Corporate language has a tell, and it is filed once a year with the SEC.
Via News examined how public companies describe artificial intelligence in their 10-K annual reports — the document where a company states, under legal liability, what it does and what could go wrong — every year since 2019. The result is a near-perfect natural experiment.
The phrase “generative AI” appears in essentially no 10-K before 2023. One filing in 2021. Two in 2022. Then 69 in 2023, 513 in 2024, and 877 in 2025. The term for the technology did not exist in the corporate record, and then, in the space of two filing cycles, it was everywhere.
| 10-K filings mentioning… | 2021 | 2022 | 2023 | 2024 | 2025 |
|---|---|---|---|---|---|
| “generative AI” | 1 | 2 | 69 | 513 | 877 |
| “artificial intelligence” | 848 | 1,156 | 1,295 | 2,436 | 3,324 |
| “machine learning” | 619 | 867 | 896 | 1,243 | 1,645 |
Read across the three rows, the story sharpens. “Machine learning” — a decade-old term — climbs steadily; this is not simply AI becoming a buzzword. “Artificial intelligence” roughly doubles between the 2023 and 2024 filing seasons, the jump from a technology mentioned to a technology that has to be accounted for. And “generative AI,” which had no prior existence in the filings, arrives exactly at the ChatGPT moment and compounds. A footnote in 2022 becomes a risk factor and a strategy by 2024.
The money moved on the same timeline
Nvidia’s data-center revenue — the meter for the AI build-out — went from about $15 billion in fiscal 2023 to more than $47 billion in fiscal 2024 to $115 billion in fiscal 2025. The largest cloud operators lifted combined infrastructure spending from roughly $155 billion in 2023 to over $215 billion in 2024. FactSet, counting mentions on quarterly earnings calls rather than annual filings, sees the same break: from a low base to 177 S&P 500 companies citing AI on their second-quarter 2023 calls — a decade high at the time — and 331 by the fourth quarter of 2025. Three independent instruments — annual filings, capital spending, and quarterly calls — cross the same line in the same window.
And where the productivity lands is already visible. When Klarna reported in 2024 that an AI assistant was handling about two-thirds of its customer-service chats within a month — work it equated to some 700 full-time agents — it was describing implementation, not a pilot. Across the market, S&P 500 net profit margins expanded from a historical 6–7% to above 12% by 2024, while the share of income going to wages kept drifting down.
What we are not claiming
Two honest limits. First, these counts are by filing date and cover the whole 10-K universe, not a fixed S&P 500 panel — the trend is unambiguous, but the absolute numbers include companies entering and leaving the filing pool. Second, a mention is not a measure of depth; a fuller pass would separate AI-as-risk from AI-as-strategy. Both refinements are on our list; neither changes the shape.
The question the filings raise
If the annual reports have stopped hedging about AI, the open question is no longer whether it reshapes the economy — the companies themselves now say it does. It is who shares in what it produces. On that, the same filings are blunt: in 2024, S&P 500 companies returned a record ~$1.6 trillion to shareholders through buybacks and dividends. The technology is set. The distribution is not.
