Big-five tech capex exceeds $5 trillion in a year

Status and baseline (computed from SEC filings)

I pulled cash capex from SEC XBRL (data.sec.gov/api/xbrl/companyconcept, tag PaymentsToAcquirePropertyPlantAndEquipment; Amazon uses PaymentsToAcquireProductiveAssets). For Microsoft I converted fiscal years to calendar years with the question's formula.

• 2025: Alphabet $91.4B, Amazon $131.8B, Meta $69.7B, Microsoft $83.1B. Oracle is about $40B (rough; only Mar–Nov 2025, $29.6B, is fully inside the calendar year). The total is about $416B. This matches the five-company series in Van Nieuwerburgh's Brookings paper (Sept 2026): $153.8B in 2023, $240.4B in 2024, $415.8B in 2025 and $800.5B estimated for 2026. That is growth of +56%, +73% and about +92%.
• First half of 2026, four companies: $294.8B. Oracle spent $16.5B in Mar–May 2026 and $28.5B in Jun–Aug 2026.
• 2026 guidance:
  • Alphabet: $195–205B
  • Amazon: about $220B
  • Meta: $130–145B, including finance-lease principal
  • Microsoft: about $175B for calendar 2026, including finance leases (cash spending is lower)
  • Oracle: $90–95B for its fiscal year ending May 2027
• Forecasts for later years:
  • Goldman: $1.2T in 2027 against a $1.1T consensus, then $1.4T in 2028, with a baseline of $1.6T a year of AI capex by 2031.
  • S&P Global Ratings: about $1.12T for these five in 2027, then flattening in 2028.
  • TrendForce: nine large cloud firms grow 60% in 2027.
  • Nvidia's chief executive still expects $3–4T of global AI infrastructure spending in 2030. Global means all buyers, not just these five.
  • Wall Street consensus has the five companies' operating cash flow at about $2T in 2030 (Apollo, cited by Yahoo).

What YES requires

$5T is 6.2 times the 2026 estimate. Each calendar year can only be established in the January–February filings after it ends:

• 2026 resolves in Q1 2027.
• 2027 resolves in Q1 2028.
• 2028 resolves in Q1 2029.
• 2029 resolves in Q1 2030.
• 2030 resolves in Q1 2031.
• Calendar 2031 cannot be reported before the end of the horizon.

Required growth rates:

• To pass $5T in 2030: about 58% a year for four straight years (about 57–61% a year over 2028–30 if 2027 lands at $1.2–1.3T).
• To pass $5T in 2029: about 84% a year.
• To pass $5T in 2028: about 150% a year.

Reference class and estimate

History of capex booms (railroads, telecom/fiber, shale): growth slows, then turns down. Nothing at trillion-dollar scale has sustained more than 55% a year for seven years in peacetime. $5T for these five alone would be:

• more than twice Nvidia's bullish global forecast;
• about 2.5 times consensus operating cash flow for 2030, implying roughly $3T a year of outside financing;
• by my rough arithmetic, roughly 100 GW a year of new data-center capacity, using about $41–50B per GW.

Physical and financial constraints are already binding:

• ASML says AI demand exceeds its chip-tool supply through 2028.
• Grid and turbine shortages, plus Oracle's force-majeure notice over power delays.
• Free cash flow is already negative at several of the companies.
• More building is moving to leases and joint ventures, which do not appear in the cash capex line this question counts.
• Wells Fargo reports most investors expect a "meaningful" slowdown in 2028.

A lognormal check with a 2030 median of about $1.7–2.0T and a right-tail log standard deviation of about 0.43–0.45 gives roughly 0.7–1.8% above $5T. Hard supply limits should make the real right tail thinner than a lognormal.

Probability by year:

• 2029: about 0.15%
• 2030: about 0.85%
• 2026–2028: tiny residuals only

Strongest case against my estimate

Growth has sped up three years running. Two-year-ahead forecasts missed 2026 by about 2.6 times (I/O Fund). Nvidia guides to +70% next year, and component inflation raises nominal spending. If an AI takeoff kept growth at 60% or more a year, $5T in 2030 would be reachable. I give this some weight, which is why I hold about 1% rather than near zero. But the most bullish supplier forecasts are still far below the threshold.

Sources

• SEC XBRL API (data.sec.gov), accessed 2026-10-17
• Brookings paper by Van Nieuwerburgh, "Financing the AI Buildout" (Sept 2026): brookings.edu/wp-content/uploads/2026/09/4c_Van-Nieuwerburgh.pdf
• Goldman Sachs, 2026-09: goldmansachs.com/insights/articles/can-ai-investment-drive-s-and-p-500-earnings-even-higher
• Bloomberg, 2026-09-25
• Motley Fool on S&P Global's projections, 2026-09-06
• TrendForce, 2026-10-14
• CNBC on Microsoft (2026-07-29), Amazon (2026-07-30) and TSMC (2026-10-15)
• Meta second-quarter 2026 release
• Business Insider on Oracle, 2026-09
• Fortune, 2026-10-07
• Yahoo / Apollo, 2026-09
• Motley Fool on Nvidia's chief executive, 2026-09-18

Question. By the last day of each quarter, has the combined capital expenditure of Alphabet, Amazon, Meta, Microsoft, and Oracle exceeded $5 trillion for a calendar year?

Definitions.

• Companies: Alphabet Inc., Amazon.com, Inc., Meta Platforms, Inc., Microsoft Corporation, and Oracle Corporation, each on a consolidated worldwide basis. Include all businesses, not just AI or cloud infrastructure. Name changes and successor holding companies count. Acquired businesses enter the perimeter when consolidated; divested businesses leave when deconsolidated. If two named companies merge, count their consolidated expenditure only once. An unrelated acquirer's other businesses do not enter the perimeter.
• Capital expenditure: gross cash purchases or additions to property and equipment reported in the investing section of the consolidated statement of cash flows, expressed as a positive amount. The reference lines are “Purchases of property and equipment” for Alphabet, Amazon, and Meta; “Additions to property and equipment” for Microsoft; and “Capital expenditures” for Oracle. Successor line labels count only for the same accounting category. Include capitalized software or construction payments if included in these lines. Do not subtract separately reported proceeds from asset sales or incentives.
• Exclude acquisitions of businesses, securities purchases, separately reported intangible-asset purchases, operating expenses, lease payments classified outside those investing lines, noncash finance-lease additions, and commitments not yet paid. Company headline capex measures that add lease principal payments or noncash lease additions must be reconciled to this definition before use.
• $5 trillion: $5,000,000,000,000 in nominal US dollars, without inflation adjustment. Use the most precise figures published in the qualifying financial statements, rather than rounded headline totals. Exactly $5 trillion does not qualify.
• Calendar year: January 1–December 31 of a single year. Apply the Observation period convention; the first eligible year is 2026. Fiscal years or trailing twelve-month periods ending on another date cannot substitute.

Calendar-year calculation. Use calendar-year statements where available. Otherwise derive nonoverlapping calendar-period spending by subtracting cumulative cash-flow figures or combining disclosed subperiods. For Microsoft, calendar-year expenditure can be calculated as the fiscal year ending in June of that year, less its first six months, plus the six months ending that December. Oracle's May fiscal year-end and February/May/August/November quarter-ends do not align with calendar quarters: do not assign an entire fiscal quarter to the calendar year containing its end, or prorate it by months or days. Use an explicit calendar-year disclosure or separately disclosed expenditure for the relevant dates. If exact calendar allocation is unavailable, a documented lower bound may establish YES: sum only nonoverlapping expenditure unambiguously incurred within the calendar year and treat undisclosed portions as zero. For example, Oracle's disclosed March–November spending lies wholly inside that calendar year. This lower-bound route may be used only after the calendar year has ended.

Resolves YES if qualifying publications establish that the sum for one eligible, completed calendar year is strictly greater than $5 trillion, either through an exact total or through the lower-bound method above. The qualifying publication is the last document needed to establish that result. A formal restatement can first establish YES. Before YES, use the latest restated figures available at the horizon; after YES, apply the Resolution window convention.

Resolves NO if no qualifying evidence establishes the threshold by the horizon. Forecasts, capex guidance, announced investment programs, annualized quarterly run rates, totals spanning multiple years, or a calendar-year estimate dependent on prorating Oracle's fiscal quarters do not qualify. Missing disclosures do not authorize estimated spending.

Resolving source. Company-authored financial statements and accompanying notes in SEC Forms 10-K, 10-Q, and amendments, and earnings releases or supplemental financial schedules published on the five companies' investor-relations sites: abc.xyz/investor, ir.aboutamazon.com, investor.atmeta.com, microsoft.com/investor, and investor.oracle.com. Company-authored supplemental disclosures may establish calendar allocation. SEC filings take precedence over conflicting releases covering the same period, with formal restatements taking precedence over original figures. These are the permitted company-authored sources under Public confirmation (A2); third-party analyst estimates cannot resolve this question. Apply Resolution date (A3), the Resolution window, and the Observation period conventions.