ROI-Metals in the Age of AI — Game-changer or hyperscale hype?: Andy Home

Artificial intelligence may operate in the cloud, but its metallic roots lie in the earth. Data centers wouldn't exist without a wide spectrum of metals, from galvanised-steel housing and copper wiring to rare earth magnets driving the servers. AI ​data centers are even more metals-intensive than traditional facilities. They are physically bigger, contain more computing hardware and use more power.

Investors have leaned into AI's metallic backbone. Copper rode the AI investment wave to all-time highs earlier ​this year before becoming collateral damage in the recent technology-stock rout.

Tin, which has leapt from the Bronze Age to the AI Age via circuit boards, ‌has been similarly ​rocked by broader tech turbulence. But is AI demand for metals really the promised game-changer, or is it hyperscaled hype?

As with any new technology, there's a bit of both in the mix. METALS TO (ARTIFICIAL) MIND

Data centers consume 60 to 75 metric tons of metals per megawatt (MW), according to a white paper by the World Economic Forum (WEF). To put that into context, standard data centers use 5 to 10 MW of power, while AI hyperscaler facilities exceed 100 MW, according to the International Energy Agency. Hyperscalers are building the fastest, propelled by US-China competition for AI dominance.

Just about every metal you can name, and a few you probably can't, is deployed in the labyrinth of servers, chips, storage, networks, electrical systems, cooling systems and backup-power facilities. Aluminium, ‌copper and steel are everywhere in the building. More exotic elements such as gold, silver, cobalt and rare earths are concentrated in the core computing hardware. And backup power is currently a battleground of competing technologies, with lithium, lead and nickel-zinc batteries all fighting for market share.

But in terms of market impact, the three metallic standouts are gallium, germanium and tin. Data center demand will account for 6.0%, 2.4% and 1.2%, respectively, of global annual usage in 2035, according to the WEF report. Gallium and germanium are tiny markets, and both are used in minute quantities, but they have an exponential impact on semiconductor performance.

China, the dominant producer of both metals, has restricted exports, which is why everyone else is scrambling to find alternative sources of two metals that are literally powering the global AI race. Tin is less exotic and also remains a small market by base-metal standards, with annual consumption of just over 400,000 tons.

But since the phase-out of lead, tin has become the dominant material for soldering circuit boards. The more ‌complex the data center architecture, the more tin it needs to bind everything together. Copper comes in a distant fourth in the demand-impact table, with data centers accounting for less than 0.9% of global usage in 2035.

Moreover, data centers are likely to start using less copper and other metal as structural designs improve and operating processes become more efficient. Wood Mackenzie expects both copper and aluminium usage to peak in the early 2030s before declining over the rest of the ‌decade. Who knows, AI itself may be the lead engineer in this technical evolution.

BEYOND THE SERVER ROOM Copper bulls shouldn't be too disheartened, though. The WEF estimates for data center metals usage stop at the door of the building, excluding the metals needed to provide the power that keeps the whole thing humming away in the clouds.

"The biggest metals story lies outside the facility itself" in the form of transmission lines, substations and energy infrastructure, according to Shashank Sriram, senior research analyst at consultancy Wood Mackenzie, speaking to DataCentre Magazine. Connecting data centers to the grid or building co-located power plants requires significant volumes of copper, aluminium and zinc for galvanising steel towers and poles.

If power connectivity is included in the equation, actual demand for both copper and aluminium is three or four times the projections based solely on the physical facilities, Wood Mackenzie estimates. And if data centers draw power from renewable sources such as solar and wind, the demand impact is magnified because these are also highly metal-intensive technologies.

Wood Mackenzie projects power-capacity additions for data centers to rise from 15-20 gigawatts (GW) a year today to 30-33 GW in the early 2030s. This is where the AI investment meme melds with the existing electrification-of-everything theme. From a market ⁠perspective, the two are self-reinforcing bull ​narratives that are tricky to separate into component drivers.

RAGE AGAINST THE MACHINES Everything, of course, depends on how many AI data centers are ⁠actually built in the next few years, which is where some of the hype comes in.

There are many estimates, all expressed in eye-watering trillions of dollars of investment. However, many of the new megawatts of announced AI computing capacity can more accurately be described as "bragawatts", according to a study by the Oxford Smith School and financial broker Marex Group.

While media and markets price in a rapid, exponential ramp-up in AI infrastructure, the study says the true outcome is likely to be "a roll-out that is delayed, lumpy, and constrained by physical realities." The single biggest bottleneck is grid connectivity. A data center ⁠can be built in 18 to 24 months, but the average US waiting time for a grid connection was four years between 2018 and 2023, the study found.

Compounding the physical hurdles is the growing political backlash against AI. The Democratic sweep in the US midterm elections on November 3 included gubernatorial wins in Texas and Georgia, where Democrats campaigned in part on a moratorium on new data centers. The incoming Democrats are planning to unleash a flood of new legislation targeting AI systems and data center development when they take control of the US ​Congress in January.

And it's not just the US where public opinion is turning against the machines. Protesters broke into the construction site of a Microsoft data center in Amsterdam last week, part of coordinated protests across 16 countries under the "Stop Dirty Data Centres" banner.

Politicians are coming under increasing pressure to rein in what many see as the reckless build-out of a technology with unmatched potential to change the world in both good and bad ways. Anthropic's ⁠warning in its IPO prospectus that AI could pose "catastrophic or existential risks to humanity" doesn't really help the industry's cause.

POLICY CONTROLS Government policy is likely to be as important as grid connection times in setting the pace of AI data center construction.

There's an analogy with the electric vehicle (EV) narrative that so excited battery-metal markets a few years ago. The road to a fossil-fuel-free future has turned out to be a bumpy one, in part due to changeable government subsidy schemes.

The most glaring example is in the US, where President Donald Trump has rolled back his predecessor Joe Biden's subsidies for battery vehicles. The result has been ⁠a collapse ​in EV sales in the world's second-largest automotive market. Sales in the first 10 months of this year were down 21% from the same period of 2025, according to consultancy Benchmark Mineral Intelligence.

The AI metals revolution could turn out to be just as much a stop-start affair. But another lesson can also be drawn from the EV sector.

The global roll-out of battery-powered vehicles may not have lived up to its early hype, but it hasn't stopped. Despite plunging sales in the US, global sales were still up 5% year-on-year in January to October. In other words, governments can restrain growth of a new technology but they can't stop broader adoption.

HERE COME THE ROBOTS Moreover, building AI's digital infrastructure is only the first part of this latest metals story.

"If digital AI is the brain of the next economy, embodied AI is the beginning of its body," according to Frank Nikolic, research principal for base and precious metals at CRU. That means "machines that can sense, decide ⁠and act in the physical world, widening the addressable market from information work to real-world labour."

The robots are coming, and they are also built with metal. Early deployment remains selective, CRU noted, but for some industrial functions such as logistics, sorting and inspections, embodied AI has already moved out of the laboratory and is being piloted in real-world environments.

"For commodity markets, the crucial point is not the novelty of the technology itself, but the fact that AI ⁠is moving from experimentation toward large-scale deployment and that needs metals and materials," according to Nikolic. If you're starting to worry we're entering the ⁠realm of science fiction, specifically the Terminator films with their eponymous liquid-metal, shape-shifting humanoids, you may have good cause. Researchers are already working on soft robots, which can both handle delicate items, such as eggs, and change shape.

Much of the work centers on room-temperature liquid metals, which make soft robots "extremely deformable with higher degrees of freedom while navigating inside narrow and constrained spaces," according to researchers at China's Shenzhen University, writing for ScienceDirect. Designers at Carnegie Mellon University in Pennsylvania unveiled a "proof of concept" nano liquid-metal robot in 2023 and, with a nice nod to the Terminator franchise, made it melt through the bars of an enclosed cage.

The metal they used was gallium. It looks like we have a clear metals winner in the ‌Age of AI.

(The opinions expressed here are those of Andy Home, a columnist for Reuters.) Enjoying this ‌column? Check out Reuters Open Interest (ROI), your essential new source for global financial commentary. Follow ROI on LinkedIn, and X.

And listen to the Morning Bid daily podcast on Apple, Spotify, or the Reuters app. Subscribe to hear Reuters journalists discuss the ​biggest news in markets and finance seven days a week. (Writing by Andy Home; Editing by Marguerita Choy and Anna Szymanski)