Tearing apart the COMEX-LME copper inventory split

Copper.

The wires are running "squeeze unwinds" headlines this week. Trafigura dumped 20,000-plus tonnes onto LME warrants in August and the cash-to-three-month backwardation collapsed from $545 to $248. CNBC brought out someone in a tie to explain that the copper panic is over. Markets can relax.

They're looking at the thermometer. The patient's fever broke for forty-eight hours. Nobody checked whether the infection cleared.

Here's what actually happened. Since April, tariff arbitrage has been hoovering physical copper out of every non-US warehouse on the planet and stacking it inside COMEX-approved vaults. As of September 15, COMEX holds 768,098 short tons in total inventory — roughly 696,000 metric tonnes. That's a record. Meanwhile, LME on-warrant stocks cratered to approximately 90,000 tonnes in early August before Trafigura's emergency delivery bumped them back up. The rest of the world didn't get more copper. America just locked it in a room.

The mechanism isn't a squeeze unwinding. It's a tariff-driven inventory dislocation splitting the global copper market in two. One side — the American side — is bloated with pre-positioned metal awaiting a Section 232 decision on refined copper tariffs expected September 28. The other side — everyone else — is running on fumes.

And ten days from now, when that decision lands, one of those two sides is going to be very wrong about where this metal belongs.

Numbers.

Goldman Sachs revised its 2026 copper outlook in June. In April, the model showed a 490,000-tonne global surplus. By June, they'd flipped it to a 640,000-tonne deficit outside the United States. That's not a tweak. That's a structural inversion driven almost entirely by tariff-driven stockpiling redirecting metal into US warehouses.

Grasberg. The world's second-largest copper mine went dark on September 8, 2025, after a wet material flood killed seven workers. Freeport declared force majeure on September 24. Benchmark Mineral Intelligence estimates 591,000 tonnes of lost copper output between the September 2025 shutdown and end of 2026 — roughly 2.6% of global mine production erased in a single event. Freeport has cut 2026 guidance by 35%. Full restart isn't expected until the end of 2027.

Chile is no better. Cochilco cut its 2026 production forecast for the second consecutive quarter to 5.27 million tonnes, down 2.6% from 2025. Codelco, Escondida, and Spence all underperformed in the first half. Structural restrictions — not bad luck. Ore grades are declining, water availability is constrained, and El Teniente's deep expansion is behind schedule.

And the concentrate market? Inverted. Spot treatment charges hit -$126.80 per tonne at end of June — meaning Chinese smelters are paying miners to take their ore. The 2026 annual benchmark was set at $0/tonne, the lowest on record. Antofagasta and Jiangxi Copper couldn't even agree on a positive number. China's top smelters responded by cutting output more than 10%.

The Trafigura delivery bought the LME a few weeks. It didn't fix the flow diagram. The pipes feeding the system — mine output, concentrate availability, refining capacity — are all running below the models that called for surplus twelve months ago.

The Section 232 decision. September 28. Ten days out.

The Commerce Department's recommendation: a phased universal tariff on refined copper of 15% starting January 2027, rising to 30% in 2028. Semi-finished copper products already face a 50% tariff since April 6. The market has front-run the outcome. Traders shipped hundreds of thousands of tonnes into COMEX warehouses, capturing the COMEX-LME premium that at one point hit $2,937 per metric tonne — an all-time record.

But here's the weak link nobody's pricing: what happens to COMEX stockpiles if the tariff doesn't land?

SocGen modeled it. The current COMEX premium over fully delivered LME metal implies only a 14.6% probability of the recommended 15% tariff by January 2027. The market has already started discounting it. If the White House delays or softens the tariff, that 696,000 tonnes of copper sitting in American warehouses becomes the most expensive metal in the wrong zip code. The arbitrage reverses. Metal flows back out — onto LME warrants, onto the global market. Backwardation collapses further.

Sounds bearish? Only if you're staring at the screen.

Because the physical reality hasn't moved. Grasberg is still dark. Chile is still underperforming. Treatment charges are still negative. ING's 600,000-tonne refined deficit doesn't include the Grasberg loss. I watched a related dynamic in 2021 — Trafigura withdrew metal from the LME, backwardation went to extremes, and the exchange imposed emergency rules to cool it. Everyone called the top. The temporary fix made the structural problem invisible just long enough for the next leg up.

The machines that shorted the LME backwardation collapse read the Trafigura delivery as a resolution. They didn't call the smelters in Jiangxi. They didn't pull up the Cochilco revision. They have not, to my knowledge, checked whether Freeport has identified the seventh body at Grasberg yet.

Two scenarios. Same underlying deficit. Different triggers, different timing.

If the tariff hits, the 696,000 tonnes in COMEX vaults become a strategic American stockpile, not a tradeable buffer. The rest of the world's refined copper market tightens faster than any model currently projects. LME stocks — already fragile — bleed toward levels where delivery queues start breaking. Physical premiums in Rotterdam and Shanghai decouple from the exchange price. I've seen that decoupling before. It's not gentle.

If the tariff is delayed or watered down, the short-term picture softens. Metal refluxes out of US warehouses. LME backwardation cools. Headlines declare victory. But 591,000 tonnes of Grasberg copper is still missing. Chilean output is still falling. Chinese smelters are still cutting refined production because they can't source concentrate without paying the miners for the privilege. The structural deficit doesn't care about tariff timing. It reasserts the moment the arbitrage unwind is done.

Where does capital go? Not into broad commodity ETFs diluted with iron ore and coal. The leverage is in unhedged copper miners producing outside the disruption zones — companies whose revenue maps directly to spot. Lundin Mining. Ivanhoe's Kamoa-Kakula, with its own smelter now absorbing DRC concentrate internally. First Quantum, if Cobre Panamá restarts. On the smelting side, the negative-TC environment is a death sentence for standalone Chinese smelters but a gift for integrated mine-to-metal operations that never needed the processing fee.

The market is treating the Trafigura delivery like the all-clear. It wasn't. It was a fire truck showing up to the first floor while the building's structural beams are cracking. The copper is in the wrong country. The mines are producing less than every model assumed. And in ten days, a political decision will determine whether the dislocation resolves quickly or gets worse.

Physical wins. It always does. The fuse length is the only question.