top of page

Your certificate is now private

CertificateBackground.png

Certificate of Completion

ErdosHorizontal.png

THIS ACKNOWLEDGES THAT

HAS COMPLETED THE FALL 2026 DATA SCIENCE BOOT CAMP

Vassu Doomra

Roman Holowinsky, PhD

November 15, 2026

DIRECTOR

DATE

clear.png

TEAM

How Reliable Is the Nuclear Power Promised to AI Data Centers?

Vassu Doomra

clear.png

AI data centers run around the clock, and tech companies are turning to nuclear to power them. Google, Amazon, Meta, and Microsoft have signed 13 deals for 9.8 GW of nuclear power. Those deals assume reactors are always on, but reactors don’t always run at full power.

The problem:
Reactors go offline for refueling, maintenance, and unexpected faults. Between 2016 and 2025, average output across US reactors ranged from 74.8% to 95.9% of full power, so the choice of reactor matters. Whenever output drops, the shortfall has to be made up from another source.

Our approach:
We use more than 25 years of NRC (Nuclear Regulatory Commission) daily power data to classify every outage and forecast each reactor’s output one quarter ahead, so data center buyers get a consistent, data-driven view of what to expect.

Screen Shot 2022-06-03 at 11.31.35 AM.png
github URL

©2017-2026 by The Erdős Institute.

bottom of page