The Physicality Of AI: Why The Mining Supercycle Is Just Beginning

Introduction Earlier this month, Glenn wrote an article on the Newmont Corporation (NEM), which is one of the world’s biggest precious metal miners. Here’s a part of his takeaway: Newmont isn’t just riding a gold rally. It’s a fundamentally transformed business with elite financial metrics, a strengthened asset base, and structural tailwinds from central bank gold accumulation. And here is its 5-year stock price (on a total return basis): [Inline image] Source: StockCharts (NEM)  The fascinating thing about gold and its miners is that it’s part of a rather large capital shift. Since 2023, headlines have been dominated by software and AI hyperscalers. While these companies are still critical, 2026, so far, has turned into a very different year, as Bloomberg confirmed in a recent article , which was titled “Mining Stocks on Cusp of Supercycle as AI Boom Stokes Metals.” Since the start of last year, the MSCI Metals and Mining Index has almost doubled. This also means it has outperformed almost every other industry and asset class, including AI hyperscalers and semiconductor companies with much higher margins.  [Inline image] Source: Bloomberg Now, there are two ways this can go: This could be an overblown rally, as investors have jumped on the FOMO (fear of missing out) train, which could lead to another decade of underperforming mining stocks once the market finds out how capital-intensive these companies really are.  There’s something bigger brewing below the surface that could create tremendous new opportunities.  I think it’s the second option, as we’re dealing with a collision of AI infrastructure, the energy transition, and what seems to be a case of chronic underinvestment in supply.  So, let’s dive into the details!  Suddenly, Metals Are A ‘Growth’ Trade I’m not bearish on semiconductor companies. Not at all.  I think demand for GPUs and related equipment will remain high for a very long time.  My point, however, is that capital will rotate, as demand for GPU chips has been priced in for years. The physical basis for the future AI, however, hasn’t.  To give you some examples, a traditional enterprise data center usually requires 1 to 10 kilowatts per rack. The ones that Nvidia systems run require up to 8x as much. That’s supportive for Nvidia, as it supplies the tools needed to modernize and build data centers. It’s also bullish for the companies that benefit from the fact that a conventional data center requires up to 15,000 tons of copper. A mega hyperscale project can consume up to 50,000 tons.  I didn’t make that up. It’s from Forbes .  [Inline image] Source: Forbes Below is a tweet that sums it up quite well: You just can't own enough Copper. Billionaire Robert Friedland sums it up perfectly..... You people have no idea whatsoever what we’re facing. “We’re consuming 30m tonnes of copper a year. Only 4m tonnes of which is recycled. That means to maintain 3% GDP growth, with no electrification, we have to mine the same amount of copper in the next 18 years as we mined in the last 10,000 years, combined.  In the next 18 years, I’ve got to mine the same amount of copper as we mined the last 10,000 years. This is without any new electrification, without data centers, without solar and wind and the greening of the world economy. You people have no idea whatsoever what we’re facing.” - X/@TheGladiatorHC The next chart shows a demand forecast of Freeport-McMoRan (FCX), one of the two biggest copper-focused miners in the world. [Inline image] Source: Freeport-McMoRan As this is the shorter-term data center outlook, you can imagine what this means for copper demand: [Inline image] Source: S&P Global  The ‘funny’ thing is that these data centers need a stronger grid. Goldman Sachs Research believes that grid and power infrastructure will drive more than 60% of copper demand growth