ai · Exposure audit

Follow the AI dollar: who pays, who gets paid, who keeps the profit

Meta's infrastructure bill consumed 97.5% of operating cash in its June quarter. Across the supply chain, company accounts reveal very different ways to own the AI buildout.

Published 2026-10-12 · 7 minute read
MacroShed Research · AI-assisted research and writing

The bill arrives before the payoff

Meta generated $31.862 billion of operating cash in the quarter ended June 30, 2026. Purchases of property and equipment consumed $30.116 billion; finance-lease principal absorbed another $962 million. That left $784 million of company-defined free cash flow. Infrastructure spending absorbed 97.5% of operating cash, calculated from the unrounded figures.

That is the tension at the center of the AI buildout. Buyers commit capital to capacity they expect to monetize over time. Equipment suppliers can earn revenue and profit as they fulfill those orders. Shareholders can own either side, but the cash-flow burden and the route to a return differ.

Our investment question: among selected public businesses building and supplying this infrastructure, where can investors already observe economic results? The answer starts with the accounts of the business you would actually own.

Source: Meta Q2 2026 results and cash-flow reconciliation

Read the buyer's bill carefully

Microsoft disclosed $41 billion of capital expenditures for its June-ended fiscal fourth quarter, alongside $35.8 billion of cash paid for property, plant and equipment and $5.6 billion of finance leases. Management described roughly two-thirds of capex as short-lived assets, mainly CPUs and GPUs supporting AI and non-AI workloads. The differently defined figures should not be forced into an exact reconciliation.

Amazon disclosed $53.1 billion of cash capital expenditures for the same calendar quarter. Its filing describes technology infrastructure, primarily supporting AWS growth, alongside fulfillment capacity. Equity investments in model companies are separate investment activity.

A buyer's capex announcement therefore answers a question about commitments to infrastructure. It provides neither an AI-only spending total nor a named supplier's receipt. Our comparison keeps the reported measures separate and makes no aggregate 'AI capex' claim.

Source: Microsoft FY2026 Q4 earnings callSource: Amazon Q2 2026 Form 10-Q

The suppliers already have an income statement

Nvidia reported $96.221 billion of revenue and $63.734 billion of GAAP operating income for the quarter ended July 26. Broadcom reported $29.591 billion and $15.955 billion, respectively, for the quarter ended August 2. Vertiv's June-ended quarter showed $3.2743 billion of sales and $637.9 million of operating profit.

MacroShed's original comparison divides each company's GAAP operating income by its own revenue: Nvidia 66.2%; Broadcom 53.9%; Vertiv 19.5%. These are whole-company operating margins for the individually labeled historical quarters. The periods differ, and the businesses have different product mixes, costs and acquisition histories.

The comparison establishes that these selected suppliers already report operating profits. It cannot isolate the profitability of AI, trace a particular hyperscaler's dollar or rank the stocks by expected return. Broadcom's software business is included in its denominator and profit. Vertiv's results include general-compute demand.

Source: Nvidia Q2 FY2027 financial resultsSource: Broadcom Q3 FY2026 financial resultsSource: Vertiv Q2 2026 financial results

Three ways to own the buildout

Compute and networking: NVDA and AVGO participate through hardware and related systems. Broadcom separately identified $16.7 billion of AI semiconductor revenue in its August-ended quarter. That equals about 56.4% of total revenue, a calculation using management's category. The category does not have a separately disclosed GAAP operating margin in the reviewed results.

Power and thermal infrastructure: VRT and ETN connect investors to the physical equipment around computing. Eaton reported $8.5 billion of June-quarter sales, including $4.0 billion in Electrical Americas. It cited data centers as a growth driver alongside other markets. Vertiv described demand from both AI and general compute. A precise AI revenue percentage for either company is outside this report's evidence.

Cloud and platform owners: MSFT, AMZN and META own businesses that can monetize installed capacity and services over time. They also bear the construction, utilization and reinvestment risks. The long-run winner could be a buyer that turns expensive infrastructure into durable customer demand.

The resulting exposure map is a set of business roles. Our judgment is that supplier accounts offer a more immediate view of buildout earnings, while buyer accounts expose the capital burden and eventual monetization. Both belong in an investment comparison.

Source: Broadcom Q3 FY2026 financial resultsSource: Eaton Q2 2026 financial resultsSource: Vertiv Q2 2026 financial results

A customer count can conceal concentration

Nvidia's filing says one direct customer represented 16% of quarterly revenue. Direct purchasers can include distributors, system builders and cloud providers; that percentage does not identify one ultimate user's share of AI demand.

Broadcom disclosed that one distributor accounted for 50% of quarterly revenue and estimated that its top five end customers represented about 55% across channels. Those are different views of concentration and must not be added together.

For investors, this creates a concrete monitoring task: examine whether a supplier's growth depends on a small number of deployment decisions. A diverse product list can coexist with concentrated purchasing power.

Source: Nvidia Q2 FY2027 Form 10-QSource: Broadcom Q3 FY2026 Form 10-Q

The counterargument to buying the suppliers

A supplier-first thesis can fail even while AI adoption grows. Buyers can improve utilization, develop internal chips, negotiate lower prices or delay the next installation. Microsoft describes its own silicon alongside Nvidia and AMD products, illustrating that customers can also develop alternatives.

Lower computing costs can expand usage. Whether that expansion outweighs lower revenue per unit depends on demand and competitive position; this report does not estimate the balance. Physical-infrastructure suppliers face their own capacity, execution and project-timing risks.

Buyers could ultimately retain more value through distribution, customer relationships and services. Their present investment burden may create assets that earn for years. Suppliers, meanwhile, can face a spending downturn after a strong installation cycle.

Source: Microsoft FY2026 Q4 earnings call

What to watch in the next results

For buyers: compare cash generation with infrastructure payments, track utilization evidence, and examine whether depreciation rises faster than the revenue the new capacity supports. A change in a capex definition or lease classification needs separate treatment from a change in physical investment.

For compute suppliers: watch repeat demand, customer concentration, gross-to-operating-profit conversion and the costs excluded from adjusted results. For power and cooling suppliers: watch order conversion, margins and working capital together. An order becomes more valuable to shareholders when delivery produces collectible revenue and cash.

Finally, connect operating progress to valuation. A company can execute well while its stock disappoints if the purchase price already assumes better results. We have not performed a current-price valuation here, so this report identifies exposures and evidence to monitor rather than an entry price.

Conclusion and method

Follow the AI dollar through three distinct questions: who commits the capital, who records the sale, and who retains the profit and cash. In this selected group, Nvidia, Broadcom and Vertiv already disclose substantial operating profits; the buyers' figures show the scale of capital they must put to work. That gives investors a practical starting point for comparison.

Evidence retrieved October 12, 2026. Figures describe the stated June-, July- and August-ended quarters, not October activity. The comparison uses GAAP company-wide operating income; Meta's free cash flow is its non-GAAP reconciled measure. No supplier allocation, thematic revenue share beyond Broadcom's disclosed category, or forecast has been invented. Alphabet is excluded because its primary transcript could not be retrieved reliably in this review.

AI-assisted research and writing checked against the linked primary sources. Calculations and business-role classifications are MacroShed analysis. Educational research; no individualized recommendation or claim about undisclosed author holdings.

Find your route into the future.

Compare the business, evidence and risks behind 20 companies and six funds.

Open the Exposure Atlas →

Sources and method

Primary sources establish the described products, reported figures and announcements. Our interpretation of exposure and risks is editorial analysis. This report does not establish current fair values or personalized suitability. Financial periods and holdings dates are shown explicitly.

  1. Meta Q2 2026 results and cash-flow reconciliation ↗

    Quarter ended June 30, 2026; retrieved October 12, 2026

  2. Microsoft FY2026 Q4 earnings call ↗

    July 29, 2026; fiscal quarter ended June 30, 2026

  3. Amazon Q2 2026 Form 10-Q ↗

    Quarter ended June 30, 2026; retrieved October 12, 2026

  4. Nvidia Q2 FY2027 financial results ↗

    August 26, 2026; quarter ended July 26, 2026

  5. Broadcom Q3 FY2026 financial results ↗

    September 2, 2026; quarter ended August 2, 2026

  6. Vertiv Q2 2026 financial results ↗

    July 29, 2026; quarter ended June 30, 2026

  7. Eaton Q2 2026 financial results ↗

    July 31, 2026; quarter ended June 30, 2026

  8. Nvidia Q2 FY2027 Form 10-Q ↗

    Quarter ended July 26, 2026; retrieved October 12, 2026

  9. Broadcom Q3 FY2026 Form 10-Q ↗

    Quarter ended August 2, 2026; retrieved October 12, 2026

Updates and corrections

2026-10-12 — Initial publication. Historical periods and accounting definitions are explicit. Alphabet excluded from this review; no current valuation or complete AI-dollar allocation claimed.

No company or fund paid for this report. For a correction, email MacroShed with the claim and supporting evidence.