Pod Insights The Real Estate Espresso Podcast What Is Data Center Demand? About this episode Today we’re asking a question that sounds simple, but it is actually the wrong question. People ask, how many data centers will be required over the next decade? The research says the better question is, how much compute, power, cooling, and land will be required? Because one 15 megawatt enterprise facility is not the same thing as one 100 megawatt AI campus. Counting buildings without counting power is like counting apartment buildings without counting units. It does not tell you what you need to know. ------------**Real Estate Espresso Podcast:** Spotify: [The Real Estate Espresso Podcast](https://open.spotify.com/show/3GvtwRmTq4r3es8cbw8jW0?si=c75ea506a6694ef1) iTunes: [The Real Estate Espresso Podcast](https://podcasts.apple.com/ca/podcast/the-real-estate-espresso-podcast/id1340482613) Website: [www.victorjm.com](http://www.victorjm.com) LinkedIn: [Victor Menasce](http://www.linkedin.com/in/vmenasce) YouTube: [The Real Estate Espresso Podcast](http://www.youtube.com/@victorjmenasce6734) Facebook: [www.facebook.com/realestateespresso](http://www.facebook.com/realestateespresso) Email: [podcast@victorjm.com](mailto:podcast@victorjm.com) **Y Street Capital:** Website: [www.ystreetcapital.com](http://www.ystreetcapital.com) Facebook: [www.facebook.com/YStreetCapital](https://www.facebook.com/YStreetCapital) Instagram: [@ystreetcapital](http://www.instagram.com/ystreetcapital)
Transcript 0:01 Speaker 00 Welcome to the Real Estate Especial Podcast, your morning shot of what's new in the world of real estate investing. I'm your host Victor Minash. Today, we're asking a question that sounds simple, but it's actually the wrong question. People are often asking me, how many data centers will be required over the next decade? In fact, we have an offer on one of our properties specifically for a data center. The research says the better question is, how much compute, how much power, cooling, and land will be required? Because 115 megawatt enterprise facilities, not the same thing as a 100-megawatt AI campus. Counting buildings without counting power is a little like accounting apartment buildings without counting apartments. It doesn't tell you what you need to know.
0:43 Speaker 00 The latest work from the International Energy Agency projects global electricity use by data centers to roughly double, about 945 terawatt hours by 2030. In the same analysis, the US alone is projected to add 240 terawatt hours of data center electricity demand by that time period, which is 130% increase from levels in 2024. The Lawrence Berkeley National Lab in work published for the US Department of Energy estimates that US data center consumption rose to 176 terawatt hours in 2023, and it could climb to 580 terawatt hours by 2028. That translates into about 74 to 130 gigawatts of average power demand.
1:26 Speaker 00 McKinsey came at the same problem from the capacity side. Their 2025 research projects data center demand reaching 220 gigawatts by 2030, with cumulative global capital spending of about 6.7 trillion. EPRI's 2026 update is even more direct about the strain on the US grid, projecting data centers will consume between 9% to 17% of US electricity by 2030. That's up from 4% to 5% today. This tells you it's no longer a niche specialty property type. It's becoming a strategic layer of national infrastructure. So, exactly how many data centers is that? Well, here's my answer. If 100-megawatt project are the new norm, as CBRE is suggesting in 2025, then 220 gigawatts of
2:10 Speaker 00 Double demand equates to about twenty two hundred one hundred megawatt campus equivalents. If the average campus ends up smaller, it's a fifty megawatts, then that count doubles. If the average campus ends up larger, then you can cut it in half. Add to that the fact that Synergy Research already counts one thousand three hundred and sixty large hyperscale data centers in operation and nearly eight hundred more in the pipeline. My working estimate is that the world will likely need the equivalent of between a thousand and twenty five hundred additional large campuses over the next decade, plus a great many expansions of existing ones. That range is my inference, not a published forecast, but it's based on capacity projections, not simply counting doors.
2:54 Speaker 00 Now, there's several forces that could pull that number down. First of all, better software can improve token efficiency. Better model architecture can do more work per watt. Hardware will keep improving. The IEA's high efficiency case shows that stronger gains in software, hardware, and infrastructure efficiency can materially reduce the electric footprint relative to the base path. And yes, efficiency matters, and it will save real money and real megawatts. Not all of the work performed by AI is time sensitive, so many data centers are being asked to implement dynamic load leveling in periods of peak demand in the power grid. Maybe if there's a hot day and people have their air conditioners running, data centers might be able to scale back and consume less power to make up the difference and soften that peak load. At the same time, several forces are pushing demand higher,
3:39 Speaker 00 perhaps faster than efficiency can offset it. Deloitte's latest survey shows AI applications are moving rapidly from pilots to production. Token demand is surging. Eighty eight percent of respondents expect AI factories to be deployed at least in a limited way by twenty twenty eight, and eighty six percent expect AI infrastructure budgets to increase over the next three years, with average budgets expected to more than triple. Now, this is what happens when AI shifts from experimentation into daily workflow. The demand absolutely multiplies. Once companies discover the real use cases, they don't need less compute, they actually use more. And then, of course, there's the edge question. Some functions will absolutely run on autonomous devices or on local hardware. Hardware, but the evidence today suggests that edge is a supplement, not a substitute.
4:24 Speaker 00 One survey from the uptime institute found only 2% of respondents identified end user devices as the primary location for AI inference and only 3% pointed to edge facilities. Central on-premise locations and colocation facilities are absolutely dominating. The final governor on growth is not desire, it's power, permits and utility interconnection. North America's primary market supply hit 1.2 terawatts in the first half of 2025. It was about 86 gigawatts in the pipeline. Vacancy fell to 1.6% and nearly three quarters of the new capacity under construction is already pre-leased. Means the limiting factor is not capital, it is time to getting power. So from a real estate perspective, the opportunity is not chasing the label data center.
5:08 Speaker 00 The opportunity isn't controlling the scarce ingredients. Land that has ready access to power, transmission access, water, fiber, tax jurisdictions and the entitlement path that can actually get it built. Data center is not simply a building with servers, it's a utility project wrapped in real estate. And the winners over the next decade will be the groups who understand that distinction. So Mike, answer the question is this. The world will require far more data center capacity than it has today. Probably the equivalent of about one to two and a half thousand additional large campuses globally over the next decade. With the exact building count determined less by demand than by design choices, efficiency gains and availability of power. So focus on gigawatts, not doors. As you think about that, have an awesome rest of your day.
5:54 Speaker 00 Go make some great things happen. We'll talk to you again tomorrow.
Previous Pivoting During A Drought with Chad Zdenek Next Leasing In An Oversupplied Market Page updated November 14, 2026 · transcript and summary are machine-generated