thetrading.tools
ValuationFrom SEC income statements · quarterly, through Q1 2026

Is AI Making US Companies More Efficient?

Overhead and operating margins at the typical large US company since ChatGPT launched, each company measured against itself, beside the profit growth of the companies building AI. The question is whether the companies using AI are earning a return on it yet, and where in their accounts that would show first.

Efficiency dividendthrough Q1 2026
Not visible
Overhead at the typical company is +0.14 points of sales from its Q4 2022 level
4-qtr pace
−0.08
norm −0.14
Margin
+0.65
pts vs Q4 2022
Software
+7.7
margin pts
Build-out
66%
of profit growth

Rule: the dividend reads as emerging once overhead at the typical company falls at least as fast over four quarters as it typically did before ChatGPT (−0.14 points).

Latest read

Through Q1 2026, overhead per dollar of sales at the typical company outside the AI build-out and the software industry is +0.14 points from its own Q4 2022 level (106 companies reporting the full SG&A line). Over the latest four quarters it moved −0.08 points; before ChatGPT the typical four-quarter change was −0.14. The typical operating margin is +0.65 points from its Q4 2022 level. Meanwhile the 23 build-out companies supplied 66% of the rise in operating profit across the 214 companies with every quarter since then.

Sources, methodology & freshnessSEC EDGAR XBRL income statements for the covered companies: revenue, operating income and selling, general and administrative expense as filed · A quarterly series. The payload is rebuilt daily, but the latest reading advances only when a new quarter of filings is nearly complete, and the page is dated by the day its content last changed.Data as of 2026-10-06 · Open ↓
Source
SEC EDGAR XBRL income statements for the covered companies: revenue, operating income and selling, general and administrative expense as filed
Methodology
Trailing four quarters per company; each company measured against its own Q4 2022 level and the median taken across companies; overhead only from companies reporting the full SG&A line; groups fixed in advance; financials excluded
Updates
A quarterly series. The payload is rebuilt daily, but the latest reading advances only when a new quarter of filings is nearly complete, and the page is dated by the day its content last changed.Data as of 2026-10-06
Maintained & reviewed by Yuriy Matso — methodology shown on the page.
01

Overhead at the typical company

Range:
011.9201620182020202220242026ChatGPT+0.14
Each company's selling, general and administrative expense as a share of sales, measured against its own Q4 2022 level, with the median taken across companies reporting the full SG&A line. Points above zero mean overhead was higher than at the end of 2022. Before ChatGPT the line fell steadily; through Q1 2026 it reads +0.14 points across 106 companies, and 52% of them carry more overhead per dollar of sales than they did then.
02

The pace against its old norm

Range:
-0.9-0.30.22017201820192020202120222023202420252026ChatGPT−0.08
The same measure over rolling four-quarter windows, the test this page's verdict uses. From 2016 to the end of 2022 the median four-quarter change was −0.14 points. At Q1 2026 it is −0.08, slower than that norm.
03

Operating margins: ordinary companies and software

Range:
-5.90.97.7201620182020202220242026ChatGPT+0.657.7
Other companies: median operating margin minus its own Q4 2022 level (points)Software and IT services (16 companies)
Each company's operating margin against its own Q4 2022 margin, the median taken across companies. Through Q1 2026 the typical company outside the build-out and software is +0.65 points higher, with 57% above their Q4 2022 margin. Most of that came in 2023, when input costs eased and prices held, which is pricing rather than efficiency. The dashed line is software and IT services, +7.7 points; those companies sell AI products as well as use them, so their line cannot settle the question.
04

Where the profit growth went

97.2187.7278.22023202420252026278249.8122.0119.2
AI build-out (23): operating profit, Q4 2022 = 100Software (14): operating profitOther companies (177): operating profitOther companies: sales
Total operating profit by group, the same 214 companies at every point, indexed to Q4 2022. The build-out's profit is up 178% on sales up 66%; other companies' profit is up 22% on sales up 19%. The build-out supplied 66% of the rise (hardware 34%, cloud 32%); Meta, counted with the other companies because its revenue is advertising, supplied 9% on its own. These are total company profits, so they show where growth occurred, not how much of it AI caused. The other companies' combined margin is 12.3%, against a high of 12.7% since Q4 2022.
05

Method

Groups, fixed in advance

Build-out hardware: ALAB, AMAT, AMD, ANET, APH, AVGO, CDNS, CIEN, COHR, DELL, GLW, KLAC, LITE, LRCX, MPWR, MRVL, MU, NVDA, SMCI, STX, TER, VRT, WDC. Cloud: AMZN, CRWV, GOOGL, MSFT, ORCL. Software: every other company our SIC mapping places in IT Services or Software. Other: everyone else outside financials, 243 companies, Meta included.

Each company against itself

Figures are trailing four quarters. Every typical-company number is the median of each company’s own change from its Q4 2022 level, never a difference of median levels, which can move because the companies in the sample change. Medians are not added together or split into shares.

Overhead

Selling, general and administrative expense, only from the 150 other companies whose filings report the full line. Some companies book support in cost of sales and engineers in research and development, so savings can appear outside SG&A; operating margin is shown beside it for that reason.

Coverage and timing

The companies are today’s largest, carried back, so earlier quarters leave out companies that shrank or disappeared. The latest quarter advances once 90% of companies have reached it in SEC’s structured data. Built from the same filings as Corporate Capex and Market P/E Ratio. Content last changed October 6, 2026.

How Corporate Efficiency Tracker Works

  1. 1
    Split the companies before looking at results
    Companies selling AI capacity (chip, memory, networking, server and data-centre hardware makers, and the cloud providers renting out computing power) form the build-out. Software and IT services companies, which sell AI products as well as use them, are reported separately. Everyone else outside financials is the group the efficiency question is about. The lists are fixed in the builder, never inferred from the results.
  2. 2
    Measure overhead only where it is reported the same way
    Overhead is selling, general and administrative expense as a share of sales, taken only from companies whose filings report the full SG&A line. Companies that report general and administrative expense alone leave out sales and marketing, so they are excluded from this measure rather than mixed in.
  3. 3
    Compare each company with itself
    Every typical-company figure is the median of each company's own change from its Q4 2022 level, the quarter ChatGPT launched. A difference between two median levels can move simply because the companies in the sample change; a median of company-level changes cannot.
  4. 4
    Hold the latest quarter until it is nearly complete
    The page moves to a new quarter only once at least 90% of the companies that reported the quarter before have filed it in SEC's structured data, which can lag the filings themselves by months.

Who Uses Corporate Efficiency Tracker

Investors weighing the AI payoff
The companies selling AI capacity report its revenue every quarter. Whether the companies buying it earn a return is harder to see, and this page tracks the first place that return should show up in their accounts.
Anyone testing an AI productivity claim
Claims that AI is raising corporate efficiency can be checked here against overhead and margins at the typical large company, with the pre-ChatGPT record as the yardstick.
Readers of the productivity statistics
Official productivity figures cover the whole economy. This page shows whether the large listed companies, where AI adoption is most reported, are behaving differently.

Pro Tips

01
Watch the four-quarter pace against its old norm
Before ChatGPT, overhead per dollar of sales fell steadily at the typical company. A four-quarter pace back at or beyond that norm would be the first sign in the accounts that savings are reaching ordinary companies.
02
Read margins and overhead together
Margins can rise for reasons that have nothing to do with efficiency, such as prices holding while input costs fall, which is what happened in 2023. A margin gain with flat overhead is pricing; a margin gain with falling overhead is closer to an efficiency gain.
03
Treat software as a separate case
Software companies sell AI products, so their margins mix their own efficiency with demand for what they sell. Their line is shown, but it cannot settle whether AI is helping its users.

Common Issues & Solutions

The profit chart is not AI profit▾
It shows each company's total operating profit, including businesses unrelated to AI. It shows where profit growth occurred since ChatGPT; it does not measure how much of it AI caused.
AI savings may not show up in overhead▾
Some companies book customer support in cost of sales and engineers in research and development, so savings can show up outside SG&A. Overhead is simply where they should be easiest to see, and operating margin is shown beside it to catch the rest.
The companies are today's▾
The sample is today's largest companies carried back, so earlier quarters leave out companies that shrank or disappeared. The changes are company-level, which limits but does not remove that bias.

Frequently Asked Questions

Is AI making companies more efficient?▾
Not visibly yet, outside software. At the typical large US company outside the AI build-out and the software industry, overhead per dollar of sales has not fallen since ChatGPT launched, after falling steadily for years before it, and operating margins have risen by an ordinary amount. Software companies have widened margins sharply, but they also sell AI products. The page updates as each quarter of filings completes.
Who has gained the most profit since ChatGPT?▾
The companies building AI capacity. The hardware makers and cloud providers in the build-out group account for about two-thirds of the rise in operating profit across the companies tracked here since the end of 2022. These are their total profits, so the figure shows where growth occurred rather than how much AI caused.
What is SG&A as a percentage of sales?▾
Selling, general and administrative expense covers the costs of running a company outside making its product: sales, marketing, finance, legal, administration and much office work. Dividing it by sales shows how much overhead each dollar of revenue carries. A falling ratio means the company is running its overhead more efficiently relative to its size.
Why compare with Q4 2022?▾
ChatGPT launched on November 30, 2022, the start of the current wave of AI adoption. Measuring each company against its own Q4 2022 level shows what has changed since, and the same measure over earlier years shows what normal change looked like before.
Why does the latest quarter lag the calendar?▾
The page uses SEC's structured financial data, which can trail the filings themselves by weeks or months. It moves to a new quarter only once nearly all companies have reached it, so a partly filed quarter never changes the answer by changing the sample.

Explore Other Tools

Corporate Capex Intensity
What the build-out costs the companies paying for it: capital spending against operating cash flow.
Labor Productivity
Output per hour across the whole nonfarm business sector, quarterly.
Corporate Profit Margin
The economy-wide margin from the national accounts, private companies included.
Market P/E Ratio
What investors pay for these profits, from the same filings.
Market Return Drivers
Revenue, margin or re-rating: where the market's gains came from.
Unit Labor Costs
The cost side of productivity: compensation per unit of output.
Last updated: 2026-10-06