AI Spending, Debt and Cash Flow: Is the Tech Selloff Panic, or a Real Warning?
Independent investigation · Published November 3, 2026 · Evidence as of November 3, 2026
On October 29, Alphabet fell about 13% in a day — its worst session since it listed in 2004, roughly $500 billion of market value gone — after reporting a quarter in which Google Cloud revenue grew 103%. The next day the Nasdaq 100 shed more than $1 trillion. A widely watched channel is now asking whether this is overdone panic or the start of a bear market, and its answer leans on AI cash burn, financing risk and profit risk. Four of the five companies in this story sit in my own portfolio, so I did not borrow the question's answer. I checked the hard numbers myself, chain by chain: how much is being spent, whether operating cash still covers it, whether cloud revenue is keeping up, how the spending is now financed, and how much balance-sheet strength is left. My verdict is split, and the split is the point: on demand, the selloff overreacted. On financing, something real has changed — and it does not reverse just because the panic fades.
Hot-Topic Investigations No. 4 — the “AI cash burn” question, checked against five chains of evidence. Data as of November 3, 2026.
Video edition
The video edition of this report was published on November 3, 2026, on the Axing's US Stock Diary channel (@axingstocks). Watch it above, or on YouTube via the link below.
After checking all five chains, the evidence as of November 3 does not support “the AI build-out is breaking”. Cloud revenue is accelerating at all four hyperscalers whose numbers I could verify — Azure +48%, Google Cloud +103%, AWS +39.3% and accelerating for a sixth straight quarter, Oracle's cloud infrastructure +121% — and the order books behind the spending are enormous (AWS backlog $578B, Oracle RPO $664B). What the evidence does support is narrower and, to me, more important: the funding model has turned. Four of the five companies printed negative free cash flow in the latest quarter; the five issued about $326B of new debt in 2026, roughly three times 2025; Meta is adding a roughly $95B equity programme on top of off-balance-sheet project debt; and the bond market has started charging real money for this paper — the $27.3B Hyperion project bond trades in the low nineties. The risk did not disappear. It moved — from “will anyone use this capacity?” to “what does it cost to finance, and who holds the paper if the music slows?”.
Scale: 2026 capex guidance across Microsoft, Alphabet, Amazon and Meta sums to about $720-740B (our addition of the four guides; external estimates cluster in the same range), and every one of them says 2027 will be higher.
Cash coverage: latest-quarter capex ran at about 68% of operating cash flow at Microsoft (cash basis) — and about 121% at Alphabet, 106% at Meta, 182% at Amazon and 123% at Oracle (our calculations from the reported lines). Only Microsoft printed positive free cash flow.
Demand: cloud growth is not stalling — it is accelerating into the spending, with contracted backlogs in the hundreds of billions. This is not “build it and hope”.
Financing: debt, equity and SPV structures now fund the gap. Meta's own filing puts its maximum exposure on the off-balance-sheet Hyperion campus at about $46B — off the balance sheet is not off the hook.
Five chains of evidence, checked one by one
Before checking anything, I fixed the standard, so it cannot move mid-argument. “Real risk” would have to show up as: cloud revenue growth stalling while capex accelerates; free cash flow negative for consecutive quarters with no path back; financing shifting from choice to necessity — rising yields, widening spreads, structures that hide leverage; and order books thinning. “Overdone panic” would show the opposite: demand accelerating, contracts funding the build, and balance sheets absorbing the gap. Here is how each chain scores today.
#
Evidence chain
Verdict
Basis
1
The spending itself — bigger than feared, and still being revised up
Confirmed, at full scale
Latest-quarter capex: Microsoft $45.2B including finance leases (call transcript), Alphabet $51.4B, Amazon $77.3B of PP&E purchases (8-K release), Meta $34.91B, Oracle $28.5B; 2026 guidance of about $170B / $195-205B / about $220B / $135-145B respectively
2
Cash coverage — four of the five now spend more than operations generate
The genuine change
Latest-quarter free cash flow: Microsoft +$18.6B; Alphabet about -$9.1B (second negative quarter in a row); Meta -$2.12B (first negative quarter in over a decade); Oracle -$5.4B on its own published table; Amazon trailing-twelve-month free cash flow -$42.4B
3
Is cloud revenue keeping up with the spending? So far, yes — it is accelerating
No demand break
Azure +48% with next-quarter guidance of +49-50% in constant currency; Google Cloud +103% to $30.7B; AWS +39.3% to $46.0B, a sixth straight acceleration, backlog $578B; Oracle cloud infrastructure +121% to $7.4B, RPO $664B
4
The financing mix — the build now runs through bond desks, equity and SPVs
The model has changed
About $326B of new debt from the five in 2026, roughly three times 2025 (S&P Global data); Alphabet's €6B sale on Nov 2 after €9B in May; Oracle's $10B offering filed Nov 2; Meta's roughly $95B equity programme; the $27.3B Beignet/Hyperion SPV bond trading in the low nineties
5
Staying power — these are still some of the strongest balance sheets on earth
Strained, not fragile
Amazon held $114.2B of cash and marketable securities at Sep 30; Meta $82.45B against about $81B of long-term debt; Microsoft still free-cash-flow positive with $10.2B returned to shareholders in the quarter; Oracle's build is partly pre-funded by $11.4B of customer prepayments
1. The spending itself — bigger than feared, and still being revised up
Confirmed, at full scale
Start with the sheer scale, because the selloff was partly a scale shock. In the September quarter alone, Microsoft spent $45.2B of capex including finance leases (cash paid for property and equipment was $39.4B, more than double the $19.4B of a year earlier); Alphabet spent $51.4B against $24.0B a year earlier; Amazon's cash-flow statement shows $77.3B of property and equipment purchases against $35.1B; Meta spent $34.91B including finance-lease principal; Oracle spent $28.5B in its August quarter — about half of its entire fiscal-2026 capex in a single quarter.
None of them blinked on guidance. Microsoft put calendar-2026 capex at about $170B and the December quarter above $50B; Alphabet held its July-raised $195-205B; Amazon held about $220B (raised earlier in the year from about $200B, with higher memory costs named among the reasons) and said 2027 will be meaningfully higher; Meta narrowed its range upward at the low end for the fourth time this year — $115-135B in January, $125-145B in April, $130-145B in July, $135-145B now. Management teams that feared a demand cliff do not guide like this. Whether the spending is wise is a separate question; that it is deliberate is not in doubt.
Company (latest quarter)
Quarter capex
Full-year guide
Microsoft (Q1 FY27)
$45.2B incl. finance leases; $39.4B cash
About $170B for calendar 2026
Alphabet (Q3 2026)
$51.4B
$195-205B for 2026
Amazon (Q3 2026)
$77.3B PP&E purchases; $75.9B cash capex per the call
About $220B for 2026
Meta (Q3 2026)
$34.91B incl. finance-lease principal
$135-145B for 2026
Oracle (Q1 FY27)
$28.5B
$90-95B for fiscal 2027 (call commentary)
2. Cash coverage — four of the five now spend more than operations generate
The genuine change
This is the chain where the story actually changed. A year ago these companies funded the AI build out of operating cash and still had cash left over. In the latest quarter, Alphabet's free cash flow was negative for a second consecutive quarter (about -$9.1B after about -$5.9B in Q2 — its first negative prints since its 2004 IPO), Meta's turned negative for the first time in more than a decade (-$2.12B, down from +$10.63B a year earlier), Oracle printed -$5.4B on the free-cash-flow table in its own release, and Amazon's trailing-twelve-month free cash flow swung to -$42.4B from +$14.8B a year earlier. Reuters said it plainly on November 2: Microsoft was the only Big Tech company still generating cash in the quarter — and even there, free cash flow fell 27% year on year to $18.6B.
Two honest qualifications. First, a negative quarter is not a solvency event at these balance sheets (chain five). Second, part of the squeeze is timing and mix, not vanishing demand: Alphabet is building TPU inventory ahead of 2027 revenue recognition, and Oracle's operating cash flow was flattered by $11.4B of customer prepayments — customers funding the build in advance, which is a demand signal wearing a financing costume. But the direction across four unrelated companies in the same quarter is not noise. The internal-funding era of the AI build is over; what replaced it is chain four.
Company
Operating cash flow
Capex
Free cash flow
Microsoft
$58.0B
$39.4B cash ($45.2B incl. leases)
+$18.6B
Alphabet
About $42.3B (derived)
$51.4B
About -$9.1B
Amazon
$42.4B
$77.3B PP&E purchases
TTM -$42.4B (company measure)
Meta
$32.79B
$34.91B
-$2.12B
Oracle
$23.1B
$28.5B
-$5.4B
3. Is cloud revenue keeping up with the spending? So far, yes — it is accelerating
No demand break
The fastest way to falsify the panic would be cloud growth rolling over while capex accelerates. The opposite printed. Microsoft's Azure grew 48% — its fastest in at least two years — and management guided the next quarter higher still, at 49-50% in constant currency, while repeating that demand exceeds supply. Google Cloud more than doubled, +103% to $30.7B. AWS grew 39.3%, its fastest pace in nearly five years and its sixth consecutive acceleration, and Amazon put its contracted backlog at $578B, up from $496B one quarter earlier. Oracle's cloud infrastructure revenue grew 121%, and its remaining performance obligations reached $664B.
Capex is growing faster than cloud revenue in percentage terms at three of the four (roughly +103-120% year on year against +39-103% revenue growth) — that gap is exactly what chain two measures, and it is why free cash flow is negative. But a build-out whose revenue line is accelerating into the spending, with hundreds of billions already contracted, is a different animal from speculative overbuilding. The demand side of the panic case is, on this quarter's evidence, the weakest side.
4. The financing mix — the build now runs through bond desks, equity and SPVs
The model has changed
Because operating cash no longer covers the build (chain two), the difference is now raised outside. S&P Global Market Intelligence puts the five companies' combined new debt in 2026 at about $326B, roughly three times the 2025 level; LSEG data via Reuters has US technology-related issuance above $250B this year, up 81%. The tape is live this week: Alphabet launched a €6B seven-tranche sale on November 2 (demand above €17B) after €9B in May and a 100-year sterling bond in February; Oracle filed a $10B note offering the same day. Meta went further than debt: in late October it announced an equity programme of roughly $95B (the reported totals run $95-97B depending on what is counted) — issuing shares to build data centres is a regime change for a company that spent a decade buying them back.
The structure to understand is Hyperion, Meta's Louisiana campus. A bankruptcy-remote vehicle, Beignet Investor, sold $27.3B of 6.58% notes due 2049; Blue Owl funds hold 80% of the venture, Meta 20%, and the debt does not sit on Meta's balance sheet. But Meta's own filings disclose a maximum exposure to loss of about $46B through its lease, equity and residual-value guarantee — off-balance-sheet is an accounting location, not an absence of obligation. And the market is pricing that distinction: the Beignet notes, issued at par in October 2025 and initially traded up, changed hands around 92 cents in late October for a yield of about 7.3% — closer to junk than to Meta's own AA-grade curve. Alphabet's filings show the same direction in quieter form: $130.5B of data-centre leases signed but not yet commenced, and not yet on the balance sheet.
5. Staying power — these are still some of the strongest balance sheets on earth
Strained, not fragile
The panic framing quietly assumes the spenders are fragile. They are not — yet. Amazon ended September with $114.2B of cash and marketable securities. Meta held $82.45B against about $81B of long-term debt, and its underlying operating income excluding a one-off $9.89B legal charge was $22.88B in the quarter — the ad engine still prints money; the charge and the capex, not the business, drove the headline miss. Microsoft generated $58.0B of operating cash in one quarter and still returned $10.2B to shareholders. Even Oracle, the most stretched of the five, collected $11.4B of customer prepayments in the quarter — its customers are co-funding the capacity they have contracted for.
That is why I score this “strained, not fragile”. The correct worry is not that these companies run out of money next quarter. It is that the margin of safety now depends on capital markets staying open and cheap, and on cloud growth holding its current slope for years — a dependence that did not exist when the build was funded from the till. Dependence is a risk even when the dependent party is strong.
Flagged, not settled
Meta's five-year CDS at about 119 basis points, described as the highest since its contracts listed — a single secondary source; I use it as colour on the credit-repricing direction, not as a load-bearing figure.
Oracle's fiscal-2027 capex guide of $90-95B comes from earnings-call commentary as summarised by data providers, not from the results release itself; labelled as call commentary wherever it appears.
Meta's equity programme total is reported at about $95B by some outlets and about $97B by Barron's, depending on which components are counted; I write “roughly $95B” and flag the range rather than picking one.
Alphabet's Q3 operating cash flow of about $42.3B is my derivation (reported free cash flow of about -$9.1B plus reported capex of $51.4B), not a line I found printed; the free-cash-flow figure itself is doubly sourced.
The other side's real cards
A check that only lays out one side is not a check. The “this time the risk is real” side holds three strong cards, and I am not going to pretend otherwise:
Negative free cash flow is a path now, not a blip
Meta's free cash flow has fallen four quarters in a row — $14.08B, $12.39B, $0.78B, then negative. Alphabet has printed two negative quarters in a row after never printing one since its IPO. And every management team simultaneously promises to spend more in 2027. A single negative quarter can be timing. Four companies, same quarter, same direction, with guidance for more — that is a trajectory, and trajectories are what credit analysts are paid to extrapolate.
The credit market is repricing faster than the equity market
The Hyperion project bond fell from above par to the low nineties in a year, at a yield around 7.3% — pricing closer to speculative grade than to Meta's own debt. Meta's CDS spreads were reported at record levels for the name. Equity investors can argue about 2028 earnings; bond investors are being paid to worry about who ultimately stands behind three decades of data-centre paper, and they are charging more for the worry than they did a year ago. When the two markets disagree, the bond market's version deserves a hearing.
The commitments you cannot see on the balance sheet are the fastest-growing line
Meta's Form 10-Q shows about $454.5B of non-cancelable purchase commitments plus $355.7B of leases that have not yet commenced — roughly $810B of committed future spending, against $82.45B of cash and securities. Alphabet shows $130.5B of uncommenced leases. None of this is hidden — it is disclosed, in the notes, in the small print — but it means the leverage ratios computed from balance-sheet debt alone now systematically understate what these companies have promised to spend. If AI revenue growth disappoints, these commitments do not shrink with it.
Where the risk actually moved
None of the three strong cards is about demand collapsing this quarter. All three are about the structure that now funds the build: consecutive negative free cash flow, repricing project debt, and commitments that live in the footnotes. The risk has not grown in the place the panic is looking — the data centres are being used, the backlogs are real. It has grown in the funding stack underneath, which is slower-moving and harder to reverse. That is a reason to watch the next two quarters closely, not a verdict that the build-out fails.
With the “bear market has begun” framing: a bear case needs demand rolling over, and this quarter demand accelerated at every hyperscaler I could verify, with backlogs growing. What rolled over was the market's willingness to fund the build at any price — a valuation and financing event, not yet a fundamental one.
With the “it's just panic, nothing changed” framing: something did change, permanently enough to matter. The internal-funding era ended this quarter — four of five free-cash-flow negative, a tripled debt pace, equity issuance, footnote commitments in the hundreds of billions. Buying the dip without updating for the new funding model is using last year's map.
With the originating video's panic-versus-bear-market binary: the honest answer is chain-dependent. Demand chain: panic looks overdone. Cash chain: the squeeze is real and guided to deepen. Financing chain: the repricing has started and is the thing to watch. One label for all five chains is exactly the kind of shortcut this column exists to avoid.
My judgment, in one paragraph
Strongly supported: as of November 3, 2026, the demand side of the AI build-out is not breaking — cloud growth accelerated across Microsoft, Alphabet, Amazon and Oracle in the latest quarter, contracted backlogs are in the hundreds of billions, and the spenders' operating businesses remain highly profitable. Also strongly supported: the funding model has structurally changed — free cash flow is negative at four of the five, 2026 new debt of about $326B is roughly triple 2025, Meta is issuing equity and shifting capacity into off-balance-sheet vehicles whose obligations it still ultimately backs, and project debt is being repriced by the credit market. My read of the selloff: mostly a pricing reset on real news — the market stopped paying “funded from cash” multiples for companies that now fund from markets — rather than evidence the build-out fails. What would change my mind: cloud growth decelerating below the capex growth rate by a wide and widening margin, a first capex guide cut, or project-bond prices sliding from the low nineties toward distress. I hold MSFT, GOOGL, AMZN and META; that position is disclosed here, and the judgment above follows the evidence, not the position.
What I am watching in the next two quarters
Q4 prints (late January): do Alphabet and Meta free cash flow stay negative, and how deep?
The first formal 2027 capex guides — every team has promised “more”; the numbers will show by how much
Meta's equity programme in practice: pricing, pace, and the Class C shares' market debut in January 2027
The Beignet/Hyperion notes: a recovery toward par would say the credit scare is fading; a slide says it is not
Cloud growth versus capex growth, each quarter — the spread between those two lines is the whole argument
How I checked this
I took the question actually circulating — is the AI cash burn a panic or a real risk? — fixed in advance what evidence each answer would require, and checked five chains against it: the scale of the spending, whether operating cash still covers it, whether cloud revenue is keeping up, how the gap is now financed, and how much balance-sheet strength remains. Every figure was cross-checked against at least two independent sources, company releases and earnings calls first. Where two sources disagreed (Amazon's $75.9B call figure versus the $77.3B cash-flow-statement line, for example), both are shown with their basis rather than silently merged. Figures I checked but chose not to use are listed at the bottom of this page, with the reason for each.
Sources
Every figure on this page comes from our own investigation of November 3, 2026. Each number was cross-checked against at least two sources before publication, company releases and calls first; where sources disagreed, the primary source governed, and where a primary figure could not be obtained, the number was dropped rather than estimated. Figures we computed ourselves from reported lines are labelled as our calculations. The originating video is a topic lead only — none of its figures, structure or narrative is cited or adopted anywhere on this page.
Microsoft — fiscal Q1 2027 earnings call (Oct 28, 2026), official transcript: capex including finance leases $45.2B, cash paid for PP&E $39.4B, operating cash flow $58.0B, free cash flow $18.6B, calendar-2026 capex of about $170B; cross-checked against the MarketBeat transcript summary microsoft.com · marketbeat.com
Amazon — Q3 2026 earnings call transcript (Oct 29, 2026, via Motley Fool): cash capex $75.9B in Q3, AWS revenue $46.0B up 39.3%, AWS backlog $578B, 2027 spending to rise meaningfully fool.com
Amazon — Q3 2026 results release (SEC Form 8-K Exhibit 99.1, Oct 29, 2026), as summarised by DatacenterFeed: purchases of property and equipment $77.3B in Q3 versus $35.1B a year earlier; full-year 2026 cash capex guidance of about $220B unchanged datacenterfeed.com
Oracle — Q1 fiscal 2027 results release (Sep 10, 2026): revenue $19.3B, OCI revenue $7.4B up 121%, RPO $664B, operating cash flow $23.1B, capex $28.5B, free cash flow negative $5.4B on the company's own table investor.oracle.com
FinanceFeeds (Sep 2026) — Oracle Q1 FY27 filing detail: $11.4B of customer prepayments with a significant financing component (Form 10-Q), deferred revenue doubling to $30.8B, about half of the $664B RPO expected as revenue within 36 months financefeeds.com
Allstream Insiders (Oct 2026) — Alphabet Q3 2026 detail from the earnings release: revenue $127.6B up 25%, Google Cloud $30.7B up 103%, Q3 capex $51.4B versus $24.0B a year earlier, full-year guidance held at $195-205B allstreaminsiders.com
AInvest (Oct 2026) — Meta Q3 2026 detail from the results release: revenue $64.6B up 26%, capex $34.91B, operating cash flow $32.79B, free cash flow negative $2.12B versus positive $10.63B a year earlier, 2026 capex guidance narrowed to $135-145B ainvest.com
TKAL (Oct 22, 2026) — Meta Q3 2026 filing analysis: the commitment book of $454.52B of non-cancelable contracts plus $355.73B of leases not yet commenced (Form 10-Q Note 10), cash and marketable securities $82.45B, the four-quarter free-cash-flow slide, and the year's capex-guidance path tkal.news
Reuters (Nov 2, 2026) — Alphabet kicks off a €6B bond sale (seven tranches, demand above €17B), after €9B in May and a rare 100-year sterling bond in February; the filing also showed $130.5B of leases not yet commenced and not yet on the balance sheet reuters.com
Reuters (Nov 3, 2026) — AI-related borrowing accelerates in Europe: US markets have absorbed more than $250B of technology-related debt issuance so far in 2026, up 81% year on year (LSEG data); a €40-60B pipeline is expected reuters.com
Reuters (Nov 2, 2026) — Oracle to offer $10B of floating- and fixed-rate notes, per its SEC filing reuters.com
Barron's (Oct 2026) — why Meta's $27B data-center bond is slumping: the Beignet Investor notes (6.58%, due 2049, S&P A+) trading around 92 cents for a yield of about 7.3% (Bloomberg data); Meta's plan to issue up to about $97B of equity; Meta long-term debt of about $81B against $82B of cash at Sep 30 barrons.com
S&P Global Market Intelligence data, as reported by BigGo Finance (Nov 2, 2026) and summarised by Bitveen: Microsoft, Alphabet, Amazon, Meta and Oracle issued a combined $326B of new debt in 2026, about three times the 2025 level bitveen.com
Bloomberg reporting, via CryptoBriefing (Oct 10, 2026) — nearly $500B of AI-related bonds sold market-wide in 2026 to date; the five hyperscalers issued roughly $200B or more of it (Amazon about $54B, Alphabet $31.5B, Meta $30B). A different measure from the S&P figure above; the two are not added together anywhere on this page cryptobriefing.com
Checked but not used
A combined “about $750B of 2026 AI capex” (Bloomberg estimate as relayed by secondary coverage) alongside a “about $725B” four-company figure from an analysis site — the two scopes conflict, so the page uses our own addition of the four published guides (about $720-740B) and cites the external estimates only as a range check.
Microsoft capex commentary framing the $45.2B print (below the earlier “over $50B” outlook) as cooling demand — the company's own call attributes the gap to delayed third-party capacity deliveries and repeats that demand exceeds supply. Dropped as a characterisation; the figures themselves are used.
Oracle “total debt of $125B” — appeared in a single secondary summary without a verifiable basis. Dropped; the page uses Oracle's dated, sourced $10B November 2 offering instead.
Alphabet's SpaceX-related unrealised loss given as $8.2B by one source and $8.7B by another — the two disagree, and the figure is not load-bearing for this report's question; it is not used in the text.
Amazon's quarterly free cash flow computed as about -$34.9B — derivable from the reported lines, but Amazon's own published measure is the trailing-twelve-month figure (-$42.4B); the page uses the company's measure and labels the derivation where the quarterly split is discussed.
Every figure, the six-question structure and the election framing of the originating TradingKey video — a topic radar only under this column's standing rule, and its political framing is excluded by rule. Never cited, never adopted.
Data as of November 3, 2026. Company figures are from the cited releases and calls (Microsoft fiscal Q1 2027, Alphabet, Amazon and Meta Q3 2026, Oracle fiscal Q1 2027); bond-market figures are as reported on the dates cited. This page is a fixed research record and is not updated with live data.
For research and education only. This page is our own independent investigation; the views are ours, and named institutions' figures are summaries of cited reporting. This page is not investment advice, not a recommendation, and not a promise of returns. Read the full disclaimer before acting on anything here.