97 per cent of S&P 500 10-Ks mention AI. Only 11 per cent disclose it with substance.
Key takeaways
- The mention–substance gap is the editorial story. 97% of S&P 500 10-Ks mention AI; only 11% disclose AI activity with substance — measured against a six-component rubric covering quantified investment, named systems, named vendors, specific use cases, real risk language, and material 8-K events.
- Five companies score the maximum 5 on the substance rubric: NVIDIA, AMD, Meta, ServiceNow, and Cognizant — three substrate / platform names plus two services-and-application names.
- Only 30 of 503 S&P 500 filings name an external AI vendor. NVIDIA leads at 13 mentions, then Scale AI 9, OpenAI 4, Anthropic 2 — sparser than the public narrative around enterprise AI adoption would predict.
- Information Technology dominates substance density. 67% of IT 10-K filers cross the threshold; 6% of Financials filers; 0% in four sectors (Real Estate, Consumer Staples, Utilities, Energy).
- Pharma is the under-told story. Pfizer (Scale AI), Moderna (OpenAI), and Vertiv in industrials (NVIDIA) all score 4 with explicit named-vendor disclosures — a cohort worth tracking through the Q3 2026 cycle.
Of the 409 S&P 500 companies that filed an annual report (10-K) in calendar 2026, 397 mentioned artificial intelligence. Forty-six crossed the substance threshold on our six-component rubric. The gap between mention and substance — between picking up the buzzword and actually disclosing material AI activity — is the central editorial finding of this research.
This is the first issue of AI in Filings — Information Matters Report #IM102, the first of a quarterly cadence keyed off the SEC EDGAR primary-source layer. The full report — methodology in detail, every table and chart, the per-sector breakdown, and the analyst commentary — is in the downloadable PDF and on the Downloads page.
The five at the top of the rubric
Five S&P 500 companies score the maximum five out of five in their 10-K. The mix is the analytically interesting bit:
- NVIDIA — substrate cohort. Names Scale AI as a partner. Quantified $-figures throughout.
- AMD — names NVIDIA, OpenAI and Scale AI. Substrate-tier story, with the unusual disclosure choice of naming its largest competitor explicitly.
- Meta Platforms — quantified capex, named systems (Llama), named vendors, specific use cases, non-boilerplate risk language. The cleanest score-5 disclosure profile in the universe.
- ServiceNow — names Glean and NVIDIA as vendors. The most operationally explicit disclosure profile in the user-cohort tier.
- Cognizant — names Anthropic and NVIDIA. The most editorially valuable name on the list because it sits in IT services rather than substrate or platform — Cognizant is buying AI to resell as service capability.
Three pharma and industrial names sit one rung lower at score 4 and are worth tracking specifically: Pfizer (names Scale AI), Moderna (names OpenAI), Vertiv (names NVIDIA). Pharma AI disclosure is among the under-told stories in the broader market commentary.
The vendor disclosure paradox
Across 503 listings, our rubric matches a named external AI vendor in only 30 of the 397 filings that mention AI. The leaderboard:
| Vendor | S&P 500 filings naming it as a vendor |
|---|---|
| NVIDIA | 13 |
| Scale AI | 9 |
| OpenAI | 4 |
| Anthropic | 2 |
| Databricks | 1 |
| Hugging Face | 1 |
| Glean | 1 |
| xAI | 1 |
Sparser than the trade-press narrative around enterprise AI adoption would predict. Two things drive the gap. S&P 500 companies are not yet in the disclosure habit of naming AI vendors with the specificity they apply to traditional supplier disclosures — most disclosures stop at “third-party AI providers, including proprietary and open-source large language models.” And naming a vendor in a 10-K creates a documentation trail that complicates later procurement renegotiation. We expect this to ease as the EU AI Act high-risk-system phase-in compels naming for regulated deployments — most likely the Q3 2026 cycle.
The corollary editorial finding: the named-vendor cohort is the right place to look for the highest-conviction enterprise AI relationships. A company that names its AI vendor in a SEC filing has chosen specificity over silence.
Sectoral picture: IT dominates, four sectors zero out
Information Technology is in a category of its own: 67 per cent of IT 10-K filers cross the substance threshold against 6 per cent of Financials filers. Four sectors register zero substantive disclosures despite filing 10-Ks and mentioning AI — Real Estate, Consumer Staples, Utilities, and (very nearly) Energy. The absence of substantive AI disclosure does not mean the absence of AI activity. It means the disclosure cadence and the regulatory pressure to disclose are different in those sectors. Transcripts (v0.2) and possibly category-specific filings will surface what the 10-K narrative does not.
The sector with the largest gap between mentioning AI and disclosing it with substance is Financials — 71 of 71 Financials filers mention AI; only 4 cross the substance threshold. The Q3 2026 disclosure cycle is when the Financials cohort will compound, on our reading, given the convergence of EU AI Act enforcement and regulator-led model-governance scrutiny in the US.
The full sectoral table — eleven GICS sectors, average score, substantive count, and total-filed numbers — is in the PDF report.
Methodology, sources, and what this is not
A composite substance score from 0 to 6 (5 in v0.1.1; the sixth point — 8-K material event — deferred to v0.2). Each company earns one point for: a quantified $ figure near an AI keyword; a named internal AI system; a named external AI vendor; a specific use case; non-boilerplate risk-factor language; and an 8-K material event referencing AI. The score is the sum of components met. Components stay separately visible so analysts can drill in. The rubric uses word-boundary matching with a proximity constraint to avoid the substring collisions that systematically inflate AI signal in disclosure data.
Sources are SEC EDGAR (machine-readable, primary): the 10-K and 10-Q for each company filed in calendar 2026.
This is not a complete account of AI in the S&P 500. It is what the S&P 500 has chosen to disclose in its SEC filings. It is not a basis for investment decisions — Information Matters does not produce investment recommendations. The substance rubric is an editorial filter for analytical reading of public filings, not a buy or sell signal.
What you can do next
Read the full report (PDF, 10 pages) — every table, the methodology in full, the per-sector breakdown. Hosted on this site; also on the Downloads page.
Subscribe to The Brief — Friday-morning ~20-minute read. The Q2 2026 issue of AI in Filings lands 17–28 August, alongside the Q2 IM Agentic AI Market Outlook.
Write to us if you find a methodology error, a per-company misread, or a vendor or system we should add to our lookup tables. Corrections are part of the analyst record.
