ISSUE #004 · WEDNESDAY, JULY 22, 2026
Copper moved higher as Chinese import demand strengthened and available metal outside the US tightened. LME inventories fell again, while COMEX stocks kept climbing as tariff uncertainty continues to pull copper into US warehouses.
The base copper story remains stable. The question this week is how much of the price is being set somewhere else entirely.
The Copper Thesis Doesn’t Need AI. The Copper Price Might.
01 THE SETUP
Two headlines caught our attention this week.
First, copper is arguably trading as a quasi AI proxy. Goldman Sachs’ latest model found AI-linked expectations have overtaken Chinese growth outlook, the dollar, and physical tightness to become the single largest contributor to copper’s cumulative price gains since early 2025.
Second, that bet is being placed as the durability of the AI build-out is being openly questioned by people who aren’t short-sellers. UBS sees hyperscaler capex growth decelerating from 76% this year to 25% in 2027 and just 6% in 2028 — and that some investors are already trimming the AI-infrastructure trade.
This comes just weeks after the Bank for International Settlements’ annual report flagged the five largest hyperscalers spending over $1 trillion on AI capex across 2025-26, commitments “outpacing earnings and the free cash flow of these firms,” and warned a disappointment in returns “could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust.”
In other words, copper has increasingly hitched its marginal price to AI sentiment at the same time that exact sentiment is looking most exposed.
02 WHY IT MATTERS
If AI expectations are the marginal price-setter, then anything that dents those expectations can unwind the rally increment without a single tonne changing hands in a warehouse.
And the trade is priced for acceleration, not just growth. That’s the trap in the UBS numbers: capex is still rising through 2028, but the rate collapses from 76% to 6%. A price re-rated on an accelerating narrative can correct hard when the narrative merely flattens. We don’t need a hard BIS-style outright bust for the AI-copper premium to deflate.
So a sharp, AI-driven correction in the copper price at some point is entirely plausible.
The question is, if it arrives, do we read that correction as a verdict on the thesis?
03 THE DECODE
A lot of coverage tends to overemphasize AI’s impact on a much broader copper story.
Even S&P Global — whose January report is literally titled “Copper in the Age of AI” — walks it back once inside the document: “While AI is creating a new vector of copper demand, it is not the largest by any means.” The title, they explain, is about what AI underlines — the foundational role of expanded electricity supply — not what it directly consumes.
Get the number right (it’s smaller than the headlines imply):
IEA’s 2025 Critical Minerals Outlook forecast that “copper use in data centres could range from 250 kt to 550 kt in 2030, equating to 1-2% of global copper demand, though this could be even higher depending on the speed at which demand for their services picks up.” Last week, the 2026 edition dropped. It didn’t swap that figure for a bigger one.
Of course, that’s not to say data centres are trivial. IEA’s 2025 estimates are for direct copper consumption within data centers — the bulk of the weight sits outside the building. As Wood Mackenzie’s June research notes, once you add the on-site power systems, transmission, and distribution, total system-level consumption reaches “an estimated three to four times the volume implied within the asset.”
Combining these two, we might reasonably estimate a total demand from AI ranging anywhere from 750 kt to 2.2 Mt by 2030. Incidentally, this back-of-the-envelope estimate lines up quite well with older numbers, including the oft-cited Sprott forecast of 638,405 kt to 1.9 Mt by 2030 with a ~1.1 Mt base case.
The bigger demand is real — but it’s a wider grid story:
Set the whole AI-and-data-centre call — direct plus a generous slice of grid — against the broader copper deficit.
Near-term, the deficit is modest. In IEA’s 2026 report shows the mine pipeline base case nearly meeting requirements at 2030 — the gap is on the order of ~2 Mt. So a genuine AI collapse — a total halt to the build-out, which nobody is actually forecasting — would meaningfully soften the 2030 picture and plausibly bring the base case close to balance.
However, by 2035, IEA has existing and announced mines covering only about 75% of primary requirements — a ~25% structural shortfall of which AI is a relative dent. By 2040, copper adds about 7 million additional tonnes of demand and the hole is deep and structural.
IEA’s 2024 Outlook acts as a sanity check against these assumptions. In 2024, the phrase “data centre” had yet to enter the conversation. Yet, the IEA had already found announced copper projects only met only about 70% of copper requirements by 2035. In other words, the deficit was fully diagnosed a full year before AI entered the frame.
The part investors might underweight:
If hyperscaler capex growth decelerates, the risk isn't to copper's deeper fundamentals but to the narrative contributing to its current price. A sentiment unwind could drive a sharp correction.
That same slowdown could potentially also take some genuine pressure off the 2030 balance, which is thin enough for data-centre demand to matter. So the near-term case for a pullback is real.
However, the broader copper thesis — the larger structural deficit into 2035 and beyond — remains without AI, and it moves on the timescale of mine permits rather than earnings calls. An investor who reads a near-term AI correction as a verdict on copper risks confusing the two.
04 WHAT TO WATCH
Now → next earnings season: Hyperscaler capex guidance is the cleanest read on the trade. UBS already has 2027-28 growth decelerating; watch whether Q3 commentary confirms the glide path or re-accelerates.
The tape that separates trade from thesis: Watch whether a risk-off move in AI equities actually drags physical copper with it, or whether ex-US tightness and the COMEX-LME spread hold firm regardless. If AI stocks wobble while the physical market stays tight, current prices may be less AI-dependent than Goldman’s model estimates.
2027–2035: The thesis test proper. Watch broad grid and electrification demand — the real load-bearing column.
The most important new long-range copper assessment of the year.
Codelco’s new chairman concedes its 1.7-million-tonne 2030 ambitions are in doubt. A look at what declining grades and troubled megaprojects mean in practice — enormous investment can amount to replacement supply rather than growth.
BMI argues that project delays, weak mine growth and rising costs could eventually push copper toward $17,000/t.
That's the wire for today. Until next week, mind the difference between a correction and a cancellation.
Know someone who keeps asking you about copper?
They can subscribe free at copperwire.news.