The AI-Driven Commodity Supercycle

The launch of ChatGPT in November 2022 was a defining moment in history. Within months, it accelerated a full-blown AI arms race and triggered an unprecedented concentration of capital expenditure into compute infrastructure. Over the past three years, major hyperscalers have announced plans to invest hundreds of billions of dollars in data centres, chips, networking and power systems. This scale of spending is now powerful enough to break decades-old relationships across global commodity markets.

In an earlier article, “The End of Cheap American Gas,” I outlined how U.S. data-centre power demand has expanded from 10 GW at the time of ChatGPT’s launch to roughly 80 GW of capacity in the pipeline today—an eightfold increase in three years. In the near to medium term, much of this incremental electricity load is likely to be met by natural gas-fired generation, contributing to upward pressure on U.S. gas prices and increasingly decoupling from global gas prices.

Source: FactSet

However, the implications of the data-centre build-out extend well beyond energy markets. The same 80 GW pipeline is creating another dramatic dislocation—this time in industrial metals.

Copper has long been known as “Dr. Copper” because its close relationship with global economic activity. Historically, periods of strong growth in manufacturing, construction, and transport were associated with simultaneous increases in copper, crude oil, and iron ore prices. Conversely, when global growth slowed, these commodities tended to weaken together.

Source: FactSet

That relationship is now broken. Since 2024, Brent crude has declined 20% to slightly above US$60/bbl, while iron ore has hovered around US$100/t. Both signalling a subdued global economy marked by stress in China’s property sector and the U.S. manufacturing teetering on recession. Over the same period, however, copper prices have risen more than 30%, reaching all-time highs.

The reason for copper’s dislocation reflects the exceptional metal intensity of AI data centres. Each gigawatt of hyperscale capacity is estimated to require around 25,000 tonnes of copper for high-voltage cables, busbars, transformers, cooling loops, and server hardware. Scaled across the full 80 GW of projected U.S. projects, that translates to approximately 2 million tonnes of new copper demand over the next five years – equivalent to about 9% of current annual global copper mine production of roughly 22 million tonnes. By comparison, the entire global electric-vehicle sector, which produced 17 million vehicles in 2024, is estimated to have consumed about 1.4 million tonnes of copper.

On the supply side, global copper mine output has grown only modestly over the past decade, rising by 3 million tonnes in total. Based on the largest projects currently in construction or advanced ramp-up – including Kamoa-Kakula, Tenke Fungurume, Kisanfu, Spence, Oyu Tolgoi and the potential restart of Cobre Panamá – the industry is on track to add about 1.2 million tonnes per year of new copper supply at full ramp over the next five years. While this represents a meaningful increase, it remains small relative to the scale of projected demand growth this decade. The pace of supply growth appears unlikely to fully offset the structural step-change in demand emerging through 2030, particularly given persistent risks around project delays, declining ore grades, energy constraints and geopolitical disruption.

Copper’s closest industrial relative, aluminium, is exhibiting a similar dislocation from traditional commodities such as oil and iron ore. Aluminium is often used as a partial substitute for copper in cables, busbars, and certain transmission applications, and current prices make substitution economically attractive. However, many critical components of data centres remain copper exclusive, which limits the potential for complete substitution.

Based on these constraints, the planned 80 GW of U.S. data centre capacity could generate incremental aluminium demand of roughly 1.5 million tonnes, equivalent to 2 percent of current global primary production. This reflects both aluminium used as a copper substitute and aluminium required in applications where copper is not essential. In essence, both metals are joined at the hip and are exposed to the same structural AI-driven investment cycle.

Source: FactSet

The AI investment surge is driving profound structural shifts across commodity markets, disrupting long-established correlations that previously guided price behaviour. Notably, copper, aluminium, and U.S. natural gas no longer follow their traditional patterns. For the first time this century, a new commodity super cycle is emerging that is not driven by China, but by the physical infrastructure of the digital economy. With much of the planned investment still underway, this capital-driven transformation is only beginning.