Demand
Total Capex of Four AI-Exposed Platforms vs. Selected Data-Center Supplier Revenue
Quarterly companywide cash capex at Microsoft, Alphabet, Meta, and Amazon versus NVIDIA and AMD data-center revenue. The fixed baskets track two sides of AI infrastructure investment, but include non-AI activity and differ in scope, timing, accounting, customers, and suppliers.
Total Capex of Four AI-Exposed Platforms vs. Selected Data-Center Supplier Revenue
- Total capex: Microsoft + Alphabet + Meta + Amazon
- Data-center revenue: NVIDIA + AMD
In 2026-06-30, companywide capex for the fixed Microsoft, Alphabet, Meta, and Amazon basket was $165.1B, while reported NVIDIA plus AMD data-center revenue was $95.7B.
Methodology
Keep both baskets fixed for the full history. The capex basket contains Microsoft, Alphabet, Meta, and Amazon because each is a public SEC registrant making material infrastructure investments for AI while also reporting a recurring companywide cash property-and-equipment measure. The revenue basket contains NVIDIA and AMD because each reports an exact, recurring data-center revenue category in public filings. Basket selection is editorial and the result is not a comprehensive market index.
For platform capex, use the cash-flow statement line called additions to property and equipment at Microsoft and purchases of property and equipment at Alphabet, Meta, and Amazon. Use the absolute cash outflow in nominal USD. When only year-to-date amounts are reported, derive a standalone quarter by subtracting the immediately preceding cumulative amount; derive a fourth quarter as the annual amount less the nine-month amount. Retain issuer restatements when a later comparative filing changes a historical amount.
For selected supplier revenue, add NVIDIA's reported Data Center market-platform revenue and AMD's reported Data Center segment net revenue. Both include meaningful non-AI business. Do not substitute NVIDIA Compute & Networking segment revenue, AMD Instinct estimates, management forecasts, analyst estimates, or run rates.
Assign each observation to a calendar-quarter cohort. Microsoft, Alphabet, Meta, Amazon, and AMD are aligned to the calendar quarter containing their quarter end. NVIDIA uses a late-January 52/53-week fiscal year, so each NVIDIA fiscal quarter is assigned to the calendar quarter containing the majority of its days: late-April to Q1, late-July to Q2, late-October to Q3, and late-January to the preceding Q4. This improves period overlap but does not make the baskets perfectly synchronous.
Convert reported USD millions to USD billions by dividing by 1,000, sum unrounded components, and publish each aggregate to three decimal places. No inflation adjustment, currency conversion, interpolation, seasonality adjustment, private estimate, or forecast is used.
Use two independent, unstacked lines because neither aggregate is a component of the other. Do not compute an implied payback period or capex-to-revenue return: platform capex creates assets over multiple years, supplier revenue includes sales to customers outside the platform basket, platform purchases may be recorded in periods different from supplier revenue, and some capex is paid to vendors outside the revenue basket.
Frequently asked questions
Is the purple line AI capital expenditure?
No. It is the total companywide cash property-and-equipment spending of Microsoft, Alphabet, Meta, and Amazon. AI infrastructure is a major reason these companies are in the basket, but their filings do not provide a comparable quarterly AI-only capex amount, and the total includes substantial non-AI assets.
Is the green line total AI revenue?
No. It is only NVIDIA Data Center revenue plus AMD Data Center segment revenue. Both categories are highly exposed to AI infrastructure demand but also include non-AI networking, CPUs, and other data-center products, while many AI suppliers and service providers are omitted.
Why compare these measures if they are not the same transactions?
They provide two public, recurring quarterly views of the infrastructure cycle in one common unit: cash investment by major platforms and accrual revenue at two large compute suppliers. The comparison is useful for direction and scale only, not for calculating ROI, payback, industry profit, or a funding gap.
Why are Amazon and Meta included with cloud hyperscalers?
The capex basket is deliberately labeled AI-exposed platform companies rather than pure cloud hyperscalers. Amazon and Microsoft operate major public clouds; Alphabet operates Google Cloud; Meta is a large internal AI-infrastructure buyer. All four also have material businesses outside cloud and AI.
Why exclude Broadcom and TSMC from supplier revenue?
Broadcom's public AI revenue disclosures are not exact and consistently defined for every quarter in this history; some are rounded or lower bounds. Adding TSMC HPC revenue would also count foundry revenue in the same supply chain as NVIDIA and AMD product revenue. Excluding both keeps the fixed supplier basket reproducible and reduces obvious vertical double counting.
How are NVIDIA fiscal quarters aligned?
NVIDIA's quarter ending in late April is assigned to Q1, late July to Q2, late October to Q3, and late January to the preceding Q4 because most days in each NVIDIA quarter fall in that calendar quarter. The mapping is explicit and should not be mistaken for exact calendar-quarter synchronization.
How are the two quarterly baskets sourced and calculated?
The chart uses cash-flow statements for each platform's cash property-and-equipment spending and filed segment or market-platform disclosures for NVIDIA and AMD data-center revenue. It converts exact reported amounts to USD billions, sums fixed components without interpolation, and derives standalone capex quarters only from reported cumulative values.