AI capex roughly doubled in a year: the four big hyperscalers alone reported about $360B in 2025, up from about $212B in 2024

Asked:
“Compare AI spending of top companies in this area, connect how budgets are ditributed across al sub-contactors, show dinamics 2022-2033 (factual and forecast)”

The rows cover seven companies — Microsoft, Alphabet, Amazon, Meta, Oracle, Apple and NVIDIA — with 66 sourced company-year capex observations for 2022–2026, 138 publicly surfaced buyer–supplier relationships (not a complete universe of subcontractors), and 144 external market forecasts through 2033. Audited AI-only budgets do not exist: the comparable basis shown is company-wide capital expenditure, which for the hyperscalers is increasingly dominated by AI and data-center build-out. Coverage by year is partial and accounting definitions differ; fiscal years differ too (Microsoft ends June, Oracle May, Apple September, NVIDIA January).

Reported capital expenditure, 2022–2026 actuals

One dot per company-year, best-sourced figure where sources conflict. Values are US$ billions of company-wide capex — not AI-only spending. Apple and NVIDIA sit near the floor because their business models differ: Apple leans on partners and device-side compute, NVIDIA is a fabless supplier whose own capex is tiny next to customers' purchases of its systems. Hover any dot for its measure definition and source.

Where the budgets flow: 138 surfaced supplier relationships

Ribbon width is the count of surfaced public relationships by category, not dollars — most contract values are confidential, and only 41 of 138 rows carry a disclosed or estimated value, so no spend widths are invented and supplier amounts are never forced to sum to buyer capex. The one large recurring estimate: Alphabet spends about $8B a year with Broadcom on TPU silicon co-design (an analyst estimate, not an audited contract). Hover a ribbon for the actual suppliers behind it; the table below holds every row.

On a consistently stated 2025 reported-capex basis: Amazon $131.8B, Alphabet $91.4B, Meta $72.2B, Microsoft $64.6B audited additions to property and equipment (other definitions reach ~$88B), Oracle $21.2B, Apple $12.7B, NVIDIA $3.24B. savrn.com
Oracle is the steepest curve in relative terms: reported capex rose from $6.9B (FY2024) to $21.2B (FY2025) to $55.7B (FY2026) — an eightfold rise in two fiscal years. savrn.com
Concentration risk is visible in the network: NVIDIA appears as supplier in 20 relationship rows, Broadcom in 14 and TSMC in 9 — accelerators, custom silicon, foundry and advanced packaging all route through a handful of firms. maktinta.com
McKinsey's accelerated-demand scenario puts cumulative global AI-related data-center capex at $7.9T through 2030 — a cumulative capex figure, not comparable to the annual market revenues in the outlook panel below. mckinsey.com

Outlook to 2033: market scenarios, not company forecasts

Reliable company-by-company forecasts do not extend to 2033, so the actual lines above stop at reported years. These are external market-research forecasts of distinct, overlapping market scopes — they are not additive and not comparable to each other or to company capex. Each panel shows the forecaster's own base and endpoint.

All surfaced supplier relationships

BuyerSupplierCategoryValue $BRelationshipSource

Method: 66 company-year spending observations (2022–2026, US$ billions, mixed definitions: cash capex, audited additions/purchases of property and equipment, finance leases) deduplicated by keeping the figure with the most independent sources per company-year; 138 publicly surfaced buyer–supplier relationships grouped into 12 categories (only 41 carry a disclosed or estimated value; ribbon widths are relationship counts); 144 external forecast observations, of which 6 well-scoped 2032–2033 endpoints are shown. Fiscal years differ by company and are plotted as labeled. Capex is company-wide, not AI-only; AI shares are generally undisclosed. Data compiled from public filings, investor-relations pages and analyst estimates as of 2026. Duplicate and non-comparable forecast scopes were cut for space.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.

Keenable SELECTAsk your own question
Made with Keenable SELECT