Board/One of 22

The AI compute core

The chips, memory and data centers AI runs on, and the big technology companies that build and buy them.

The AI compute core is one of 22 market stories StoryVector measures. 396 US ETFs carry it, with $682 billion between them; the largest holder, VGT, holds 23.8% of its capital. This week it ranks tied 4th of 22 for attention, in the WARM band. Over the past year its ETFs returned +50.2%, against +16.8% for broad US ETFs, to October 2, 2026; past performance, not a recommendation.

57.0WARM level, 4 of 22
last four weeks
396ETFs carry it
68.5%held by the top five
VGTlargest holder, 23.8%
$682 bncapital, 5th of 22

Where it sits this week

Its level among all 22 on the fixed 25 to 60 scale. The band is where the level sits this week, not a direction.

25QUIET below 41.4HOT from 57.460
What its ETFs returnedFRI OCT 2market price total return

Its ETFs ranked among the narratives, lowest to highest. The white notch is this one; the thin tick is where broad US ETFs would sit. Measured and past, not a forecast.

+1.23%Friday · US +0.79%lowest −1.17%highest +3.50%7th of 22
+9.5%Last month · US +0.8%lowest −5.8%highest +15.1%2nd of 22
+47.8%This year · US +14.6%lowest −3.8%highest +47.8%1st of 22
+50.2%Last 12 months · US +16.8%lowest −37.2%highest +50.2%1st of 21
0.22% cost to own, a year, weighted · largest holder VGT 0.09%Every period and the 10-year chart ↓

How it is talked about

What the story is about, where its attention comes from, what sets it off, and what it shares ETFs with.

The argumentsigned off

AI spending is huge and still growing. What is it evidence of?

Demand

There is not enough computing power. The orders and the shortages are real, and the money is a response to demand that already exists.

Financing

The build has moved from cash to debt, and sellers now lend to their own buyers. The money says as much about how it was raised as what it bought.

What would separate them. Who pays for the build. When a chipmaker invests in a company that buys its chips, the first reading sees a partnership and the second sees a seller funding its own customer. Same deal, two readings of what the revenue means.

Both readings use the same facts, which is what makes this an argument and not a dispute about data. We measure the attention and the capital, not which reading is right.

Where the attention comes from
Forums65.2
Video61.6
News59.0
Reference pages54.4
Search53.7
Investor message boards48.2
057.0 combined100
What sets it off
AI compute and semiconductor results23 stories
Who carries the AI buildout's credit risk9 stories
Power supply for the compute buildout6 stories
ETF launches and liquidations6 stories
Index levels and records5 stories
Wire stories filed to this narrative since August 24, by the kind of event. A count of stories, not of outcomes.

Who owns it

396 ETFs carry this narrative; 344 have it as their main one. Here is how its capital divides among them.

Its slice of all 22
8.4%
$682 bnof the $8.1 tn across all 225th by capital
How its capital divides
top five hold 68.5% · 391 other ETFs hold 31.5%
ETFShare1 yearCost
1VGT23.8%+36.0%0.09%
2XLK20.5%+40.4%0.08%
3SMH12.1%+87.0%0.35%
4SOXX7.8%+109.7%0.33%
5IYW4.3%+36.0%0.37%

What its ETFs returned

The asset-weighted return of the unlevered ETFs that carry this narrative, against broad US ETFs. Measured and past, not a forecast.

FRI OCT 2market price total return
1M3MYTD1Y3Y*5Y*
The AI compute coreits ETFs, unlevered+9.5%+8.7%+47.8%+50.2%+41.6%+24.5%
rank among the narratives2nd of 226th of 221st of 221st of 212nd of 211st of 21
The marketbroad US ETFs+0.8%+2.6%+14.6%+16.8%+22.9%+12.9%

1M, 3M, 1Y, 3Y, 5Y: one month to five years. YTD: year to date. * A year, annualized. Rank is among the narratives with a full record for the period. A dash means the narrative’s ETFs do not cover the whole period.

$10,000 invested October 31, 2016

The AI compute core $105,992The market $42,209
$0$25k$50k$75k$100k$125kOct 2016Oct 2018Oct 2020Oct 2022Oct 2024Oct 2026
$0$25k$50k$75k$100k$125kOct 2016Oct 2020Oct 2026

Year by year

The AI compute coreThe market
−40%−20%0%20%40%60%2019 +50.4%2019 +31.3%20192020 +46.8%2020 +21.4%20202021 +32.8%2021 +27.1%20212022 -31.2%2022 -19.8%20222023 +57.1%2023 +27.9%20232024 +27.5%2024 +23.1%20242025 +29.0%2025 +16.8%2025YTD +47.8%YTD +14.6%YTD
−40%−20%0%20%40%60%2019 +50.4%2019 +31.3%20192020 +46.8%2020 +21.4%20202021 +32.8%2021 +27.1%20212022 -31.2%2022 -19.8%20222023 +57.1%2023 +27.9%20232024 +27.5%2024 +23.1%20242025 +29.0%2025 +16.8%2025YTD +47.8%YTD +14.6%YTD

Total return on the market price, each distribution reinvested on its ex-date. The narrative is the asset-weighted return of the unlevered ETFs that carry it; the market is the asset-weighted return of broad US index ETFs. Past performance is no guarantee of future results.

What is inside it

9 categories, each claimed by a written rule. The bar is each one’s share of the capital. The table names what decides each.

Broad technology wrappers 57.1%Semiconductors 19.7%Levered and inverse wrappers 9.7%Communication services 4.6%Memory 3.8%4 more 5.0%
CategoryETFsCapitalShare of the ETFsShare of the capital
Broad technology wrappersWhether the build is a story inside technology, or the thing technology currently is35$390 bn10%57.1%
SemiconductorsWhether the chip tier is where the constraint sits and where the return on it is booked15$135 bn4%19.7%
Levered and inverse wrappersHow fast an issuer can list a wrapper against how fast capital arrives in one223$66.3 bn65%9.7%
Communication servicesWhether the largest buyers of AI compute earn back what they spend on it4$31.3 bn1%4.6%
MemoryWhether high-bandwidth memory is the part that runs short6$26.2 bn2%3.8%
AI thematic wrappersWhether the story's return shows up in one tier or across the largest names generally38$25.0 bn11%3.7%
Everything else7$5.8 bn2%0.8%
The physical plantWhether the binding constraint sits in the chip or in everything built around it10$2.8 bn3%0.4%
Photonics and optical interconnectWhether the limit on a training cluster is the parts or the connection between them6$0.9 bn2%0.1%

The widest gap: levered and inverse wrappers, 65% of the ETFs and 9.7% of the capital.

How the market built it

What issuers have listed with this as the main narrative, by wrapper.

121Unlevered long90.3% of the capital
184Levered long9.3% of the capital
39Inverse0.4% of the capital

What issuers have listed, not what anyone expects. A wrapper can be listed before anyone wants it, so a count of ETFs and a share of capital answer different questions.

Common questions

Which ETFs give exposure to the AI compute core?

396 US ETFs carry it. The largest holders of its capital are VGT (23.8%), XLK (20.5%) and SMH (12.1%).

Which ETF holds the most of the AI compute core?

VGT, with 23.8% of the capital in ETFs carrying the story; the top five hold 68.5%.

How loud is the AI compute core right now?

Its attention level is 57.0 on a scale of 25 to 60, tied 4th of 22 this week, in the WARM band. The level averages six channels: search, video, news, forums, investor boards and reference pages.

How has the AI compute core performed?

Asset weighted, its ETFs returned +50.2% over the past year, against +16.8% for broad US ETFs, to October 2, 2026. Past performance only.

Run an ETF in the AI compute core? StoryVector shows issuers where their ETFs stand against the closest rivals, every week.

Request a pilot
See all 22 ranked on the board →
Every figure on this page is this week’s. Returns are market price total return as of October 2, 2026, measured and past, not forecasts. The level is attention on a fixed 25 to 60 scale across six channels; the band is where it sits among the 22. Ownership is Capital Share of Narrative, exposure weighted. Capital is the assets of the ETFs with this as their main narrative; ownership shares are measured on exposure-weighted capital, a different total. ETFs are listed by measured share; placement cannot be bought. Nothing here is investment advice. How the measurement works →