Panic of 1873: The Railroad Bubble That Shut Wall Street, and What Big Tech's Q3 2026 Filings Say About AI Capex
The Panic of 1873 began when Jay Cooke & Co. suspended payments on September 18, 1873; the New York Stock Exchange closed for ten days and at least 100 banks failed. Beside that railroad bubble, a dated reference on the 2026 AI buildout: eight quarters of capex for Microsoft, Alphabet, Amazon and Meta, each company's stated capex outlook, cloud growth in the September quarter, and how the spending is financed, every figure from the company's own 8-K, 10-Q or earnings call.
For Halloween, a real Wall Street horror story and a dated reference page to go with it. The Panic of 1873 began on September 18, 1873, when Jay Cooke & Co., the banking house financing the Northern Pacific Railway, suspended payments; the New York Stock Exchange closed on September 20 and stayed shut for ten days, and at least 100 banks failed nationwide. The boom behind it was railroads: real, useful infrastructure, financed through stocks and bonds years before the paying traffic arrived.
This week Microsoft, Alphabet, Amazon and Meta filed their results for the September 2026 quarter, so the same questions can be asked of the AI buildout with numbers rather than analogies. This page pairs the two: the 1873 chronology from the institutional histories, then eight quarters of capex for each of the four companies, each company's stated outlook, cloud growth in the quarter, and how the spending is financed, every figure from the company's own 8-K, 10-Q or earnings call. It is explicitly not a crash forecast. The 1873 story is the two-minute version in the Short below; the tables are what the Short could not fit.
What caused the Panic of 1873?
The setting is the Gilded Age, the decades after the Civil War. Wartime bond sales had introduced many small investors to securities for the first time, and railroad promoters found buyers for their stocks and bonds among them (Harvard's Baker Library). Stocks meant ownership; bonds were loans the companies promised to repay with interest. Construction needed the cash immediately, while the passengers, freight and new towns that would pay for it could arrive years later. Promoters expected settlement and business to follow the track (Library of Congress).
Early success drew more money, and building outran demand: new projects added capacity faster than traffic could fill it, which pushed down the returns on the investment (Federal Reserve History). The trigger came from abroad. Financial turmoil in Europe led investors there to sell American securities, railroad bonds in particular, and fresh financing for unfinished lines became hard to find (U.S. Treasury). Jay Cooke & Co. was deeply committed to the Northern Pacific; construction problems and a funding gap put the bank under pressure, and on September 18, 1873 it suspended payments (New York Fed). Investors sold, depositors ran on banks, and on September 20 the New York Stock Exchange closed. Trading stayed suspended for ten days. At least 100 banks failed across the country, and business failures and unemployment followed.
Three distinctions the histories insist on. The 1873 episode was a railroad financing boom, with stocks, construction bonds and fragile banks together, rather than a story of share prices alone. It is a different event from Britain's Railway Mania of the 1840s. And the railroads themselves mostly survived; it was the securities sold to build them, and the banks that sold them, that failed. This page quotes no market-return percentage for 1873, because no documented price series with named constituents and dates was used.
Railroad bubble vs AI bubble: what rhymes and what does not
What rhymes is the shape. A transformative technology; construction that must be paid for before the revenue exists; capital raised from investors on the expectation that demand will follow; early success drawing more capital in. What does not rhyme is in the filings below, and it comes down to two facts. First, the paying demand is arriving now rather than later: the three largest clouds grew 39% to 103% in the September quarter, and each company reports a contracted backlog in the hundreds of billions. Second, the builders are among the most profitable companies in the world, and in the quarter their operations generated between $32.8 billion and $58.0 billion of cash each. In 1873 the weak point was the lender, one bank carrying one railway's bonds. In 2026 the borrowers are rated Aaa and AAA (Microsoft's October 29 term sheet states both) and the debt is spread across public bond markets, leases and, since this summer, new equity.
None of that settles whether the returns will justify the spending. It changes which questions matter, which is where the Short ends and the tables begin.
Big tech capex 2026 by quarter: Microsoft, Alphabet, Amazon and Meta
Eight calendar quarters, each figure from that quarter's 8-K press release. The four companies define capex differently, and the table keeps each company's own line: Microsoft's cash "additions to property and equipment" from its cash flow statement, with the "including finance leases" figure its CFO gives on each call in parentheses; Alphabet's and Amazon's "purchases of property and equipment" (Amazon's is gross, before proceeds from sales and incentives); and Meta's "capital expenditures, including principal payments on finance leases", the line Meta itself headlines.
| Quarter ended | Microsoft, cash capex (incl. finance leases) | Alphabet | Amazon, gross | Meta, incl. finance lease principal |
|---|---|---|---|---|
| Dec 31, 2024 | $15.8B ($22.6B) | $14.3B | $27.8B | $14.8B |
| Mar 31, 2025 | $16.7B ($21.4B) | $17.2B | $25.0B | $13.7B |
| Jun 30, 2025 | $17.1B ($24.2B) | $22.4B | $32.2B | $17.0B |
| Sep 30, 2025 | $19.4B ($34.9B) | $24.0B | $35.1B | $19.4B |
| Dec 31, 2025 | $29.9B ($37.5B) | $27.9B | $39.5B | $22.1B |
| Mar 31, 2026 | $30.9B ($31.9B) | $35.7B | $44.2B | $19.8B |
| Jun 30, 2026 | $35.8B ($41B) | $44.9B | $54.2B | $31.1B |
| Sep 30, 2026 | $39.4B ($45.2B) | $51.4B | $77.3B | $34.9B |
Over the eight quarters the filed figure rose 2.5 times at Microsoft, 3.6 times at Alphabet, 2.8 times at Amazon and 2.4 times at Meta. The September 2026 column adds to about $203 billion against about $73 billion for the December 2024 column; that is arithmetic across four definitions, so read it as an order of magnitude. Amazon's September quarter is the largest single step in the table, $54.2 billion to $77.3 billion, and its cash capex net of proceeds and incentives was $75.9 billion. Microsoft's measure is the one that moved in both directions: the finance-lease figure fell from $37.5 billion to $31.9 billion in the March quarter before rising again, and the CFO said on October 28 that $7.4 billion of the September quarter's $45.2 billion was finance leases, recognized in capex at the full principal value when a lease commences. Meta's quarterly figures sum to $72.2 billion for 2025, the figure its fourth-quarter release printed.
Hyperscaler capex guidance: what each company said this week
| Company | September quarter, the company's own measure | Calendar 2026 outlook | 2027 | Where it was said |
|---|---|---|---|---|
| Microsoft | $45.2 billion including finance leases; cash paid for property and equipment $39.4 billion; "roughly three quarters" of it short-lived assets, primarily CPUs and GPUs | December quarter "over $50 billion"; "for the calendar year, spend should be approximately $170 billion due to the 3P capacity delays"; was "approximately $175 billion" in July | Capex to "accelerate" in the second half of fiscal 2027 (the first half of calendar 2027) | October 28 call, company-hosted transcript |
| Alphabet | $51.4 billion, "the vast majority" in technical infrastructure; about 65% servers and networking equipment (six-year useful lives), 35% data centers (seven to 40 years) | "$195-$205 billion remain unchanged" | "We continue to expect our CapEx to increase significantly in 2027", details on the fourth-quarter call | October 28 call, company-hosted transcript |
| Amazon | $75.9 billion of cash capex (purchases of $77.3 billion less $1.4 billion of proceeds and incentives), "the majority for AWS" | "We continue to expect 2026 cash capital expenditures to be approximately $220 billion" | "Expect to spend meaningfully more in 2027", per a published transcript of the October 29 call, not a company-posted document | October 29 press release; call per a published transcript |
| Meta | $34.91 billion including principal payments on finance leases | "$135-145 billion, narrowed from our prior outlook of $130-145 billion" | "Still developing our budget and capacity plans for next year, but we expect to continue investing aggressively" | October 21 press release, CFO outlook commentary |
On four definitions, the four calendar-2026 outlooks add to roughly $720 billion to $740 billion. Microsoft is the one of the four whose stated figure went down between July and October, and it attributed the change to delayed third-party capacity deliveries that shifted into the second half of its fiscal year, not to a smaller plan. Alphabet's range, against $132.0 billion spent in the first nine months per its 10-Q, implies roughly $63 billion to $73 billion for the December quarter; Meta's, against $85.8 billion in nine months, implies roughly $49 billion to $59 billion. Both are arithmetic on the company's range, not company figures.
Will paying demand keep pace with construction? Cloud growth in the September quarter
The first of the Short's three questions. In 1873 the freight arrived years after the track. In the September 2026 quarter, each company's own release reports the revenue line that pays for its data centers, and each reports contracted work not yet recognized.
| Company | Line | Q3 2026 | Q3 2025 | Growth | Contracted, not yet recognized |
|---|---|---|---|---|---|
| Microsoft | Azure revenue (disclosed in dollars under the new segment reporting Microsoft adopted this quarter) | $33.2B | $22.4B | +48% (48% in constant currency) | Commercial remaining performance obligation $677 billion, up 73% |
| Microsoft | Microsoft Cloud revenue | $63.9B | +30% (31% in constant currency) | ||
| Alphabet | Google Cloud revenue | $30.7B | $15.2B | +103% | Revenue backlog $567.7 billion, of which $560.8 billion is Google Cloud; "just over 55%" expected as revenue over the next 24 months (10-Q) |
| Alphabet | Google Cloud operating income | $11.1B | $3.6B | ||
| Amazon | AWS net sales | $46.0B | $33.0B | +39% | $578 billion backlog, per a published transcript of the call (not in the release) |
| Amazon | AWS operating income | $17.8B | $11.4B | ||
| Meta | Revenue (Meta sells advertising, not cloud capacity) | $64.6B | $51.2B | +26% | n/a |
AWS has now accelerated for six straight quarters, from 16.9% in the March 2025 quarter to 39.3%, which Amazon's release calls its fastest growth in nearly five years; its CEO's quote in the release projects "revenue growth rates of over 40% both in 2027 and in 2028". Microsoft guided Azure to 49% to 50% constant-currency growth for the December quarter and said "customer demand continues to exceed available supply". Alphabet's Google Cloud doubled, with TPU system sales "a meaningful contributor"; its CFO also said on the call that the infrastructure "will continue to weigh on the P&L in the form of higher depreciation expenses and related data centers' operation costs such as energy". Meta's operating income of $13.0 billion includes a $9.89 billion legal charge; the release says it would have been $22.9 billion without it.
A backlog is a contract, not cash, and a growth rate says nothing about the price paid for it. What the table does show is the thing 1873 lacked: the customers are already paying, at rates that are rising rather than falling.
Who carries the debt? How the AI buildout is financed
The second question. Each line is from the company's September-quarter cash flow statement and balance sheet, with the post-quarter capital raises from the 8-Ks and term sheets filed since.
| Company | Operating cash flow, Q3 | Capex, Q3 (filed) | Free cash flow | Debt raised | Long-term debt | Equity raised | Cash and securities |
|---|---|---|---|---|---|---|---|
| Microsoft | $58.0B | $39.4B cash; $7.4B of finance leases on top | $18.6B (CFO) | None in the quarter; repaid $4.85B. On October 29 it priced $25.0 billion of notes in ten tranches, 2028 to 2066, two floating and fixed coupons of 4.850% to 6.350%, settling November 4 | $31.1B plus $4.4B current, at Sep 30 (vs $31.1B and $9.2B at Jun 30) | None | $64.5B |
| Alphabet | $42.3B | $51.4B | ($9.1B); $19.7B trailing twelve months | $29.0B net in the quarter ($28.7B of senior unsecured notes per the release); $85.2B in the nine months. The 10-Q: $45.0B of dollar notes and $35.8B of foreign-currency notes issued during 2026 | $125.1B at Sep 30 (vs $46.5B at Dec 31, 2025) | $12.4B of common stock in the quarter through its ATM program ($4.3B Class A, $8.1B Class C; $27.6B remains available); $42.9B of common and $19.1B of mandatory convertible preferred in the nine months. No buybacks in 2026 (vs $39.8B in the first nine months of 2025) | $265.1B, including $90.4B of marketable equity securities |
| Amazon | $42.4B; $168.3B trailing twelve months | $77.3B gross, $75.9B net of proceeds and incentives; $210.7B net trailing twelve months | ($42.4B) trailing twelve months, vs $14.8B a year earlier, which the release attributes to a $94.8B rise in net capex | $48.1B of long-term debt in the quarter; $115.1B in the nine months | $173.1B at Sep 30 (vs $65.6B at Dec 31, 2025); long-term lease liabilities $98.8B | None | $114.2B |
| Meta | $32.8B | $34.9B including $1.0B of finance lease principal | ($2.1B), its first negative quarter in the eight | None in the quarter; $25.0B of notes in May 2026 ($24.9B net) after $30.0B in November 2025 | $80.9B at Sep 30 (vs $58.7B at Dec 31, 2025) | On October 29: 250 million depositary shares of 6.50% mandatory convertible preferred, each one-twentieth of a $1,000 share ($12.5B at liquidation preference), plus a forward sale of 40 million Class A shares at an initial price of $622.91875 (about $24.9B if fully settled, by arithmetic). No buybacks in 2026 (vs $26.2B in the first nine months of 2025) | $82.45B |
Three readings. In the September quarter, operating cash flow covered cash capex with room to spare at Microsoft and fell short at Alphabet, Amazon and Meta, by about $9 billion, $33 billion (against Amazon's net cash capex) and $2 billion. The gap is being closed with bonds, in the nine months $85.2 billion at Alphabet and $115.1 billion at Amazon, and this season with equity: Alphabet's at-the-market stock sales and mandatory convertible preferred, Meta's preferred and forward sale. Microsoft was the holdout in the quarter, repaying debt rather than issuing it; its CFO said on October 28 that leasing had been the financing vehicle and that the company was "fortunate to have access in the debt markets" that it had not yet tapped, and the $25 billion term sheet followed the next day.
The commitments off the balance sheet are larger than the debt on it. Alphabet's 10-Q lists $130.5 billion of future payments for leases not yet commenced, primarily data centers and energy infrastructure, commencing between 2026 and 2031; $75.7 billion of backstop obligations for data centers, including lease payments and energy costs; and $924.4 billion of material purchase commitments, of which $802.1 billion is multi-year commitments for technical infrastructure and inventory, energy take-or-pay contracts and content licenses, the significant majority to be fulfilled through 2031. Amazon's CEO said on the call, per the published transcript, that "most of our capacity is either owned or long leased" and "we can fund this plan". None of the four repurchased stock in the quarter, and both Alphabet and Meta stopped buybacks entirely in 2026 after $39.8 billion and $26.2 billion respectively in the first nine months of 2025. In 1873 the question "who carries the debt" had one answer, a banking house; in 2026 it has several, and the public bond market and the companies' own shareholders are now among them.
Will future profits justify today's price? Three questions, not a forecast
The third question is the one this page cannot answer from filings, because the price is set in the market and the profits are in the future. What the filings do give is the shape of the bet: Alphabet's servers depreciate over six years and its data centers over seven to 40; Microsoft says roughly three quarters of its quarter's capex went to short-lived assets, CPUs and GPUs; the free cash flow lines at Alphabet, Amazon and Meta are negative on the quarter or the trailing year while the backlogs are counted in hundreds of billions. Whether the second decade of these assets earns a return depends on demand holding up after the current contracts are recognized, and no table above can see that far.
So the Short's framework, with the numbers behind each question:
- Will paying demand keep pace with construction? The growth rates and backlogs in the cloud table, read each quarter against the capex table. The two series have risen together for eight quarters; the question is whether they keep doing so.
- Who carries the debt? Long-term debt rose 2.7 times at Alphabet and 2.6 times at Amazon in nine months, Meta's rose by $22 billion, and Microsoft joined the issuers this week. The leases not yet commenced and the backstop obligations sit beside the debt, not inside it.
- Will future profits justify today's price? Depreciation is already weighing on margins, in Alphabet's CFO's words, and useful lives are the assumption that decides how fast.
The businesses and the financing differ from 1873, so history is not a crash forecast. It gives the questions, and this season's filings give the first answers. Real progress can still make a scary investment, and it can also make a sound one; the difference is in the numbers above, not in the analogy.
Related
Two earlier Shorts on the same channel cover the adjacent ground without repeating it here: Is AI a Bubble? What Big Tech's Own Filings Show and AI Energy Bottleneck Explained: Is Power the Real Limit?. The pre-report preview for these three cloud companies, with consensus, each company's guide and the eight-quarter cloud series through June, is Big Tech Earnings This Week. The other side of the capex, the AI lab buying the compute, is in Anthropic's IPO prospectus, explained.
Where the numbers come from
The 1873 chronology is from Federal Reserve History, the Library of Congress, the New York Fed, the U.S. Treasury and Harvard's Baker Library. The September 2026 quarter is from each company's 8-K press release on EDGAR: Microsoft (October 28), Alphabet (October 28), Amazon (October 29) and Meta (October 21); the seven earlier quarters are from each quarter's own 8-K press release. Alphabet's backlog, 2026 note issuance and off-balance-sheet commitments are from its third-quarter 10-Q. Microsoft's capex outlook and finance-lease figures are from the company-hosted transcript of its October 28 call, and its $25 billion note offering from the pricing term sheet filed with the SEC. Alphabet's outlook is from the company-hosted transcript of its October 28 call. Amazon posts no transcript, so its 2027 and backlog remarks are from a published transcript and are marked as such. Meta's October 29 capital raise is from its 8-K of that date. Ratios and sums across companies are arithmetic on those figures and are labeled where they appear.
Each company's earnings history, the KPI series behind these tables and the measured reaction to every report are on SentiSense's earnings pages and the earnings analysis endpoint of the API; nothing on this page needs a subscription to check, because it is all in the filings.
See this season's earnings calendar and each report's reaction, live →
Filed figures are as printed in each company's SEC filing and are rounded to one decimal in billions. Capex definitions differ by company and are stated with each table. Company outlooks were read from the documents linked above on October 31, 2026, and can change. The historical account is drawn from the institutional sources linked above and quotes no market-return figure. This article is educational and is not investment advice.