Research/Research note

Every AI bottleneck gets an ETF

StoryVector · October 28, 2026

Demand for AI data centers is outrunning supply, from memory chips to the transformers that connect them to the grid. Each gap has drawn its own ETF.

This year 37 AI chip ETFs have launched, more than twice last year's 16, and 28 of them bet on a single slice. Most opened after chip stocks peaked, and nearly all have lost money, though the shortages are far from over.

37AI chip ETFs launched in 2026, against 16 in 2025
28 of 37opened after chip stocks peaked on June 22
32 of 34launched after the first memory fund are below their first price

The shortages are real

Micron says its new Taiwan plant's 2027 output of AI memory is already sold out. Large power transformers can take four years to arrive. Forecasters expect most of these gaps to last into 2027 or beyond.

The shortages behind the funds

Where demand is running ahead of supply, and the ETFs built on each.

ShortageWhat forecasters sayETFs in 2026
Memory chipsMicron says its new Taiwan plant's 2027 output of AI memory is already sold out. Big new supply arrives from 2027.7
Optical partsThe fastest optical links could stay 30% to 40% short of demand through 2029.6
Data centers, power and coolingLarge power transformers can take up to four years to arrive, and their prices are up about 80% in five years, pv magazine reports.3
CapacitorsOrders at the biggest makers are the strongest since the pandemic, and the fourth quarter could decide whether supply tips into a full shortage.3
Chip substratesThe base layer each AI chip sits on is set to swing into shortage this year, with the gap widening in 2027 and 2028.1 cleared

A fund for every shortage

The funds followed the shortages, one at a time. Memory went first: Roundhill's DRAM opened in April and became the fastest ETF ever to reach $6.5bn, and six more memory funds followed. Then came photonics, data centers and power, chips by country, inference chips, capacitors and, in September, chipmaking. Tuttle Capital has 14 more AI funds cleared to launch, among them AI Substrate and AI Test Equipment.

AI chip ETFs launched in 2026, by slice

One dot per fund, sized by its assets. Hollow dots: Tuttle Capital's 14 cleared funds. Hover or tap for detail.

launched, a narrow slicelaunched, broad AI and big techcleared October 7, not yet in our data

The shortages kept going. The stocks did not.

Spending has not slowed: the four biggest cloud companies plan $720bn to $745bn this year, after three of them raised their plans this summer. Chip stocks peaked anyway. SMH, the biggest chip ETF, rose 79% to June 22, then fell 21%. One reading: by June the good news was in the price.

The funds came after. 28 of the 37 opened after the peak, most in August. Each slice arrived later than the last: memory 81 days before the peak, capacitors 65 days after, chipmaking 80 days after.

Chip stocks, and AI chip ETF launches, in 2026

SMH, $100 invested on January 2. Bars: launches each month. Hover or tap for any day.

SMH, total returnAI chip ETFs launched that month

How the slices have done

Badly, mostly. In every narrow slice the typical fund has trailed SMH since it opened: the memory funds after DRAM are down 25%, photonics funds 28%. Only the broad AI and tech funds beat it. The later slices also drew little money; most hold a few million dollars or less.

Each fund since launch, against SMH

The dot is the fund; the white mark is SMH over the same days. Hover or tap a row for detail; select a ticker to open its page.

the fund, total return since its first daySMH over the same days

Could they still pay off?

Possibly. If the shortages last and chip stocks recover, narrow funds should gain more: memory funds have moved about 1.3 times as much as SMH.

How far each kind of fund moves with SMH

The typical fund, for each 1% SMH rose or fell in a week.

Memory fundsup 1.1% when SMH rose 1%down 1.6% when SMH fell 1%
Photonics fundsup 0.8% when SMH rose 1%down 1.7% when SMH fell 1%
Broad AI and big techup 0.4% when SMH rose 1%down 0.5% when SMH fell 1%

What to watch

Watch the supply, not just the demand. A fund built on a shortage is a bet on how long it lasts.

Notice when the slices multiply. A run of ever-narrower funds can mean the story has already run.

Check the size. Small funds can cost more to trade and are more likely to close.

The bottom line

AI's shortages are real, and every one now has an ETF. But most of this year's funds arrived after chip stocks peaked, and in every narrow slice the typical fund has trailed the broad chip ETF. The shortage and the fund are not the same bet.

Notes and method

Funds: unleveraged ETFs our taxonomy places in the AI compute story by their holdings, first traded in 2026; slices are the theme each fund names, grouped by us. Data at the close on October 23, 2026 (XIGV, which did not trade that day, at October 22). Returns are total returns, against SMH over the same days; the sensitivity table uses weekly returns and the median fund in each group. Sizes are from the issuers where published, otherwise market data. Sources: Yahoo Finance, May 13 (DRAM); SEC EDGAR (Tuttle funds, cleared October 7, none trading in our data by October 23); TMT Finance, August 18 (cloud spending); Micron, October 16, and Counterpoint Research, October 22 (memory); McKinsey, 2025 (optics); PwC and pv magazine USA, May 11 (power); TrendForce, July 6 (capacitors); DigiTimes, August 17 (substrates). All dates 2026 unless shown.