Amazon paid $38.7 billion into OpenAI and Anthropic in the first half of 2026, on top of $98.4 billion in cash purchases of property and equipment. Meta made no share repurchases during the same period. Alphabet received $100.5 billion from reported equity proceeds and net long-term borrowing. Those are three distinct financial choices behind the familiar capex headline. ([2], [1], [3])
Together, the three companies still produced substantially more revenue, operating profit and operating cash than a year earlier. The investment bill grew faster. But the consequences depend on what else each company buys, which payments it can change, and how it chooses to finance them.
That is the central finding of this comparison: capex alone understates the range of cash commitments, while a capex-versus-cash-flow gap says too little about the response available to management. These are whole-company financial statements; “cash PPE” below means gross cash purchases of property and equipment, not an AI-only spending series. ([1], [2], [3])
The operating engines are growing. The investment burden is growing faster.
Across Meta, Amazon and Alphabet, H1 revenue rose 21.5% and operating income rose 26.2% from 2025 to 2026. Operating cash flow increased 35.2%, to $220.4 billion. Gross cash purchases of property and equipment rose 80.6%, to $228.1 billion. ([1], [2], [3])
Figure 1. Three-company pooled growth
Pooled H1 2025-to-H1 2026 growth for Meta, Amazon and Alphabet: revenue 21.5%, operating income 26.2%, operating cash flow 35.2%, and gross cash purchases of property and equipment 80.6%. Cash purchases grew fastest.
Pooled H1 2025–2026 growth
View exact data: Pooled H1 2025–2026 growth
| Metric | H1 2025 (USD billions) | H1 2026 (USD billions) | Pooled growth (%) |
|---|---|---|---|
| Revenue | 599.861 | 728.928 | 21.51615124170432816935923490 |
| Operating income | 137.450 | 173.426 | 26.17388141142233539468897780 |
| Operating cash flow | 163.014 | 220.366 | 35.18225428490804470781650660 |
| Gross cash purchases of property and equipment | 126.324 | 228.122 | 80.58484531838763813685443780 |
This is a different starting point from a story about collapsing underlying businesses. At the combined level, cash PPE purchases grew much faster than operating cash flow. That increases the importance of investment timing and capital allocation even before one makes a judgment about the eventual return.
The differences between companies are already substantial. H1 operating cash less gross cash PPE purchases remained positive at Meta ($15.0 billion) and Alphabet ($4.3 billion), while Amazon’s was negative $27.0 billion. These are a limited comparison, not the companies’ reported free-cash-flow measures or their total financing needs. Pooling them would hide the fact that each company controls a different balance sheet. ([1], [2], [3])
Amazon is paying for model-company stakes as well as infrastructure.
Amazon’s investment cash extends well beyond equipment. Its H1 payments included $28.7 billion for OpenAI and $10.0 billion for Anthropic. Adding those disclosed equity payments to gross cash PPE gives $137.1 billion of selected investment uses, compared with $71.4 billion of operating cash flow. This is a selected-use comparison, not a reconstruction of Amazon’s entire cash-flow statement. ([2])
Figure 2. Amazon: selected investment cash uses
Amazon H1 2026 gross cash purchases of property and equipment were $98.4 billion, actual OpenAI equity cash was $28.7 billion, and actual Anthropic equity cash was $10.0 billion. The selected-use sum is $137.1 billion. A separate operating-cash-flow reference is $71.4 billion; it is not a funding bridge.
Selected H1 2026 cash uses
View exact data: Selected H1 2026 cash uses
| Item | USD billions |
|---|---|
| Gross cash PPE purchases | 98.411 |
| OpenAI equity cash | 28.7 |
| Anthropic equity cash | 10 |
The economic distinction matters. Infrastructure and equity stakes give Amazon different kinds of exposure to the AI market. One commits capital to operating assets; the other adds ownership exposure to private model companies. Their values and cash returns need not develop on the same timetable. A capex chart captures the first category and misses the second.
The timing also changes the forward picture. Amazon’s filing says the remaining $21.3 billion of its OpenAI commitment was funded after June 30. That payment is outside the H1 chart. Anthropic’s separately described financing facility is conditional; the $15 billion remaining after the Series H investment is potential additional funding, not another H1 cash outlay. ([2])
Amazon also reported $64.2 billion of net long-term debt cash proceeds, while cash and marketable securities were almost unchanged between December and June. Those observations show how active the financing side had become. They do not establish that a particular dollar of borrowing funded a particular AI investment. ([2])
The implication is that investors need to track two investment exposures: the operating return on new assets and the value and eventual cash return of the stakes. Faster cloud demand alone would not settle both questions. Conversely, a successful equity investment could create value before it delivers cash available for the next construction bill.
A strong earnings headline can still overstate the cash available to invest.
Amazon’s H1 net income rose from $35.3 billion to $92.9 billion. Operating income rose from $37.6 billion to $51.3 billion, and operating cash flow rose from $49.5 billion to $71.4 billion. All three improved, but at very different speeds. ([2])
Figure 3. Amazon: earnings and cash generation
Amazon consolidated H1 2025 versus H1 2026: operating income rose from $37.6 billion to $51.3 billion, net income from $35.3 billion to $92.9 billion, and operating cash flow from $49.5 billion to $71.4 billion. H1 2026 included a $62.8 billion pretax noncash investment revaluation gain; no adjusted net income is calculated.
Consolidated earnings and operating cash
View exact data: Consolidated earnings and operating cash
| Metric | H1 2025 (USD billions) | H1 2026 (USD billions) |
|---|---|---|
| Operating income | 37.576 | 51.313 |
| Net income | 35.291 | 92.902 |
| Operating cash flow | 49.53 | 71.419 |
The filing identifies $62.8 billion of noncash investment revaluation gains during H1. Those gains help explain why the net-income headline moved so much more than operating performance. They did not themselves produce cash to pay for equipment or additional equity investments. ([2])
This is also why the evidence should not be reduced to a bearish accounting story. Operating income grew about 37% and operating cash flow about 44%. The operating business improved materially. The analytical mistake would be to use the much larger rise in net income as evidence that the investment programme had already funded itself.
A more useful test is whether future operating earnings and cash generation keep strengthening independently of private-company valuation marks. Valuation gains can be economically meaningful, but they answer a different question from the one facing the treasury team.
Meta’s cash mix changed sharply as buybacks stopped.
Meta’s gross cash PPE purchases increased by $19.6 billion, or 66.6%, in H1. Yet cash share repurchases fell from $22.9 billion to zero, while dividends stayed close to $2.7 billion. Across those three selected uses, total payments actually fell from $55.1 billion to $51.8 billion. ([1])
Figure 4. Meta: three selected cash uses
Meta H1 2025 versus H1 2026 selected uses in USD billions: gross cash purchases of property and equipment $29.5 and $49.1; share repurchases $22.9 and $0.0; dividends $2.7 and $2.7. The three selected-use totals are $55.1 billion and $51.8 billion. No causal link is asserted.
Selected uses, excluding other cash activity
View exact data: Selected uses, excluding other cash activity
| Metric | H1 2025 (USD billions) | H1 2026 (USD billions) |
|---|---|---|
| Gross cash PPE purchases | 29.479 | 49.113 |
| Share repurchases | 22.921 | 0 |
| Dividends | 2.656 | 2.699 |
That does not prove that capex caused the buyback pause. It does show why a spending-growth headline is an incomplete description of the change in cash allocation. Less cash went to repurchases while much more went to property and equipment.
For shareholders, the trade-off is tangible. The absence of repurchases removed a major cash use, while one channel for returning cash to shareholders stopped. That is a change in the form of capital allocation, even though Meta’s operating cash flow still exceeded the gross PPE purchases used here.
Meta also recorded $24.9 billion in net proceeds from long-term debt issuance. Separately, it disclosed $10.8 billion in restricted money-market escrow tied to multiyear infrastructure purchase agreements, expected to be released during 2028–2030. That escrow was unavailable for general corporate use. A balance-sheet asset can therefore represent a future commitment rather than freely deployable liquidity. ([1])
The next evidence to watch is whether operating cash growth can support both the enlarged investment programme and a resumption of repurchases. Continued zero buybacks by itself would not prove financial distress; it would remain a capital-allocation choice whose cost and benefits need to be assessed.
Alphabet enlarged its liquidity buffer with outside capital.
Alphabet’s H1 financing provides a third pattern. It received $49.6 billion in common and mandatory convertible preferred equity proceeds, plus $51.0 billion in net long-term debt cash proceeds. Cash and marketable securities rose from $126.8 billion at December 31 to $242.5 billion at June 30. ([3])
Figure 5. Alphabet: funding flows and liquidity stocks
Separate panels show Alphabet H1 2026 selected funding cash flows—$30.5 billion common equity, $19.1 billion mandatory convertible preferred, and $51.0 billion net long-term debt cash flow, selected total $100.5 billion—and cash plus marketable securities balances of $126.8 billion at December 31, 2025 and $242.5 billion at June 30, 2026. Financing is not attributed to the balance change.
Selected financing cash flows
View exact data: Selected financing cash flows
| Item | USD billions |
|---|---|
| Common equity | 30.499 |
| Mandatory convertible preferred | 19.063 |
| Net long-term debt cash flow | 50.973 |
Cash plus marketable securities — balance-sheet stocks
View exact data: Cash plus marketable securities — balance-sheet stocks
| Balance date | USD billions |
|---|---|
| 2025-12-31 | 126.843 |
| 2026-06-30 | 242.474 |
The financing flows and the liquidity balances are different measures. The chart does not attribute the entire change in holdings to those financing lines. The practical point is that a large liquidity balance should not be mistaken for cash generated entirely by the operating business. Alphabet described the financing as serving general corporate purposes, including AI infrastructure; the filing does not trace proceeds to individual projects. ([3])
There were also significant uses outside the equipment line. Alphabet closed its Wiz acquisition on March 11 at a stated adjusted purchase price of $29.5 billion. Its H1 cash-flow line for acquisitions, net of acquired cash, and purchases of intangible assets was $33.7 billion. That aggregated line cannot be labelled “cash paid for Wiz,” and the cybersecurity acquisition should not simply be relabelled AI capex. ([3])
The implication is a wider financial buffer, coupled with the obligations and ownership consequences of the funding instruments used to create it. The relevant follow-up is the cost and terms of that capital, the pace of subsequent cash deployment, and the cash the enlarged business generates. A healthy liquidity stock alone does not establish investment payback.
The next test is how those choices evolve.
The H1 evidence supports a more differentiated reading than a single Big Tech cash-shortfall number. Amazon’s selected investment outlays include large model-company stakes; Meta’s mix shifted sharply away from buybacks; Alphabet combined strong operations with substantial outside funding and a larger liquidity buffer.
Three developments would change this reading:
- Operating cash growth catching up with cash PPE purchases would reduce reliance on other sources of funding for that particular use. Slowing operating income and cash flow together would weaken the current picture of growing operating franchises.
- Fewer strategic-investment payments after the disclosed commitments are met would ease a source of cash demand outside capex. Further large payments would extend it. Amazon’s post-period OpenAI payment must be counted in its own period, not twice.
- Resumed shareholder returns alongside sustained investment, or less need for fresh financing, would provide evidence of greater financial flexibility. Larger debt or equity issuance would require closer attention to financing terms and future claims on cash—not an automatic verdict of distress.
The opportunity is substantial if today’s investment generates durable operating cash or valuable ownership stakes. The risk is that investors judge that progress using the wrong evidence: a paper gain as cash payback, a pooled deficit as a company’s funding problem, or a large liquidity balance as proof of internally funded expansion.
Following the cash choices alongside the operating results gives a clearer picture of what the AI investment cycle is asking of each company—and what would have to happen for those choices to pay off.
Explore AI Economy Radar’s methodology for how we connect company accounts, operating performance and capital allocation.
Data note
Financial comparisons use the six months ended June 30, 2025 and 2026. Liquidity compares December 31, 2025 with June 30, 2026; later events are labelled separately. Cash PPE means gross cash purchases of property and equipment, excluding noncash additions and finance-lease principal; it is not AI-only capex. GitHub’s logged cash-PPE and cash-flow inputs were reconciled to the three issuers’ Q2 2026 filings, with primary filings supplying the additional earnings, investment, financing and liquidity disclosures. Pooled growth is calculated from the three-company totals, not an average of growth rates. Selected-use charts are incomplete cash-flow comparisons. Amazon’s $62.8 billion revaluation gain is pretax; no adjusted after-tax net income is calculated by subtracting it. Cash and marketable securities are balance-sheet holdings; restricted escrow is identified separately. These measures do not estimate AI ROI.
Primary sources
1. Meta Platforms, Q2 2026 Form 10-Q, statements of income and cash flows, Notes 2 and 9
2. Amazon.com, Q2 2026 Form 10-Q, financial statements and Notes 2 and 6
3. Alphabet, Q2 2026 Form 10-Q, financial statements and Notes 6, 8 and 11