How data center developers are powering AI
AUTHORS’ NOTE: In the short time between when we wrote this paper and its publication, the Chinese AI company DeepSeek released a report claiming that its AI model uses 10-40 times less energy than similar U.S. AI technology. Greater energy efficiency is a good thing, of course. But it’s not likely to eliminate massive new power requirements, just as virtualization—which dramatically improved servers’ computational efficiency—didn’t alter data centers’ growth trajectory. U.S. companies are not likely to use DeepSeek’s models, but if they replicate its more efficient training approach, that may well spur even greater levels of AI adoption that continue to drive data centers’ growing power requirements.
- Already massive, accelerating data center demand is driving unprecedented levels of demand for power capacity—potentially growing at three times the pace of the last 20 years.
- Power supply is constrained by a number of factors, including permitting timelines and equipment supply chains.
- There is a range of ways savvy developers are getting the power they need: on-site generation, co-development of power infrastructure with utilities, proactive land development and power procurement, and (to a lesser extent) moving to new markets.
Massive data center demand is driving huge new power requirements
Exponential growth in demand for data center capacity continues to accelerate— the U.S. colocation data center market doubled in size from 2020 to 2024, and that doesn’t account for all the capacity brought online by hyperscalers themselves. AI could drive demand even faster; Dell CEO Michael Dell predicts that AI will drive a 100x increase in data center demand over the next 10 years.
Demand is rising so much faster than supply that vacancy rates in major data center markets are at their lowest levels ever—below 3% nationwide.
EXHIBIT 1: Data center supply is growing, but not as fast as demand

As we’ve written before, the primary task of a data center is to securely house, power, and cool the IT equipment on which virtually every aspect of modern life depends. Challenges abound: land is constrained in some markets, and data center operators are having to adopt new cooling strategies to support AI. But power is the predominant gating factor in data center developers’ ability to support growing demand.
Conclusion: Investor considerations
Data center developers face a ‘perfect storm’: massive (and accelerating) demand for data center capacity, driving unprecedented rates of demand for power, in an environment in which getting power to the data center takes twice as long as before. Not all developers are equally well-equipped to navigate these challenges. For investors, realizing the full return potential requires a keen understanding of which developers are poised to be able to deliver more data center capacity, more quickly than others.