Prices 2026-03-29 – 2026-09-09
Economics
Cloud rate cards and training-run costs; hyperscaler capital expenditure.
Cloud rate cards and training-run costs; hyperscaler capital expenditure.
Each line compares a hyperscaler’s capital expenditure as a share of revenue in 2019 with its latest complete four-quarter period. The heavier line shows combined capex divided by combined revenue for those 6 issuers.
A line is tagged rose or fell where the share moved by more than 0.5 percentage points. Changes within that range, including its endpoints, are labelled level.
An issuer appears only if both periods contain four quarters of capex and revenue data.
Each issuer uses its own reporting periods. Oracle has quarter ends outside March, June, September and December. Baseline periods end between 2019-11-30 and 2019-12-31; the latest periods end between 2025-12-31 and 2026-02-28. The table below gives the capex and revenue used in each ratio.
From 2019 to the latest complete periods, the 6 hyperscalers multiplied their capital spending by 5.5× while multiplying revenue by 2.2×. Capex as a share of revenue moved from 8.4% to 20.5%, a change of 12.2 percentage points, calculated before rounding. The largest change was at Oracle, 4.0% to 75.3%, a move of 71.3 percentage points. The smallest change was at Apple, which fell 0.7 percentage points.
| Issuer | Capex 2019 | Revenue 2019 | Capex, latest | Revenue, latest | Share, then → now |
|---|---|---|---|---|---|
| Oracle · four quarters to 2026-02-28 | $1.6B | $39.6B | $48.3B | $64.1B | 4.0% → 75.3% |
| Meta · four quarters to 2025-12-31 | $15.1B | $70.7B | $69.7B | $201.0B | 21.4% → 34.7% |
| Microsoft · four quarters to 2025-12-31 | $13.5B | $134.2B | $83.1B | $305.5B | 10.1% → 27.2% |
| Alphabet (Google) · four quarters to 2025-12-31 | $23.5B | $161.9B | $91.4B | $402.8B | 14.5% → 22.7% |
| Amazon (AWS) · four quarters to 2025-12-31 | $16.9B | $280.5B | $131.8B | $716.9B | 6.0% → 18.4% |
| Apple · four quarters to 2025-12-31 | $9.2B | $267.7B | $12.1B | $435.6B | 3.5% → 2.8% |
| All 6, summed · capex ×5.46, revenue ×2.23 | $79.9B | $954.6B | $436.4B | $2.13T | 8.4% → 20.5% |
Vintage filed quarters 2018-01-31 to 2026-02-28 · each of the 25 issuers at its own latest filed quarter781 quarterly rows · 29 without a quarterly capex figure
The series below shows quarterly changes between the periods compared above. Bars stack by the calendar quarter end each issuer reports, so Microsoft’s fourth fiscal quarter and Alphabet’s second share a bar when both close on 30 June. The second mode restates each bar as a share of that quarter’s revenue.
Quarter by quarter. One bar per calendar quarter end, holding only the issuers whose own quarter closed on that date.
Bars in one period are directly comparable only where issuers share a quarter-end calendar: 32 of 65 periods have a single filer, and no period has more than 5 of the 6. The 6 do not share a fiscal calendar, so the tall-then-short alternation counts how many of them closed a quarter on each date.
The pale band behind a bar marks a period fewer than all 6 filed in — on this axis every one of the 65, because the 6 never close a quarter on the same date.
The line sums each issuer’s own trailing four quarters, read at the last quarter that issuer had filed as of the period being plotted. Four consecutive bars span four or five months of calendar time and hold about two quarters from each issuer. The line is drawn only where all 6 have a complete window (56 of 65 periods). A window counts only when all four of its incremental-capex cells are computable; where one is not, the line breaks, because an older window substituted there would draw a movement the data does not contain.
Not computable29 of 781 rows have no incremental-capex cell: a late listing such as Arm or GlobalFoundries has no prior quarter inside the extract for its first fiscal year, and neither do the extract’s own opening quarters. They are left null, appear as gaps in the series, and drop out of the cohort sum for those periods.
| Issuer | Latest filed quarter | Capex, that quarter | Revenue, that quarter | Share of revenue |
|---|---|---|---|---|
| Hyperscalers · 6 | ||||
| Microsoft · MSFT | FY2026 Q2 · closed 2025-12-31 | $29.9B | $81.3B | 36.8% |
| Amazon (AWS) · AMZN | FY2025 Q4 · closed 2025-12-31 | $39.5B | $213.4B | 18.5% |
| Alphabet (Google) · GOOGL | FY2025 Q4 · closed 2025-12-31 | $27.9B | $113.8B | 24.5% |
| Meta · META | FY2025 Q4 · closed 2025-12-31 | $21.4B | $59.9B | 35.7% |
| Oracle · ORCL | FY2025 Q3 · closed 2026-02-28 | $18.6B | $17.2B | 108.4% |
| Apple · AAPL | FY2026 Q1 · closed 2025-12-31 | $2.4B | $143.8B | 1.7% |
| Chip designers · 7 | ||||
| Nvidia · NVDA | FY2025 Q4 · closed 2026-01-31 | $1.3B | $68.1B | 1.9% |
| AMD · AMD | FY2025 Q4 · closed 2025-12-31 | $222M | $10.3B | 2.2% |
| Intel · INTC | FY2025 Q4 · closed 2025-12-31 | $3.5B | $13.7B | 25.5% |
| Qualcomm · QCOM | FY2026 Q1 · closed 2025-12-31 | $549M | $12.3B | 4.5% |
| Broadcom · AVGO | FY2026 Q1 · closed 2026-01-31 | $250M | $19.3B | 1.3% |
| Marvell · MRVL | FY2025 Q4 · closed 2026-01-31 | $114M | $2.2B | 5.2% |
| Arm · ARM | FY2025 Q3 · closed 2025-12-31 | $179M | $1.2B | 14.4% |
| Foundries · 2 | ||||
| TSMC · TSM | FY2025 Q4 · closed 2025-12-31 | $10.5B | $30.7B | 34.2% |
| GlobalFoundries · GFS | FY2025 Q4 · closed 2025-12-31 | $208M | $1.8B | 11.4% |
| Semi equipment · 5 | ||||
| ASML · ASML | FY2025 Q4 · closed 2025-12-31 | $527M | $11.4B | 4.6% |
| Applied Materials · AMAT | FY2026 Q1 · closed 2026-01-31 | $646M | $7.0B | 9.2% |
| Lam Research · LRCX | FY2026 Q2 · closed 2025-12-31 | $261M | $5.3B | 4.9% |
| KLA · KLAC | FY2026 Q2 · closed 2025-12-31 | $106M | $3.3B | 3.2% |
| Teradyne · TER | FY2025 Q4 · closed 2025-12-31 | $63M | $1.1B | 5.8% |
| Memory · 1 | ||||
| Micron · MU | FY2026 Q2 · closed 2026-02-28 | $6.4B | $23.9B | 26.8% |
| Networking · 1 | ||||
| Arista Networks · ANET | FY2025 Q4 · closed 2025-12-31 | $37M | $2.5B | 1.5% |
| Colocation REITs · 2 | ||||
| Equinix · EQIX | FY2025 Q4 · closed 2025-12-31 | $2.0B | $2.4B | 84.3% |
| Digital Realty · DLR | FY2025 Q4 · closed 2025-12-31 | $0 | $1.7B | 0.0% |
| AI-DC power infrastructure · 1 | ||||
| Vertiv · VRT | FY2025 Q4 · closed 2025-12-31 | $93M | $2.9B | 3.2% |
Quarterly corporate fundamentals from a licensed financial-fundamentals database, held under academic subscription and not redistributed. The underlying primary documents are the 10-Q and 10-K filings each issuer makes with the SEC; the database re-serves them under standardised column codes (capxy, revtq). The numbers are the XBRL figures the issuer filed, read one subscription layer away from the filing, which puts the page on the second of the four authority tiers Scrutica grades a source on: tier 1 the primary filing or measurement, tier 2 a database re-serving one, tier 3 press and analyst reporting, tier 4 an estimate. What the licence permits.
capxy is year-to-date capital expenditure and resets to zero at the start of each fiscal year, so a naïve quarter-over-quarter difference produces a large negative number at every fiscal-year boundary. The decomposition is Q1 = capxy, and Q2 through Q4 = capxy[fqtr] − capxy[fqtr−1] within one fiscal year. Where the prior-quarter row falls outside the extract, the incremental cell stays null.
Four quarterly incrementals sum to within ±$2M of the source’s own annual capx across 180 of 180 closed fiscal years. Fourteen partial fiscal years carrying one or more null quarters, and eleven still open, are left unreconciled.
Fiscal labels are not comparable across issuers: Apple’s fiscal year ends in late September and Oracle’s quarters close in February, May, August and November. Apple’s quarters nonetheless close on the calendar ends: at Apple the mismatch is in the label, at Oracle in the date. The series stacks on the calendar quarter end the source reports; the same fact governs the figure above, where a trailing-four-quarter window is built per issuer over that issuer’s own consecutive quarters, because four consecutive rows of the calendar-keyed series would collect roughly two quarters from each issuer.
The 25 issuers are fixed by the extract pipeline: the AI-infrastructure-relevant set spanning hyperscalers, chip designers, foundries, semiconductor equipment, memory, networking, colocation REITs and data-centre power infrastructure. The figure and the series draw the 6 hyperscalers; the register has all 25. Membership says an issuer builds or supplies AI infrastructure; what fraction of its capital budget went to that is a question no column in the source answers.
One cross-reference table in the database maps the data vendor’s own key for a company to Scrutica’s, so adding an issuer is a single row and the next load of fundamentals resolves through that same table.