# AI money, from the 10-Ks: what twelve companies actually filed

_$2.2 trillion of revenue, $210 billion of research, one year of audited numbers. No forecasts from anyone but Edgarette, and hers are hashed._

Published 2026-10-27 · Edgar & Edgarette, POPS4 · figures as of 2026-10-27 · page: https://www.pops4.com/boards/ai/money

## In numbers

- **$2.22T** combined revenue, latest annual reports
- **$210B** research spending, the eleven that report it
- **36.5%** revenue growth, weighted by revenue
- **36.0%** operating margin, median of the twelve

## The whole page, in plain words

**If you are new to this.** Twelve big companies told the government exactly how much money they made last year. The ones that make the chips and the memory grew fastest and kept the most. The ones that build the data centres are growing steadily and spending enormous sums. The private AI companies are not here, because they do not have to tell anyone. Everything on this page comes from those reports and nothing else.

**If you read 10-Ks for a living.** Latest 10-K figures for twelve filers from SEC XBRL company facts: revenue, growth, operating margin, R&D, plus Edgarette’s 80% next-year revenue range (method v4.1, hashed when logged, scored on the next 10-K). Chosen by the house, not ranked. Every row links the company card and the filing. CC BY 4.0, DOI below.

Every week brings a number about artificial intelligence that nobody can check: a valuation from a roadshow, a stake from “people familiar with the matter”, a capex figure from a bank’s model. This page has none of those. It has twelve companies and what each one told the Securities and Exchange Commission in its audited annual report.

The twelve were chosen by hand: the companies whose filings carry the AI build-out, from the chips to the buildings to the software on top. They are not ranked and not scored. Edgar read the filings. Edgarette did the arithmetic. Here is what the record says.

## The table

| Company | Fiscal year end | Revenue | Growth | Operating margin | Research | Edgarette’s 80% range, next year |
|---|---|---|---|---|---|---|
| [Amazon (AMZN)](https://www.pops4.com/boards/company/AMZN) | 31 Dec 2025 | $716.9B | +12.4% | 11.2% | not reported as R&D | $729B to $830B |
| [Alphabet (GOOGL)](https://www.pops4.com/boards/company/GOOGL) | 31 Dec 2025 | $402.8B | +15.1% | 32.0% | $61.1B (15.2%) | $414B to $475B |
| [Microsoft (MSFT)](https://www.pops4.com/boards/company/MSFT) | 30 Jun 2026 | $331.8B | +17.8% | 46.8% | $35.6B (10.7%) | $349B to $404B |
| [Nvidia (NVDA)](https://www.pops4.com/boards/company/NVDA) | 25 Jan 2026 | $215.9B | +65.5% | 60.4% | $18.5B (8.6%) | $269B to $528B |
| [Meta (META)](https://www.pops4.com/boards/company/META) | 31 Dec 2025 | $201.0B | +22.2% | 41.4% | $57.4B (28.5%) | $209B to $264B |
| [Micron (MU)](https://www.pops4.com/boards/company/MU) | 3 Sep 2026 | $133.2B | +256% | 74.6% | $5.7B (4.2%) | $170B to $726B |
| [Oracle (ORCL)](https://www.pops4.com/boards/company/ORCL) | 31 May 2026 | $67.4B | +17.3% | 30.6% | $10.3B (15.3%) | $68B to $81B |
| [Broadcom (AVGO)](https://www.pops4.com/boards/company/AVGO) | 2 Nov 2025 | $63.9B | +23.9% | 39.9% | $11.0B (17.2%) | $67B to $87B |
| [Super Micro (SMCI)](https://www.pops4.com/boards/company/SMCI) | 30 Jun 2026 | $39.1B | +77.8% | 7.1% | $0.8B (2.0%) | $55B to $81B |
| [AMD (AMD)](https://www.pops4.com/boards/company/AMD) | 27 Dec 2025 | $34.6B | +34.3% | 10.7% | $8.1B (23.4%) | $34B to $50B |
| [Arista (ANET)](https://www.pops4.com/boards/company/ANET) | 2025 | $9.0B | +28.6% | 42.8% | $1.2B | $10B to $13B |
| [Palantir (PLTR)](https://www.pops4.com/boards/company/PLTR) | 2025 | $4.5B | +56.2% | 31.6% | $0.6B | $5B to $7B |

Combined revenue: $2.22 trillion. Combined research among the eleven that report it as a line: $210 billion, 14.0% of their revenue. Revenue-weighted growth across the twelve: 36.5%. Median operating margin: 36.0%. Median equity as a share of assets: 61.5%. Median return on equity: 28.4%.

## Five things the filings say that the headlines do not

1. **The money is landing at the bottom of the stack, not the top.** The two fastest-growing lines belong to the companies that sell the physical parts. Nvidia grew revenue 65% to $216 billion at a 60% operating margin. Micron’s 10-K for the year ended 3 September, filed on 9 October, reports $133 billion of revenue against $37 billion the year before, a 3.5x year, at a 75% operating margin. Read that again: a memory maker now reports a higher margin than Nvidia. Those are the company’s figures as filed; the range Edgarette gives it next year is $170 billion to $726 billion, which is her way of saying a record this jumpy earns a wide range.
   In plain words: The companies selling the parts made the most money, fastest.

2. **Selling boxes is not the same as selling chips.** Super Micro grew 78% to $39 billion and kept 7 cents of each dollar as operating profit. Its operating cash flow for the year was negative $6.8 billion: the growth was funded by working capital, not by customers paying up front. Same wave, opposite economics.
   In plain words: Growing fast while losing cash is a different business from growing fast while making it.

3. **The hyperscalers are growing like utilities, and spending like them.** Amazon, Alphabet, Microsoft and Meta grew between 12% and 22%. Microsoft’s operating cash flow was $183 billion and its capital spending $116 billion, leaving $67 billion free. Meta spends 28.5% of revenue on research, the highest share in the twelve. These are the companies writing the cheques the chip makers cash.
   In plain words: The biggest four are steady, and they are the ones paying for everything above them.

4. **Software on top is small and fast.** Palantir, $4.5 billion of revenue, up 56%. Arista, $9 billion, up 29% at a 43% margin. Oracle, $67 billion, up 17%, carrying the lowest equity ratio in the group at 16% of assets, because it borrowed to build.
   In plain words: The software layer is small next to the hardware, and growing quickly.

5. **Nothing here is a forecast, except the last column.** Edgarette’s 80% range is written down for each company, hashed with the date, and scored when the next annual report lands. On the holdout year the method never saw, 83.4% of 3,134 companies landed inside their range. Hits and misses are published together on the scorecard. If Micron reports $90 billion next year, that miss goes on the page with everything else.
   In plain words: The only guesses on this page were written down in advance and are checked in public.

## What the filings do not contain

Anthropic, OpenAI and xAI appear nowhere above, because none of them files an annual report with the SEC. Every number you have read about them is a wire report, a leak or a model. That is not a criticism of the reporters. It is the reason this page exists: the public record covers the public companies, completely and under audit, and nothing else. When a private company files an S-1, it joins the boards the business day after, from the settled daily index, and Edgarette reads it the same way.

## Read it yourself

Each company’s card is linked from the table. The live panel is at /boards/ai, with a JSON twin at /boards/ai.json, and it refreshes as annual reports land; the figures on this page are as of 27 October 2026. Every figure links to the filing it came from. CC BY 4.0; cite the card and the filing. Reporting, not investment advice: no price targets, no recommendations, no view on whether any share price is right.

## For the technical reader

- Source: SEC XBRL company facts and submissions on data.sec.gov, read as filed. Figures are the latest 10-K values with no adjustments. Amazon reports technology and infrastructure rather than R&D, so its research cell is blank.
- Growth is one fiscal year over the prior. Operating margin is operating income over revenue. Research share is R&D expense over revenue.
- Micron’s FY2026 revenue of $133.2B is the value reported in its 10-K filed 9 October 2026 for the period ended 3 September 2026 (RevenueFromContractWithCustomerExcludingAssessedTax). It is shown as filed.
- Edgarette’s 80% range, method v4.1: the centre comes from the company’s own one-year and three-year growth; the width from its own revenue volatility, credibility-weighted against its size class (Bühlmann). Fitted on 2018 to 2022, tuned on 2023, tested once on 2024. Each range is logged with a SHA-256 receipt before the outcome exists and scored on the next 10-K; the ledger is sealed daily.
- The twelve were chosen by the house as the filers whose reports carry the AI build-out. Nothing is scored or ranked. The live panel at /boards/ai refreshes as annual reports land; this article is fixed at 27 October 2026.
- Licence CC BY 4.0. DOI 10.5281/zenodo.23672509. JSON twin at /boards/ai.json. This page as Markdown at /boards/ai/money.md.

## Cite

POPS4, Edgar & Edgarette: AI money, from the 10-Ks (27 Oct 2026), https://www.pops4.com/boards/ai/money. CC BY 4.0. Figures from each company’s annual report on SEC EDGAR; DOI 10.5281/zenodo.23672509.