AI’s brave new world is going nowhere without the old

THE hype around artificial intelligence (AI) has the hallmarks of a classic gold rush: hordes of hopeful prospectors rushing to unearth the treasure, no matter the cost.

But more people prospered from selling picks, shovels and pans during the 1849 California Gold Rush than from finding nuggets.

If AI is today’s gold rush, then data centres are its picks and shovels. And investors should look to the old-world commodities – especially uranium and copper – that are essential for those data centres and the AI-driven economy they enable.

Demand is growing exponentially for the computing capacity to train and deploy ever more powerful AI models, fuelling a scramble to build new data centres, along with the power generation and transmission infrastructure they will need. While attention has focused on bottlenecks in computing power as the main constraint on AI growth, the biggest constraint will soon be energy – especially AI’s need for high and constant supplies.

The European Union this month ordered data centres to report key energy parameters by September in a bid to understand likely power needs. In Ireland and Denmark, two countries that have actively sought to attract data centres, demand for energy may rise twofold and sixfold respectively by 2030. The International Energy Agency in its 2024 electricity outlook said data centre demand could double in the coming two years.

This pace of growth creates a dilemma: AI is devouring baseload power from often carbon-intensive sources at a time when the world’s efforts are focused on installing renewable capacity to meet climate goals. On top of that, data centres are becoming political hot potatoes because they strain national grids, lack transparency over their climate impact, suck up large amounts of land and capital, and create few jobs.

For societies to accept the trade-off, they need to work harder to reduce their carbon footprints and improve transparency around their purpose; by some estimates, 88 per cent of their memory banks are clogged by kitten videos, selfies and other once-off content that’s never accessed again.

Increasing the supply of renewable energy will obviously be essential to making AI and the data centres it depends on more sustainable. But the AI boom is also likely to lean heavily on the clean baseload power offered by nuclear energy – and it’s no coincidence that some leaders in the field are investing in the sector, including Sam Altman and Microsoft.

Nuclear option

The nuclear industry provides around a quarter of the world’s carbon-free power, but has proved controversial in the climate-change debate, especially in the wake of the Fukushima nuclear disaster. That led Germany to shutter all its plants, even as the country bought record amounts of electricity from France, whose reactors generated around 320 terawatt hours of power last year.

Nuclear is a natural fit for energy-hungry, always-on AI servers, providing a steady source of zero-emissions energy. The price of uranium has already doubled since the beginning of 2023, reflecting increasing acceptance of its place in the energy transition, while supply is constrained by decades of under-investment. These factors suggest a bright future for this heavy metal.

Uranium has been used as a source of energy for more than 60 years – but the use of copper goes back to the dawn of civilisation, and it will be just as important to the AI era. In particular, copper has been a key metal of the electrical age, in which it has been used in cables, motors, circuits, lights, air conditioners, heaters, transformers and power transmission grids.

Copper will still be king in the renewable energy age. It is used in electric vehicles, solar arrays, wind turbines and even green hydrogen production. These renewable energy systems are also highly copper-intensive, with many requiring six times the amount of copper than an oil-fired thermal power station for each megawatt of electricity produced.

Copper’s ubiquity mean its future is both assured and demand for the metal difficult to estimate. Prices have tested historical highs in recent weeks, even as consensus forecasts suggest a surplus in production. Delays in mine production and licences are also holding back output. Yet the slower than expected uptake of electric vehicles in many countries, and weakness in the Chinese economy, are pulling prices in the other direction.

Equally, it’s not yet clear just how much additional energy will be required because of the proliferation of AI applications running in data centres. Nvidia’s latest Blackwell chip is both much more powerful than its predecessors and much less power-hungry.

These are among the big questions that we’re going to be tackling at Tribeca’s Future Facing Commodities conference here in Singapore next week. But one thing is clear, though: valuations are skewed. The shares of Nvidia’s biggest rivals and cloud customers trade at around 70 times their earnings. That’s more than 10 times where Jiangxi Copper, China’s biggest supplier of the metal, is trading.

So, if you believe that AI changes everything, it’s worth looking past the hype cycle for AI stocks themselves and considering what it will take for this technology to fulfil its promise. That means data centres – and increasing data centre capacity means increasing the supply of clean energy. We believe that in turn means growing demand for uranium and copper, among other time-honoured commodities. We won’t build the brave new world without old-world resources.

The writer is country head of Tribeca Singapore