Copper Is the AI Trade That 13F Filings Barely Show

Copper settled at an all-time high on COMEX in early August 2026 — $6.70 per pound — with the LME print a day later at $6.56. That is not a squeeze in a thin contract. It is the market repricing the one physical input that every version of the AI buildout requires.

And yet if you go through the Q2 2026 13F filings looking for the institutional copper trade, it is surprisingly hard to find. A handful of macro-influenced books added Freeport-McMoRan and uranium names. Most large equity managers did not. The materials sector remains one of the thinnest allocations across the funds we track — a pattern that shows up in the sector view of our holdings data.

That gap between the commodity and the positioning is the story.

Why copper became an AI input

The connection is less abstract than it sounds. Data centers do not just consume power — they consume the infrastructure that delivers it. Connecting a 100MW-plus facility to a high-voltage grid means substations, transformers, switchgear and busbar, all of which are copper-intensive in ways that never showed up in prior technology cycles.

The projected numbers are still small as a share of global demand: data centers ran about 0.15% of copper consumption in 2024, and forecasts put them near 1% by 2030, averaging roughly 400,000 tonnes a year over the next decade with a peak around 572,000 tonnes in 2028. One percent of a market does not usually move prices.

It matters here because it lands on top of a market that is already short. The International Copper Study Group sees a 150,000-tonne refined deficit in 2026. UBS models more than 400,000 tonnes. Wood Mackenzie expects 2026 to exceed the 304,000-tonne shortfall of 2025. In a balanced market, incremental data-center demand is noise. In a market already running a deficit, it is the marginal tonne that sets the price.

Layer in the slower structural drivers — EVs, renewables, grid replacement in the US and Europe, and now defense spending — and S&P Global's January 2026 study concluded the shortfall widens rather than closes.

Why the filings lag the trade

There are three reasons institutional copper exposure looks light in the data, and only one of them is a real signal.

First, 13Fs do not capture commodities. A fund expressing a copper view through futures, physical, or an offshore swap does not report it on a 13F. The form covers US-listed equities and certain derivatives on them. A macro fund can be structurally long copper and show nothing in the filing. This is the same blind spot we wrote about with defense exposure and with gold earlier this summer.

Second, the data is old. Q2 filings landed on August 14 and describe positions as of June 30. The record copper prints came in August. Any fund that bought the breakout will not be visible until the Q3 filings on November 14. September through mid-November is the annual dead zone where 13F watchers are working from the most stale data of the year.

Third — and this is the part that is actually informative — the miners are not the commodity. Freeport-McMoRan reached 15-year highs on the back of record copper plus a $1.25/lb US premium from tariff clarity, worth roughly $1.7 billion annually on its US sales. That premium is a policy artifact, not a mining outcome. A manager who wants copper exposure without a bet on trade policy, Indonesian permitting, or Grasberg operations has good reasons to skip the equity.

What to actually watch

The useful question is not "who owns copper" but "who was already positioned for the physical constraints of the AI buildout." That is a wider net than the miners. It includes the utilities and grid equipment names that showed up as the top-bought sector in Q2, the electrical equipment suppliers, and the industrials that hedge funds rotated into as they trimmed semiconductors.

Seen that way, the copper trade is not missing from the filings. It is expressed one layer downstream, in companies that sell the copper-intensive equipment rather than the metal. Comparing which managers hold both ends of that chain is the kind of thing the overlap tool is built for.

The November 14 filings will be the first real read on whether large equity books chased the August breakout or stayed downstream. Until then, be skeptical of anyone claiming to know what the smart money did in copper this summer — the filings that would show it have not been written yet. If you want to track which managers have historically leaned into commodity cycles early, the investor pages hold the position history, and the macro consensus view aggregates the asset-class signal across funds.

---

Data sourced from public SEC 13F filings. Educational research only — not investment advice.