Data Center Growth Challenged Despite AI Demand

AI has taken on a hype of its own. Software companies feel compelled to extoll the wonders of their products’ artificial intelligence aspects to say nothing of the ones making everything into generative AI.

But talk is cheap and building data centers for AI implementation is more complicated than raising money, making plans, and a ribbon-cutting. BlackRock’s iShares recently released a report on AI and geopolitics. It also contains some observations about creating the infrastructure for AI that will have implications for the CRE property aspect.

The important part of the report for this aspect is the second one. The more demand for AI, particularly generative AI, the greater the need for materials, semiconductors, and energy.

As they note, many companies are looking to integrate generative AI into their operations with the promise of more work with fewer salaries to pay, while enabling automated features to simplify customer support. The report mentions a Gartner estimation that by 2026, “more than 80% of enterprises will have used generative artificial intelligence (GenAI) application programming interfaces (APIs) or models, and/or deployed GenAI-enabled applications in production environments, up from less than 5% in 2023.”

This should be taken with some salt because tech market research firms are often unrealistic in their predictive scenarios. Eventually, technologies settle in, but may be in smaller forms with far more time needed for maturity of implementation. Whether the paperless office, automated demand prediction, quick adoption of enterprise resource planning, dot com madness, or the year of practical cryptocurrency, many sure things have faded to the back.

“We are now building ever-larger AI factories to manufacture intelligence. This buildout phase is only in its second year and I expect it to continue throughout the decade, to become the largest infrastructure investment in history,” Tony Kim, BlackRock head of fundamental equities global technology, said in prepared remarks.

The report said that markets vastly underestimate the amount of money that will be needed for the development of data centers over the next five or six years. Fair enough, they may be, but many CRE investors and developers have financial concerns about whether they can afford to refinance major property holdings. Many market analysts underestimate the financial pressures facing many in commercial property.

Power is another factor, as iShares quoted Goldman Sachs expecting U.S. data centers to represent 8% of all power consumption in the country by 2030. Another six years out could face large increases in the cost of power. “Our Fundamental Equities team sees AI servers as costing roughly 40 times the cost of traditional data centers of the past and could require about $10 billion-$12 billion to build each gigawatt of additional data center power,” they wrote.

And then there are materials, with copper possibly becoming the “chokepoint” for energy infrastructure and digital infrastructure. The International Energy Agency wrote last fall in its “Energy Transition Mineral Price Index, which tracks a basket price of copper, major battery metals and rare earth elements, tripled in the two years following January 2020, but relinquished most of the increase by the end of 2023 — although copper prices remained at elevated levels.” That will make electrical systems more expensive to afford.