LearnBits

L’argent a atteint un niveau record en janvier, puis a perdu près de la moitié de sa valeur au moment où la guerre entre les États-Unis et l’Iran a éclaté en février. Il s’échange désormais autour de 58-59 dollars l’once, grattant de petits gains après une période particulièrement brutale qui a vu les contrats à terme sur l’argent enregistrer la pire baisse sur une seule journée depuis les années 1980. Ce type d’aller-retour — record historique, quasi-effondrement, puis reprise encore hésitante, le tout en l’espace de six mois — est la véritable histoire derrière toute discussion sur le « meilleur commodity pour 2026 » en ce moment, et pas seulement le chiffre du prix en une journée.

Des sirènes d’alerte aérienne se sont déclenchées deux fois à Bahreïn en l’espace de quelques heures, dimanche matin. Entre 2 h et 3 h (heure locale), les Gardiens de la révolution iranienne ont lancé des missiles balistiques et des drones depuis la base aérienne d’Ali Al Salem, au Koweït, ainsi que contre le siège de la Ve Flotte américaine à Port Salman, affirmant avoir détruit huit éléments d’infrastructure américaine. La IRGC a présenté cela comme une riposte à une frappe américaine récente sur cinq positions côtières iraniennes, quelques heures plus tôt — le même cycle qui s’est répété au moins cinq fois depuis mars.

A single AI data center, the kind hyperscalers are now building by the dozen, can chew through copper at a rate that would have sounded absurd five years ago. The going estimate runs anywhere from 30 to 150 tonnes per megawatt depending on whether grid infrastructure is counted, and at gigawatt scale that adds up to roughly 50,000 metric tons per facility. With around 15 gigawatts of new AI capacity going up annually, that's 750,000 tonnes of fresh demand a year from data centers alone — before a single transmission line upgrade or substation gets counted.
That timing couldn't be worse for a market already running short. Disruptions at Indonesia's Grasberg mine and the Kamoa-Kakula complex in the Democratic Republic of Congo have knocked a meaningful chunk of global refined output offline at precisely the moment demand is accelerating. Ore grades at legacy mines have fallen roughly 40% since 1991, and building a new mine from discovery to first production routinely takes 15 to 29 years once permitting is factored in. The International Copper Study Group has revised its 2026 forecast from an expected surplus to an outright 150,000-ton deficit — a swing large enough to reshape how every major bank is pricing the metal.
That repricing is already visible. J.P. Morgan projects copper averaging around $12,075 a tonne for 2026, with UBS reaching further out to $15,000 by early 2027. Not everyone agrees on the timeline — Goldman Sachs still expects a modest surplus this year before structural deficits take hold in 2027, and Red Cloud actually trimmed its near-term forecast on softer US demand even while raising its 2028 and 2030 numbers. The disagreement isn't really about direction. It's about how soon the math catches up with itself.
What makes this cycle different from past copper rallies is that AI isn't competing with electrification for the metal — it's stacking directly on top of it. EVs, grid modernization, and renewable buildouts were already going to strain supply before a single AI campus broke ground. Robert Friedland, the mining veteran behind the Kamoa-Kakula discovery, put it about as bluntly as anyone in the industry has: the world is heading toward a shortage measured in the billions of tonnes over the coming decades, not the thousands. Whether 2026 ends up the year prices break out or merely the year everyone agreed they eventually would, the supply side has nowhere near enough lead time to catch up on its own schedule.