Energy Production the Silent Killer of A.I.

A funny thing happened the other day, my wife scheduled a car detailing and they plugged into the only outlet in the front of the house. They were running a power washer, vacuum and buffer on a 15 AMP circuit. The end of that story is it tripped the breaker because of too much draw and they were dead in the water. Good news for them is I had 2 dedicated 20 AMP circuits put in my garage when I built the house. They hooked up to those and were back in business. What does this have to do with A.I., well A.I. workloads are my detailing crew and current datacenters are the shared 15 AMP circuit in the front yard that can’t handle the load.

The legacy standard for datacenters has been 10-25 kWh racks to support “normal” IT operations. A.I. workloads will initially require at least 50 kWh per rack. I recently had a customer purchase an A.I./HPC solution and they couldn’t turn on their $10M shiny new toy because they didn’t have enough power in the datacenter. Someone must have missed a meeting or memo, or they weren’t concerned about not having enough power. Future A.I. workloads will very quickly get into the 100-200 kWh range per rack and as A.I. continues to advance the energy demands will only increase. The cherry on top is that current datacenters were not built to handle power increases of this type and the investment required to upgrade these facilities might cost as much and the building.

CoLo’s are being built to support 100-200 kWh racks but is that the future? At the current time are CoLo’s are equipped for the future? As we all pull our power from the grid does the grid have enough power to fuel the coming A.I. surge. So, what to do? How do we introduce the 20 AMP dedicated circuits with the electricity behind them to get cars cleaned at a national scale?

Believe it or not the answer to this problem is micro nuclear reactors. Without significant upgrades to the United States power grid through increases in solar, nuclear or fossil fuels A.I. will not have enough power to run at a national level, which might not be a bad thing given that whole SkyNet risk 😊. Micro nuclear reactors are the best path forward for A.I. and Rolls Royce and others might be on the right track.

Rolls-Royce is developing a Micro-Reactor designed to provide clean, affordable, and reliable power. Here are some key points about their micro nuclear reactor:

  • High-Power Density: The reactor has a high-power density, making it capable of reliably supporting various operational demands.
  • Scalability and Transportability: It is designed to be scalable and easily transportable by rail, sea, and even space.
  • Safety: The reactor uses an inherently safe and robust fuel form, with each particle of uranium surrounded by multiple protective layers.
  • Versatility: It can provide both power and heat output, making it suitable for a wide range of applications.

Energy… A challenge for A.I. workloads that might not be currently talked about.

Scroll to Top