Copper Tops $14,000 as Goldman Widens Its Deficit Call, and the AI Build-Out Is the Buyer

Copper Tops $14,000 as Goldman Widens Its Deficit Call, and the AI Build-Out Is the Buyer A Goldman supply-driven upgrade pushes copper within $500 of its record, exposing the metals squeeze behind the data-center power surge. Copper is trading just above $14,000 a ton in London, roughly $500 shy of the all-time high it set in January. Goldman Sachs lifted its end-2026 price target by more than 10% this week (week of 2026-06-01), to $13,735 a ton from $12,465, and Wall Street thinks the metal has further to run.⁶ What is driving the move is supply, not a fresh burst of demand. Goldman cut its global mine supply estimate by 350,000 tons, citing operational disruptions at Indonesia's Grasberg complex and Ivanhoe Mines' Kamoa-Kakula operation, and now sees the copper deficit outside the United States exceeding 640,000 tons this year, up from a prior forecast of just 60,000 tons. That is a tenfold revision in the shortfall, and it lands while inventories are already thin.⁶ The Kamoa-Kakula numbers show how fast the picture has deteriorated. The mine, once on track for 420,000 tons of copper this year, has cut its 2026 guidance to 330,000 tons after seismic disruptions in 2025 slowed its ramp-up. Copper is up roughly 10% year to date on the London Metal Exchange, outperforming gold over the same stretch.⁶ That matters to energy desks because copper is the physical bottleneck in the electrification and data-center story they have been pricing through power and gas. Every gigawatt of new compute load needs cabling, transformers, switchgear and grid connections, and copper sits inside all of it. The International Energy Agency, in its World Energy Outlook released on Wednesday (2026-05-20), warned of a more fragile energy-security picture and an AI-driven power surge, and called for greater diversification of supply.⁵ The chokepoint problem cuts across fuels too. Import-dependent economies remain exposed to recurring energy-security risks from chokepoints even amid abundant oil and LNG, the climate and energy think tank E3G said in a study reported by Montel on Tuesday (2026-05-19). Those risks, it argued, cannot be designed away by a supply glut.⁴ The demand signal from the power-equipment side is concrete. Fluence Energy reaffirmed its 2026 revenue target of roughly $3.2bn to $3.6bn, with 85% of the midpoint already contracted, and pointed to new master supply agreements with two hyperscalers as it pushes into data-center energy storage.¹ Babcock & Wilcox, pivoting its industrial generation business toward data-center baseload, booked a $2.4bn design-build contract with Base Electron for 1.2 GW of gas-fired power that drove its backlog up 470% to $2.8bn; the stock closed at $14.54, up 129% year to date.² So the buyers are real and the equipment orders are being signed. The constraint is upstream, in the mines, and a supply-led rally is harder to fade than a demand-led one. A demand spike can cool with sentiment. A 90,000-ton hole at a single mine does not refill on a soft tape. Cheap gas underlines the contrast. US working gas in storage fell by 52 Bcf in the week reported on 2026-05-21, well below the five-year average draw of 168 Bcf, leaving inventories 141 Bcf above a year earlier, about 8% higher; NYMEX Henry Hub front-month was trading near $2.86 at the time.³ Energy itself is plentiful and cheap; the metal needed to move and consume it is neither. There is a wrinkle in the bull case worth flagging. Goldman's new end-2026 target of $13,735 sits below the current spot price above $14,000, so the upgraded forecast still implies the market has run ahead of the bank's modelled fundamentals.⁶ That gap is either a sign of more upgrades to come or of froth that mean-reverts. Watch the mine-supply headlines first. Further guidance cuts at Grasberg or Kamoa-Kakula would validate the deficit call and keep the LME bid; a stabilisation in output would test how much of the rally is structural shortage versus positioning.⁶ For power and gas traders, the read-through is that the binding limit on the AI build-out may turn out to be copper and transformers, not megawatt-hours.⁵