Who is on each side · Part 14
Is AI revenue arriving fast enough for the biggest spenders to pay for their data centers out of their own cash?
22 voices who disagree about The Cash Arrives Unevenly. Each voice is attributed and paraphrased, dated where the date is known, with a link where one exists.
The DeepStack callSplit
Microsoft pays for its build from its own cash. Alphabet and Meta spend most of theirs; Amazon spends more than its operations bring in.
Our Call takes neither side alone.
13 voices
The cash still covers it
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…
Modeled bottom-up, 100 billion dollars of annual data-center capex sustains roughly 160 billion of steady-state annual revenue, a 1.6x revenue-to-capex multiple and about a 21 percent IRR, even assuming a five-year chip life and a one-third annual…
A cracking labor market is a real risk for anything tied to the consumer, and the AI buildout being funded by hyperscaler cash flows and multi-year contracts is why, for now, it stays on the watchlist and doesn't change the thesis.
Because every cloud and hardware dollar traces back to someone paying for AI at the top, the application and model boxes are the demand test for the whole map; in 2030 the stack would spend $3-4T a year on hardware while apps and models bring in…
Anthropic itself is not borrowing the money; it is the hyperscalers (he names Meta, Google, Amazon, Microsoft, Oracle) who are borrowing to use Anthropic's products, so if Anthropic were to go bust, that would be an equity problem rather than a…
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.
Hyperscaler capex has come in more than 56% above what the market was forecasting a year ago, with spending estimates continuing to be revised upward without signs of slowing.
Citing management's account of two distinct capital cycles, roughly two years of data center construction before any revenue against 30-plus years of monetization, and about three years to break even on servers with a five to six year life, Invest…
Amazon expects roughly $220 billion in cash CapEx in 2026, up from an earlier $200 billion estimate, driven primarily by higher memory chip costs.
Because Oracle could not keep funding the AI buildout on its own without loading hundreds of billions in debt that could endanger the company once the AI boom cools, Oracle shifted to a 'saving move' in which customers either bring their own GPUs…
Kakashii estimates that after excluding roughly 200B of legacy and non-AI RPO (including ByteDance's U.S. datacenter deal), Oracle's remaining RPO of 464B is attributable to AI Compute Cloud, of which close to 300B-350B is attributed to OpenAI…
Axios reported, and the Financial Times later confirmed the same day, that OpenAI's annual recurring revenue is nearing $70 billion, with enterprise sales more than doubled since July according to sources familiar with the financials.
Tae Kim believes that if OpenAI were public, its stock would have gapped up 30% or more on hitting $70 billion in ARR, a blowout versus expectations on the level of Nvidia in May 2023, and that the figure proves his August report that OpenAI's ARR…
13 voices
The bill has moved to debt
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.
The sustainability and predictability of incremental returns on invested capital in hyperscaler AI projects matters more to valuations than the absolute level of those returns, because the market values certainty.
Hyperscalers including Google, Microsoft, Amazon and Meta funded more than half of compute growth so far but now do not generate cash, spending everything on CapEx and raising debt on top, so for a roughly $3 or $4 trillion 2028 CapEx level the…
Headline AI CapEx figures of $40-50 billion cover only critical IT such as servers, networking, fiber, transceivers and optics, not the data center shells or power plants, which must be built ahead of time, so a 100-gigawatt-a-year build is much…
Google spent $45 billion and Amazon $54 billion on capital expenditures in a single quarter, and both ended up with negative cash flows despite huge cash intake.
The AI investment cycle is entering a new phase focused on economic returns rather than just capital expenditure.
Meta is moving toward the upper end of its $130-145bn capex range, fueling broader market concerns about negative free cash flow and the lack of clear returns on massive AI investments.
Across the five hyperscalers, uncommenced, non-cancelable data-center leases and purchase commitments together total roughly $2.7 trillion in off-balance-sheet contractual commitments as of the disclosed filing dates, with the total potentially…
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.
Hyperscaler capex for 2027 tracks as high as 1.2 trillion dollars with about 400 billion of debt issuance penciled for next year; AI-related paper doubled from about 3 percent to 6 percent of the investment-grade index in a year, Amazon became the…
A $5 billion Pentagon loan to an $18 billion company signals that government is done leaving the pace of AI investment to private hurdle rates, and while private capital keeps working, it is now being joined by capital priced to different…
Given the cost for compute, Meta's free cash flow is going to crater in 2027 to a very suspect degree, and like Oracle, betting on a product tied to accelerating compute demand at any cost is borderline insane to Michael Spencer.
A Financial Times report found OpenAI's annualized revenue as of late September 2026 was approximately $50 billion, roughly $20 billion lower than the $70 billion figure that had previously circulated in financial media.
14 voices
The buyers have not paid yet
Analysts covering tech expect the sector's operating cash flow to more than double to roughly $2.4 trillion by 2028, an increase of over $1.2 trillion, while analysts covering the other S&P 500 sectors, which are tech's customers, expect those…
Consensus expects hyperscalers' operating cash flow to more than triple by 2030, outpacing the sharp increase in AI-driven capex, based on FactSet and Apollo Chief Economist sources for Google, Meta, Amazon, Microsoft and Oracle.
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…
Although SpaceX reports AI infrastructure payback periods of less than one year, maintaining a one-year payback across the entire hypothetical 2027 buildout would require $1.9-2.28 trillion in annual compute payments, and if AI labs footed that…
OpenAI told investors it hit roughly $50 billion in annualized revenue at the end of September, a fairly large gap from the $68 billion figure widely reported the prior month.
OpenAI also told investors it achieved 77% run-rate growth in the third quarter and 107% growth in its enterprise business, but even that growth was completely overshadowed by the $20 billion gap in revenue, with Wall Street reacting in many ways…
The AI trade's foundations are shaky because roughly seventy percent of the hyperscalers' AI revenue traces back to just two customers, OpenAI and Anthropic, both of which are burning enormous amounts of cash, so if either stumbles the entire…
George Noble's bear case for AI rests on the question of actual return on investment; he argues that Nvidia announcing massive funding packages just to keep customers like OpenAI able to keep buying chips shows the money underneath the boom is not…
They are uncertain about the sustainability of the current and projected level of AI capital spending and the prospect of it generating reasonable returns on the aggregate investment.
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…
He states it is equally possible that AI transforms business and society yet fails to earn an adequate return for the companies that built it, since revenue can arrive well before owner earnings if depreciation understates obsolescence or…
AI capex has become a prisoner's dilemma financed increasingly by leases, purchase commitments, debt and in one case an equity raise: each firm may invest rationally to protect existing businesses while the industry collectively commits too much…
They say it remains uncertain to them how likely it is that AI spending dramatically slows as Wall Street and the AI funders realise they cannot get a return.
They conclude either the AI companies will not earn a return on the trillions they are spending, or they will earn it at the cost of very large short-term unemployment, or most likely a combination of both.
8 voices
Voices that cross sides
That gap can hold for a while because hyperscalers fund capex from non-AI businesses, debt covers part of it and hardware is paid off over several years, but eventually it has to close from the top, and if the $2.1T arrives late every floor…
AGI is not the investable trade; compute is, because Astra adds a second, inference-driven demand chain (more capable models, more agents, more delegated work, longer inference trajectories, more tokens, more compute) alongside the existing…
In the data center buildout, hyperscalers have been able to raise all the capital they have asked for, but they are paying more for it: spreads on their longest-dated bonds have widened, and most 2026-issued paper trades wider today than where it…
The credit story for hyperscalers depends on a single consensus assumption that operating cash flow triples from $600 billion to $2 trillion.
Oguz Erkan does not regard the AI capex fear as valid for this cycle: capex went from $28 billion in 2023 to $72 billion last year and is expected at $130-$145 billion this year and $202 billion by 2030, but committed capital is already earning ROI…
The loop's binding constraint is capex finance: hyperscaler free cash flow may go negative in 2027 and investors dislike leverage contingent on speculative returns, as the depressed valuations of Oracle and CoreWeave show, so continued spending…
Labs have not yet reached the point where their own cash flows can fund the scale of CapEx required, since labs are projected to generate hundreds of billions in revenue next year while total CapEx across the compute supply chain will be well north…
His current modeling has $5 trillion of credit-funded and $6 trillion of cash-funded infrastructure investment through 2029, and this is not enough compute relative to projected AI demand growth, so revenue per megawatt has to keep rising.
Notes
Not a disagreement.
Where two voices only seem to differ, or where the thread runs back to an earlier Part.
The thread
Part 4 said the bill had arrived; Part 8, that the cash must triple. Part 14 reads, company by company, who still pays in cash.
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.
Ren qualifies Michael Spencer, on the mechanism
Guinness Global Investors qualifies Torsten Slok, on the evidence
Guinness Global Investors contradicts Torsten Slok, on premise
Michael Spencer qualifies Torsten Slok, on the mechanism
Torsten Slok qualifies Hayden Capital, on the mechanism
Steve Eisman contradicts Michael Burry, on the mechanism
Torsten Slok qualifies Longriver, on the mechanism
Michael Spencer qualifies Ren, on a condition
Ren contradicts Nathaniel Whittemore, on premise
Michael Spencer contradicts Steve Eisman, on the mechanism
Torsten Slok qualifies First Pacific Advisors, on the evidence
Michael Burry qualifies JunkBondInvestor, on the mechanism
Michael Spencer qualifies Ren, on the evidence
Bronte Capital qualifies Kakashii, on a condition
Michael Spencer qualifies Ren, on the mechanism
Michael Burry refutes Steve Eisman, on the mechanism
Bronte Capital qualifies JunkBondInvestor, on a condition
Torsten Slok qualifies Steve Eisman, on the evidence
Torsten Slok qualifies Hayden Capital, on a condition
Torsten Slok qualifies Market Sentiment, on a condition
Bronte Capital qualifies Market Sentiment, on a condition
Hayden Capital qualifies Michael Spencer, on the mechanism
Guinness Global Investors qualifies Invest in Assets, on the mechanism
Aurelion Research refutes Michael Burry, on the evidence
What would settle it
The dated tests.
The same tests the story set, on the Docket; results land on the Results page.
Alphabet's September quarter: whether capital spending takes 80 cents or more of each dollar of operating cash, after 49.5 cents a year earlier and 95.0 cents in the first half of 2026.
The four's September-quarter reports on EDGAR: whether combined purchases of property and equipment topped combined operating cash, after 96.1% in the June quarter of 2026.
The 2027 budgets of Alphabet, Amazon and Meta: whether the midpoints sum to at least 25% more than the three spent in 2026.
The full year: whether the four's purchases topped their operating cash across calendar 2026, after 91.4% in the first half.
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.