Themes

Exposure by the company's own filings— how much of a 10-K business description is spent on a theme, measured against a versioned term list and gated on recurrence across consecutive years. A tier describes what a company says it does, not what its stock will do; the word "beneficiary" appears nowhere on this page by design.

Artificial intelligence

lexicon 2026-08-03.1 · latest 10-K 2026-08-03

pure-play

significant

156

mentions

1,696

risk-only

1,359

Value chain

Where each scored name sits in the theme's supply chain — a hand-curated taxonomy (v2026-08-04.0-draft), not a model output. Only pure-play and significant names appear: a chain position is a claim, and "mentions" isn't enough filing evidence to make it.

Chip design

Designs the processors and IP that AI computation runs on, including the EDA tools chips are designed with.

NVDA 74AMD 57AVGO 28QCOM 49AMBA 109CEVA 67SYNA 27GSIT 26MBLY 29CDNS 46SNPS

Fabricationempty

Manufactures the chips. Empty by construction: the fabs file foreign annual reports, not 10-Ks, so the corpus cannot score them.

Equipmentempty

Builds the tools fabs manufacture with. Empty because these companies' own filings describe semiconductor equipment, not AI — their exposure by their own words is below the significant bar.

Memory & networking

Moves and stores the data AI training and inference depend on.

ANET 36NTAP 35

Cloud & datacenter

Operates the compute AI workloads run on, from hyperscale clouds to AI-pivoted datacenter operators.

AMZN 35MSFT 39GOOG 60ORCL 52IBM 28HPE 38DOCN 37RXT 54KD 45

Power & coolingempty

Supplies the electricity and thermal infrastructure datacenters consume. Empty for the same reason as equipment: these businesses do not describe themselves in AI terms.

Appliers

Builds AI into software and services sold to end customers — the widest layer, and the one where 'exposure' most needs the tier qualifier next to it.

AI 180NOW 76CRM 42ADBE 66PLTR 42SNOW 45PATH 52PEGA 72SOUN 93

Tiered names

Every name whose exposure clears a recurrence-gated bar, densest first. The 1,696 "mentions" and 1,359"risk-only" names are counted above but not listed — a tier that weak describes a paragraph, not a business.

Ticker	Tier	Density	Recurrence	Latest 10-K
AI	pure-play	180.4	3/3	2026-06-24
GDYN	pure-play	109.5	3/3	2026-03-05

Density is weighted theme-term hits per 10,000 words of 10-K Item 1 (Business), from the most recent filing; multi-word phrases weigh full, the bare "AI" token half, and modal language ("we may explore…") is discounted. Pure-play requires density ≥ 70, significant ≥ 25, and both require the two most recent filings nonzero — so a single-year spike or a first-time filer caps at "mentions" regardless of density. Risk-factor mentions are counted separately and never scored: "risk-only" names discuss the theme as a threat, not a business. Scores are comparable only within lexicon version 2026-08-03.1.

Theme tiers are descriptive statistics of public filings — weighted term densities in 10-K business descriptions — and are provided for information only. They are not investment advice, not a recommendation to transact, and not a forecast: a company's decision to describe itself in a theme's words says nothing about its future returns. The one measured fact worth carrying here: specialized ETFs built on exactly this kind of thematic framing underperformed the market by roughly 30% risk-adjusted over their first five years after launch (Ben-David, Franzoni, Kim & Moussawi, "Competition for Attention in the ETF Space," Review of Financial Studies, 2023). Past performance does not predict future results.