AI Stack Report · Q2 2026 · published October 28, 2026
AI Stack Report: Q2 2026
What the filings for Q2 2026 say about the AI build-out: how much the biggest buyers spent, which suppliers got paid and kept it, whether cloud revenue is catching up, and what changed in the financing behind it. Published October 28, 2026.
Key findings
- The five largest buyers spent $182B on capital projects in Q2 2026, +22% on Q1 2026 and +87% on Q2 2025. That was 97% of their operating cash flow, against 72% a year earlier.
- Revenue at the four big cloud businesses reached $132B, +40% on Q2 2025; Oracle cloud infrastructure grew fastest, +93%.
- Among suppliers, memory grew fastest: +253% on Q2 2025, and the biggest margin move was in memory (+43.3 pts of gross margin).
- Amazon, Alphabet, Oracle and CoreWeave spent more on capex than they generated from operations in the quarter.
What changed since the last report
This is the first edition, so it compares Q2 2026 with the quarter before (Q1 2026) and the same quarter a year earlier (Q2 2025). From the next edition this section lists what changed between reports: new financing links in the filings, changes in leases and in the useful lives the buyers state for their servers, and the move in the buyers’ capex-to-cash-flow ratio.
1. The spending
Capital spending is where every quarter’s story starts, because the buyers’ capex is the revenue of nearly every other company in the stack. In Q2 2026 Microsoft, Amazon, Alphabet, Meta and Oracle together spent $182B, against $148B in Q1 2026 and $97.3B in Q2 2025. Their operating cash flow was $186B, so capex used 97% of it.
The fastest increase on a year earlier came from Microsoft (+110%), the slowest from Amazon (+68%). Depreciation and amortization across the five was $46.2B in the quarter: capex ran at 3.9 times the expense it will eventually become.
Companies report on different fiscal calendars, so each one’s quarter is aligned to the nearest calendar quarter: Oracle’s quarter to the end of May counts as Q2, for example. Capex is cash paid for property and equipment; equipment under finance leases and leases signed for data centers not yet built come on top, and are covered below.
| Company | Capex, Q2 2026 | vs Q1 2026 | vs Q2 2025 | Capex ÷ operating cash flow |
|---|---|---|---|---|
| Microsoft | $35.8B | +16% | +110% | 65% |
| Amazon | $54.2B | +23% | +68% | 119% |
| Alphabet | $44.9B | +26% | +100% | 115% |
| Meta | $30.1B | +59% | +82% | 95% |
| Oracle | $16.5B | −11% | +82% | 113% |
| CoreWeave | $6.4B | −17% | +162% | 946% |
The five buyers’ capex, operating cash flow and depreciation
Per calendar quarter, combined, as of this report
- Operating cash flow
- Capital spending
- Depreciation and amortization
Source: company 10-K and 10-Q filings (SEC XBRL), frozen at publication. Finance leases excluded.
Show the numbers
| Quarter | Capex | Operating cash flow | D&A |
|---|---|---|---|
| Q3 2023 | $38.3B | $109.8B | $18.9B |
| Q4 2023 | $44.0B | $99.8B | $21.1B |
| Q1 2024 | $46.0B | $104.5B | $19.3B |
| Q2 2024 | $55.6B | $114.6B | $20.2B |
| Q3 2024 | $61.2B | $123.0B | $27.0B |
| Q4 2024 | $76.3B | $136.3B | $30.4B |
| Q1 2025 | $77.8B | $120.2B | $29.5B |
| Q2 2025 | $97.3B | $134.6B | $32.0B |
| Q3 2025 | $105.8B | $167.1B | $35.8B |
| Q4 2025 | $130.7B | $180.9B | $40.5B |
| Q1 2026 | $148.4B | $157.9B | $42.6B |
| Q2 2026 | $181.5B | $186.4B | $46.2B |
2. The return: cloud revenue
If the spending pays, it shows up first in the cloud segments. AWS, Microsoft Cloud, Google Cloud and Oracle’s cloud infrastructure brought in $132B in Q2 2026, against $117B in Q1 2026 and $94.2B in Q2 2025. AWS earned $16.6B of operating income, +64% on a year earlier; Google Cloud earned $8.8B of operating income, +212% on a year earlier; Microsoft’s Intelligent Cloud segment earned $16.0B of operating income, +31% on a year earlier.
Compare the growth rates: cloud revenue +40% on a year earlier, capex +87%. When capex grows faster, the spending is running ahead of the revenue it is meant to earn, which is normal while capacity is being built and a concern if it persists after the capacity is live.
| Cloud business | Revenue, Q2 2026 | vs Q1 2026 | vs Q2 2025 |
|---|---|---|---|
| AWS | $42.2B | +12% | +37% |
| Microsoft Cloud | $59.3B | +9% | +27% |
| Google Cloud | $24.8B | +24% | +82% |
| Oracle cloud infrastructure | $5.8B | +18% | +93% |
3. Who got paid
Down the stack, the clearest single read is Nvidia’s data-center revenue: $89.0B in the quarter, +18% on Q1 2026 and +117% on Q2 2025. Micron’s DRAM revenue, which includes the high-bandwidth memory beside every accelerator, was $31.3B, +343% on a year earlier.
By layer, using the companies on our map that reported all three quarters, servers +48%, chips +74%, memory +253%, networking +30%, power & dcs +26%, foundry & tools +29% on a year earlier. Gross margins moved most in memory (+43.3 pts) and chips (+6.8 pts): rising margins mean scarcity and pricing power, falling ones mean supply catching up or costs passing through.
| Segment | Revenue | vs Q2 2025 |
|---|---|---|
| Nvidia Data Center | $89.0B | +117% |
| Broadcom Semiconductor Solutions | $20.8B | +127% |
| Micron DRAM | $31.3B | +343% |
| Dell AI-optimized servers and networking | $16.4B | +100% |
| AMD Data Center | $6.7B | +107% |
| Layer | Revenue | vs Q2 2025 | Gross margin |
|---|---|---|---|
| AI clouds | $494B | +22% | 61% (61%) |
| Servers | $91.7B | +48% | 18% (14%) |
| Chips | $166B | +74% | 67% (61%) |
| Memory | $58.0B | +253% | 81% (37%) |
| Networking | $39.5B | +30% | 53% (52%) |
| Power & DCs | $56.8B | +26% | 27% (27%) |
| Foundry & tools | $26.8B | +29% | 52% (51%) |
| Software | $46.5B | +22% | 76% (77%) |
Layer totals cover the companies on our map with at least $100 million of revenue that reported all three quarters; gross margin a year earlier in brackets.
At the top of the stack, the software companies we track took in $46.5B in the quarter, +22% on a year earlier, at a gross margin of 76%. Their growth is the slowest-moving test of whether AI is making software more valuable to the businesses that buy it.
4. The money loop and the commitments outside capex
The filings we read disclose 21 financing links between suppliers and their customers: investments, credit and guarantees, capacity buy-backs, warrants and long power contracts, each quoted on the money-loop page. Separately, the buyers disclose $1.04T of data-center leases signed but not yet started, outside capex and the balance sheet.
| Link | What the filing says | Source |
|---|---|---|
| Microsoft → OpenAI | $11.9B funded of $13.0B committed | Microsoft 10-K |
| OpenAI → Microsoft | $24.1B of Microsoft revenue in FY2026 | Microsoft 10-K |
| Amazon → Anthropic | $8.0B in convertible notes, 2023–2025 | Amazon 10-Q |
| Amazon → Anthropic | $5.0B in preferred stock, 2026 | Amazon 10-Q |
| Amazon → Anthropic | Up to $20B facility, released as compute is delivered | Amazon 10-Q |
| Anthropic → Amazon | Commitment to AWS expanded by more than $100B over 10 years | Amazon 10-Q |
| Nvidia → OpenAI | Lease guarantees for OpenAI, capped at $105B | Nvidia 10-Q |
| Nvidia → Nvidia’s private holdings | $99B invested, $25B more committed | Nvidia 10-Q |
| Nvidia → AI cloud partners | $36B of cloud capacity it has agreed to buy back | Nvidia 10-Q |
| Nvidia → Nvidia’s suppliers | $279B of supply and capacity commitments | Nvidia 10-Q |
| Nvidia → CoreWeave | $2.0B share purchase, January 2026 | CoreWeave 10-Q |
| Microsoft → CoreWeave | 67% of CoreWeave’s 2025 revenue | CoreWeave 10-K |
| Meta → CoreWeave | Up to $21.0B through 2032 | CoreWeave 10-Q |
| OpenAI → CoreWeave | Up to $6.5B through May 2031 (one order form) | CoreWeave 10-Q |
| AMD → OpenAI | Warrant for up to 160M AMD shares at $0.01 | AMD 10-Q |
| AMD → Meta | Warrant for up to 160M AMD shares at $0.01 | AMD 10-Q |
| HPE → Oracle | Warrant for up to 4.2M HPE shares, tied to networking in Oracle’s AI data centers | HPE 10-Q |
| Meta → Meta’s data-center ventures | $41.5B of residual value guarantees, $68.4B maximum exposure | Meta 10-Q |
| Meta → Third-party clouds | $454.5B of non-cancelable commitments, mostly cloud capacity | Meta 10-Q |
| Microsoft → Constellation | 20-year PPA for a restarted nuclear plant | Constellation 10-Q |
| Meta → Vistra | 2,609 MW of nuclear power under 20-year PPAs | Vistra 10-Q |
How to read this report
How to read the comparisons. Each figure is compared with the quarter before and the same quarter a year earlier. The year-on-year comparison is the one to trust for trends, because several of these businesses are seasonal and capex lumps up around new data-center openings. The quarter-on-quarter change shows momentum, and is most useful when it agrees with the yearly one.
What the layer figures include. Layer totals add up the companies on our map with at least $100 million of revenue that reported all three quarters being compared, so a newly listed company doesn’t inflate the growth rate. They measure the direction and the split of the money, not the whole market: private companies and foreign filers without quarterly SEC reports are outside them.
What to watch next quarter
- Capex guidance on the next round of earnings calls, and whether the capex-to-cash-flow ratio rises further.
- Cloud segment growth against capex growth: the gap needs to start closing as new capacity is sold.
- Gross margins at the chip and memory makers, the first place supply catching up with demand would show.
- New investments, guarantees, warrants or capacity buy-backs in the filings.
- Any change in the useful lives the buyers state for their servers.
The live numbers, updated with every data refresh, are on the AI stack map and the capex tracker. This report’s figures are frozen as of publication; data last pulled from SEC filings on October 28, 2026.
Free · quarterly · from SEC filings
Get the next AI Stack Report
After each Big Tech earnings season we update the map from the new filings and publish what changed: who spent, who got paid, who kept it and what the filings reveal about the loop. Free, in Rob May’s Investing in AI newsletter.
Form not showing? Subscribe on Substack. No spam; unsubscribe any time.
About this report
How much did Big Tech spend on AI capex in Q2 2026?
Microsoft, Amazon, Alphabet, Meta and Oracle reported $182B of capital spending in Q2 2026 (calendar-aligned), +87% on Q2 2025. That was 97% of their combined operating cash flow.
What is the AI Stack Report?
A quarterly report from Investing in AI that updates the AI stack map from the latest SEC filings after each earnings season: how much the largest buyers spent, which supplier layers were paid and kept the money, how cloud revenue is tracking, and what changed in the financing behind the build-out.
Why are the numbers frozen?
Each report is a dated publication. Its figures are saved when it is published, so later data refreshes or restatements don’t change what a past report said. The live AI stack pages always show the latest data.
Where do the numbers come from?
From companies’ 10-K and 10-Q filings with the SEC: XBRL financial data, segment disclosures and quoted text for financing links and leases. No estimates or paid data are used.
How do I get the next report?
Subscribe to the free Investing in AI newsletter; each new report is announced there when it is published.
Research and education, not investment advice: nothing here is a recommendation to buy or sell any security. Figures come from companies’ SEC filings as reported; companies named were chosen because they show how the stack works, not as a view on their shares.