the next dated test, oldest first.

  1. PublishedPart 08 published; the Call: Lenders are mispricing scarcity.
  2. Call updatedThe Call: Lenders are mispricing scarcity. Lenders are right to finance capacity and wrong to assume it stays scarce. Scarcity is a season; debt is a contract. Our Call takes the side “The cash has to triple”. by Marco Cauduro
  3. Next testNext: Oct 29, 2026, September-quarter results test cash flow toward $2 trillion and whether maintenance costs enter accounting. Pre-specified
Sources (8)
  1. TechCrunch - Oracle force majeure notice
  2. Seeking Alpha via TradingView - Oracle CDS record
  3. FRED - ICE BofA CCC spread
  4. Apollo - technology cash-flow mismatch
  5. Apollo - hyperscaler credit assumption
  6. Apollo - AI spending concentration
  7. Apollo - AI financing market
  8. Aurelion Research - Nvidia customer base

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.