How we calculate exposure

Updated September 11, 2026.

Published estimates of the S&P 500's AI weight currently sit about ten points apart, because each publisher draws the line in a different place. Here is ours, in full.

Computed from holdings, not tags

Every figure is a sum over a fund's actual holdings and their weights, refreshed each pipeline run. We do not hand-assign funds to themes — a fund's theme weight is whatever its holdings add up to, which is why the number can be checked against the fund's published constituents.

Pure-play vs broad

For a theme, a company can count fully (it is the theme) or partially (it has a meaningful segment). We publish a strict pure-play figure internally and a broad figure that includes partial members; the app shows one number per theme — the broad figure — because a single, honest number is more useful than two a reader has to reconcile.

Bars are scaled to the most-exposed fund

On a fund page, a theme bar is scaled to the highest exposure any ETF we track has for that row — so a full bar means no fund holds more. Size and geography each total 100%; theme rows overlap and do not add up (the Magnificent 7 sit inside AI; memory sits inside semiconductors).

Resolution and exclusions

A fund is included only when at least 90% of its weight resolves to identifiable companies. A leveraged fund holding swaps, or a fund-of-funds holding other ETFs, will not resolve and is excluded rather than shown with a misleadingly low number.

What we publish and what we do not

These are computed aggregates — percentages, counts and ranks. Underlying holdings, weights, prices, fees and yields are licensed and are not published here.

What each row means

Size

Every holding is placed in large, mid or small by market capitalisation and recomputed nightly as prices move — so a fund's size mix drifts on its own, without the manager changing anything, which is why there is no one-year column here.

Geography

Every holding is placed in US or international by its country of domicile and recomputed nightly. A fund's geographic mix drifts as prices move, so there is no one-year column here.

Tech

Tech is the data provider's broad Technology sector, computed from every holding — no hand-tagging and nothing to maintain. It is wider than a pure-IT screen, so a fund's Tech figure will usually sit above its AI or semiconductor figure.

Nvidia

Nvidia is a single company rather than a theme, so the figure is that one holding's weight in the fund. It is computed the same way as every other row — from the fund's actual holdings — and shown on its own because one position can carry more of a fund than an entire theme row.

Magnificent 7 · The ETF Copilot Magnificent 7 List — 8 companies

The Magnificent 7 is a fixed list — Nvidia, Microsoft, Apple, Amazon, both Alphabet share classes, Meta and Tesla. Because it is a closed set, this figure is exact and checkable.

AI · The ETF Copilot AI List — 54 companies

A holding counts toward AI when the business builds or supplies the technology itself — chip designers and foundries, memory and networking, hyperscale cloud, and the model and software makers. A company that merely uses AI tools internally is not counted. Adoption is not exposure, and conflating the two is the biggest reason published figures disagree.

Semiconductors · The ETF Copilot Semiconductor List — 502 companies

Semiconductors starts from companies whose primary industry is chip design or manufacturing, then adds diversified firms with a major semiconductor segment and removes distributors and general contract manufacturers.

Memory · The ETF Copilot Memory List — 54 companies

A holding counts toward Memory when it designs or manufactures memory and storage — DRAM, NAND flash and the equipment and controllers behind them. Companies that merely buy memory as a component are not counted.