AI needs data centers. What this means for metals.

None of these firms are profitable just yet, but they, along with many others, will be spending around $3 trillion on data centers by 2028, according to Morgan Stanley. In fact, in the most recent round of earnings calls, many corporate executives noted that, if anything, estimates on how much they will spend on AI during the next five years are too low. Microsoft CFO Amy Hood noted: “I thought we were going to catch up. We are not. Demand is increasing,” she warned.

The impact of this spending, along with an assortment of other items, including GPUs, network gear, server farms, high-performance chips, HVAC systems, transformers, gas turbines, power cables and power plants, have had a significant impact on the US economy. In fact, AI spending accounted for more than half of US GDP growth for the first half of 2025. Specifically, of the 1.6% inflation-adjusted GDP growth seen in the US during the first half of the year, 0.8% was AI-related.

AI also accounts for the bulk of business investment. Strip it away and private business spending has been essentially flat since 2019, according to Deutsche Bank. In real estate, outside of data centers where building is booming, commercial construction of shopping centers and office buildings are both struggling.

Data centers

Anywhere from 100 to 5,000 people are needed to build a data center, a senior executive at Turner Construction recently told the Wall Street Journal. But shortages of skilled workers and materials are ongoing challenges, as are costly and lengthy lead times for critical items like transformers and turbines.

Moreover, many of these parts have to be in place concurrently. If power is available but the transformers needed to connect it to buildings are on a waiting list, operations will not move ahead. And so, while the level of announced expenditures by a number of AI companies sounds intimidatingly high, this spending will be spread out over many years given the current bottlenecks.

Of all the LME metals that are in play in the AI space, copper is the one of most interest. Estimates as to how much copper data centers will eat up vary significantly, but the consensus range is between 400,000-800,000 tons of copper has gone into AI data centers this year.

Aluminum does not play as prominent a role in the AI revolution as copper does, but it will ride on its coattails to some extent. We see aluminum being used in the server racks, cooling units, radiators, and HVAC system in many AI facilities, but aluminum cables are likely not going to be used as extensively given poor conductivity and larger cable sizes that make laying the lines more expensive. However, if the aluminum/copper ratio continues to expand from the current reading of about 4.5:1 to 5.5:1 or 6:1, it may usher in some aluminum substitution in the critical component category.

Aluminum should be used more extensively in the construction of the AI plant itself, such as in structural panels, doors, and roofing. In all, we have seen estimates that roughly 800,000 tons of aluminum will be needed for AI data centers by 2030, or a little more than 1% of current production. This is not a massive quantity in a 75 million-ton market, and, unlike copper, could be accommodated rather easily.

Even if aluminum demand is doubled for alternative green uses, such as for EVS and solar panels, the market will still likely be able to produce enough metal without getting tight.

Doubts remain 

The US stock market has been getting queasy about AI of late. US equity markets are growing skeptical, not so much with regard to AI’s promise, but more to do with the vast amount of capital being deployed and the lack of visibility on the revenue side. As a result, share prices for a number of AI stocks have been down sharply this past month, with many retreating by 20%-30%.

Markets are concerned about the following issues:

Even though AI demand is strong and financed (mostly) by companies with deep pockets, Morgan Stanley estimates of the $3 trillion that will be spent on AI over the next 3-5 years, only half will be financed by the companies themselves. The bank sees the rest being outsourced to private equity, the corporate bond market, and asset-backed securitization deals. This is fine, as long as the demand is there and market conditions remain strong. However, if we get anything like the dot.com bust, the outsourced market could retreat quickly.

And what about the US government? The Trump administration intends to play a role but will likely not write big checks. Instead, it will contribute a modest amount towards R&D, infrastructure and permitting help, and even those amounts have yet to be defined by Congress.

Nonetheless, Open AI’s CFO last month was brash enough to suggest it might look to Washington to provide a funding backstop. CEO Sam Altman quickly shot that down, saying the firm is not looking for any federal bailout, while insisting company revenues would “grow to hundreds of billion[s] by 2030.” Altman even chastised a skeptical investor and offered to buy his shares back if he was unconvinced by the company’s potential.

Such is the drive (dare we call it hubris?) of the CEOs leading the AI charge, strikingly reminiscent of remarks made during the dot.com era. Indeed, in August 2000, Cisco’s ex-CEO John Chambers unveiled quarterly earnings growth of more than 60% (similar to Nvidia’s) and confidently predicted that the “Second Industrial Revolution” is just beginning. A year later, Cisco’s stock was down by nearly 80% amid a broad sector slump.

Of course, there are differences between the two eras, with the most important being that firms in the AI space are better capitalized and enjoy stronger demand, something that eluded many internet firms at the time. In addition, many companies in the 1990s spent massive amounts on infrastructure while waiting for internet traffic to catch up. Although it ultimately did, it was too late to save many.

Circular funding

Although money is pouring into the AI industry, the circularity behind these investments is also raising concern. As one analyst observed: “OpenAI has agreed to pay $300 billion to Oracle for new computing capacity, Oracle is paying Nvidia billions for chips to install in OpenAI’s data centers, while Nvidia has agreed to invest up to $100 billion in OpenAI as it deploys Nvidia chips.”

Nvidia recently said it will invest up to $15 billion in Anthropic, an Open AI competitor. Anthropic, in turn, would buy $30 billion of computing capacity from Microsoft Azure running Nvidia AI systems. All this is eerily reminiscent of the dot-com era as well, where there was considerable cross-pollination going on before many firms ultimately went down together.

On the revenue side, investors are questioning some of the assumptions on how long it will take for AI earnings to materialize. Morgan Stanley expects tech companies to spend nearly $3 trillion on AI through 2028 but only generate enough cash to cover half that amount. JPMorgan calculates that the four majors must draw an extra $650 billion in revenue every year—early double Google’s total sales in 2024—for AI investments forecast through 2030 to earn a modest 10% return..

OpenAI’s revenue is not public, but Altman expects revenues of $13 billion-$20 billion this year, and $30 billion next year before doubling again in 2027. But losses are predicted to roughly triple to more than $40 billion by 2027 and to $74 billion in 2028.

On the consumer side, although AI chat boxes and image generators are already being used by hundreds of millions of people, many users are on free versions and so it is not certain how many are going to pay for a service they could conceivably get for free from “no-name” competitors. But for now, AI companies are clearly betting on the more stable—and far more profitable—business sector.

Productivity claims not materializing

There are other nagging concerns about all the productivity contributions AI conceivably could make. An MIT report found 95% of organizations surveyed said that they are getting no productivity return on their AI spend. The best AI system ranked by Scale AI and the Center for AI Safety calculates it successfully completed a measly 2.5% of tasks done by humans. A Yale study released last month agreed, concluding that AI has not yet disrupted the labor market in a significant way. Similarly, a McKinsey study revealed that while eight in 10 businesses said they are using AI technologies, most said this had “no significant bottom-line impact.”

Despite questions about productivity contributions, we think AI will displace quite a number of workers. According to a recent estimate from Challenger, almost 50,000 US job cuts announced so far this year were a direct result of AI-related efficiencies, concentrated mainly on repetitive, high-volume work. But as “inference AI” starts to take hold, the job erosion could intensify and spread to more skilled areas of the workforce (including to commodity analysts like yours truly!). It is no wonder therefore, that AI carries a decidedly negative perception, in stark contrast to the enthusiasm that greeted the internet revolution.

Why this matters

We know that the tech industry is spending massive amounts of money on AI, with these funds originating from both the companies themselves and from outside investors. At some point, the industry has to show its revenues are real and significant.

For society at large, excerpts from an Atlantic article summed up the AI revolution best, albeit in somewhat Orwellian terms:

“The biggest lesson of the past two decades of Silicon Valley is that Meta, Amazon, and Google—and even the newer AI labs such as OpenAI—have remade our world and have become unfathomably rich for it, all while being mostly oblivious or uninterested in the fallout. They have chased growth and scale at all costs, and largely, they’ve won. The data-center build-out is the ultimate culmination of that chase: the pursuit of scale for scale itself. If AI companies deliver on their massive investments, it would likely mean producing a technology so capable and revolutionary that it wipes out countless jobs and sends an unprecedented shock wave through the global economy before humans have time to adapt. (Perhaps we will be unable to adapt at all.) If they fail, there will likely be unprecedented financial turmoil as well. In all scenarios, the outcome seems only to be real, painful disruption for the rest of us.”

The quote is the real takeaway of the imminent AI revolution; how much copper, aluminum and power we ultimately need is an interesting analytical exercise, but it pales in comparison to what is conceivably at stake for society in so many respects.