Data Center Funds: What the Biggest One Actually Holds, Read From Its Own Filing

A named fund, taken apart from its own Schedule of Investments. Less than half of the Global X Data Center & Digital Infrastructure ETF is data center owners — 28% is cell towers and 22% is semiconductors — and the index methodology explains exactly why it has to be.

Search for a data center fund and you will find several products whose names contain the words "data center". The useful question is not which one to buy. It is what is inside one, and whether the thing inside matches the reason you went looking.

So this page does one thing properly: it takes the largest fund with "Data Center" in its name and reads its holdings out of its own regulatory filing, to the dollar. Then it explains, from the index rulebook, why the answer could not have been otherwise.

The short version: less than half of it is data centers.

The fund

The Global X Data Center & Digital Infrastructure ETF (DTCR) launched on 27 October 2020, holds 25 positions, and charges a total expense ratio of 0.50% (fund fact sheet). It tracks the Solactive Data Center REITs & Digital Infrastructure Index.

Its Schedule of Investments at 31 August 2026 reports total common stock of $2,138,751,543, which the filing puts at 99.7% of net assets (Q3 complete holdings). Every figure below is computed from that filing.

What it holds, by what the company actually owns

Those four groups add to the filing's total exactly, with no remainder. The grouping is ours; the dollar amounts are the filing's.

• Data centers (41.7%): Digital Realty Trust, Equinix, Keppel DC REIT, NEXTDC, GDS Holdings, Vnet Group, SUNeVision.
• Cell towers (28.2%): American Tower, Crown Castle, SBA Communications, China Tower.
• Semiconductors and hardware (22.3%): NVIDIA, Broadcom, AMD, Intel, Micron, Marvell, Microchip, SK Hynix, Taiwan Semiconductor, Super Micro Computer, Hewlett Packard Enterprise.
• Arguable (7.8%): Applied Digital, Whitefiber, Uniti Group. We put these aside rather than count them in our favour — the first two build and operate AI compute facilities, the third owns fiber. A reader could fairly call the first two data centers; we did not, because the point of the exercise is to avoid classifying things to suit the argument.

Which holding is the largest depends on the day you ask, and the answer crosses the category line. In the 31 August filing the top three are Digital Realty at $273.2 million, American Tower at $270.1 million and Equinix at $265.3 million — a data center company first, by less than 1%. On the fund's own live holdings page dated 4 November 2026, American Tower showed 15.15% against Digital Realty's 13.76%, so a cell tower company had moved to the top.

Ten weeks apart, the largest holding in a fund called "Data Center" went from a data center to a cell tower, without anybody changing anything.

Why it has to be this way: the index rulebook

This is the part that makes the fund's composition predictable rather than surprising, and it is published.

The index methodology summary states the selection test — the emphasis is ours:

"The index is designed to track performance of companies that have business operations in the field of data centers and/or cellular towers"

and requires that constituents "[g]enerate at least 50% of their revenues from data center and/or cellular tower related business operations". It then lists what the index may include:

"i. Own, operate, and or develop data centers (including data center REITs) ii. Own, operate, and or develop cellular towers (including cellular tower REITs) iii. Manufacture the servers and/or other hardware often used in data centers and cellular towers, including semiconductors, integrated circuits, and processors"

(index methodology summary).

So a pure cell tower company satisfies the revenue test completely, on towers alone. Nothing in the rules requires any minimum data center exposure.

The sentence that explains the 22.3% of chips

Then comes the mechanism, and it is worth reading twice:

"All Data Center Companies and Cellular Tower Companies are included as index constituents, up to a limit of 40. If there are fewer than 25 constituents, Digital Infrastructure Hardware Companies are included until the index reaches 25 index constituents"

Now count the filing. Data center owners and tower companies together: 11 names. The index needs 25. The fact sheet confirms the fund holds exactly 25.

The chipmakers are not a view. They are a quota being filled. There are not 25 listed companies that own data centers or towers, pass a $200 million market cap screen, pass a $2 million daily-volume screen and sit in an eligible country — so the rulebook tops the index up with hardware manufacturers until it has 25 names. That is why a fund called "Data Center" holds NVIDIA, Intel, Micron and Taiwan Semiconductor.

And the weighting rules show the design intent clearly:

A data center company and a tower company get the same 12% ceiling — the index does not prefer one over the other. And the 2% hardware cap limits each chipmaker, not chipmakers in aggregate: eleven names at up to 2% each is how 22.3% of the fund ends up in semiconductors without any single rule being broken.

One loose end, stated as the open question it is: American Tower read 15.15% on 4 November, above the 12% cap. The index "follows a semi-annual rebalance schedule", adjusted on the last business day of January and July, so the most likely explanation is simple drift since the July rebalance — a 12% cap is applied at rebalance, not maintained daily. We have not confirmed that with the index provider, so treat it as the probable reading rather than a fact.

What this means for the decision you are making

If you wanted exposure to data center real estate, this fund gives you less than half of that, and the other half is a bet on cell towers and semiconductors. Both are real businesses; neither is the one you searched for. Towers and data centers are often lumped together as "digital infrastructure", but a tower is a steel structure with a long lease to a mobile carrier and almost no power problem. Data center economics turn on kilowatts and interconnection; tower economics do not.

If you wanted exposure to the AI buildout broadly, this is a reasonable instrument and the semiconductor sleeve is a feature. But then compare it with a cheaper broad technology fund before paying 0.50% for a 25-stock portfolio.

If you wanted exposure to a specific building or project, a fund cannot give you that at all. A listed fund owns shares in large landlord companies; it is the opposite end of the spectrum from a single-asset private deal. That trade-off is the whole of REITs vs private deals, and what the liquid route dilutes is set out in AI data center REITs.

What we could not check, and will not pretend to

• The other funds in this category. paceretfs.com returned HTTP 403 to this machine, so the Pacer Data & Infrastructure Real Estate ETF (SRVR) is not analysed here. We saw commercial sites quoting its weights; we did not use them, because the issuer's own filing is the source that matters and we could not open it. One fund read properly is worth more than three read from aggregators.
• Private and non-traded data center funds, where the minimums run to hundreds of thousands or millions of dollars and the holdings are not published daily. Those are a different instrument with a different disclosure regime — see non-traded REITs for what that regime costs, and exit and liquidity for why getting out is the hard part.
• Whether any of this is a good investment. Nothing here is a recommendation. The point of the page is that the composition is a fact you can check in twenty minutes, so you should check it rather than infer it from a name.

Five checks on any fund with "data center" in its name

1. Open the issuer's own holdings file, not an aggregator's page, and write down the as-of date.
2. Group the holdings by what the company owns — data centers, towers, fiber, chips, servers — and add up each group as a share of the total. The name of the fund is not evidence about this.
3. Read the index methodology summary. It tells you what the fund is allowed to hold, which is more durable than what it holds today. Look specifically for an "and/or" in the theme definition and for any rule that pads the constituent count to a minimum.
4. Check the per-name weight caps against the aggregate. A 2% cap per company is not a 2% cap on the category.
5. Check the overlap with what you already own, then compare the fee with a broad fund that holds the same large names.

The general version of this method — and the naming problem across the whole category — is in data center ETFs and funds. What a fund cannot give you, and what the private routes ask in exchange, is in how to invest in AI data centers.

For the engineering of what these companies actually build, the free vendor-neutral reference is *The Definitive Guide to AI Data Centers*. They cover how it is built; we cover whether to put money into it.

Nothing here is investment advice. See how this site is funded.