What Do Data Centers Require?
Welcome to the Real Estate Espresso podcast. I’m your host, Victor Menasce, and on today’s show we’re examining the impact of data centers on local economies, focusing on the energy component and the technology lifecycle. These variables will significantly influence employment in areas where these data centers are located.
We often hear that energy drives the economy. It’s a fact that for each unit of GDP, an equivalent unit of energy is consumed somewhere globally. However, with data centers, energy isn’t just driving the economy; it’s also the lifeblood of the AI industry. Despite the global push for AI, the world isn’t generating nearly enough electricity to sate the increased demand from these emerging data centers.
What piqued my interest in the past week wasn’t the litany of proclamations of fresh power-generating capacity— though that in itself is impressive. We are venturing into something more critical, anchored in the aggregate power generation announcements in just the previous month alone. While many of these announcements revolve around the two gigawatts range, we need to keep in mind that two gigawatts of power is sufficient to supply one and a half million homes.
Recently, Oracle and OpenAI unveiled an additional four and a half gigawatts of data center capacity, with much of it positioned in their Stargate facility in Abilene, Texas. This colossal project, a joint venture between OpenAI, Oracle, and SoftBank, is already under construction with parts of it up and running.
However, apart from electricity, the significant expense for these data centers is the AI hardware housed within them. The revenue of companies like Nvidia, which currently holds 90% market share, is eye-opening. Still, the swift technology obsolescence curve is even more shocking.
The magnificent strides in innovation in the technology sphere mean that AI chips that are two to three years old might be viewed as dreadfully slow compared to their newer counterparts. As such, the average lifespan from release to final use for frontier training requirements is roughly 3.9 years. The physical lifespans of these chips can even be incredibly short due to their high workloads, with estimates saying they could last only one to three years under heavy use.
These realities translate to frequent regeneration within these physical data centers, necessitating vast reserves of personnel, incessant rewiring, and continuous maintenance. Unlike traditional data centers teeming with equipment but relatively devoid of humans, AI data centers will require a significantly higher number of personnel due to shorter product lifespans and increased maintenance requirements.
This creates vast business opportunities for those located near these burgeoning data centers. As you ponder on that, have an excellent rest of your day. Go make some fantastic things happen. We’ll talk again soon.
Stay connected and discover more about my work in real estate by visiting and following me on various platforms:
The Real Estate Espresso Podcast:
- 🎧 Spotify: The Real Estate Espresso Podcast
- 🌐 Website: www.victorjm.com
- 💼 LinkedIn: Victor Menasce
- 📺 YouTube: The Real Estate Espresso Podcast
- 📘 Facebook: www.facebook.com/realestateespresso
- 📧 Email: podcast@victorjm.com
Y Street Capital:
- 🌐 Website: www.ystreetcapital.com
- 📘 Facebook: www.facebook.com/YStreetCapital
- 📸 Instagram: @ystreetcapital
