Microsoft has signed a 20-year nuclear power purchase agreement dedicated to powering AI data centers.
Why Nuclear Power Is Needed
A single cluster of H100 GPUs can consume megawatts of power. Microsoft plans to build a training cluster with 500,000 GPUs by 2027—the electricity demand at this scale is equivalent to that of a mid-sized city.
Wind and solar power are unreliable (output drops sharply at night and on cloudy days), and data centers require 24/7 uninterrupted power. Nuclear power is currently the only solution that can stably provide zero-carbon electricity.
What Other Tech Giants Are Doing
- Google: Signed deals for geothermal and small modular nuclear reactors
- Amazon: Purchased a nuclear-powered data center campus in Pennsylvania
- Meta: Building a data center in Texas paired with wind power and energy storage
The Hidden Cost of AI
When people discuss AI, they talk about model size and parameter counts, but rarely mention energy consumption. Yet training a GPT-5-level model reportedly produces carbon emissions equivalent to 500 cars running for a full year.
Microsoft's environmental commitment is to achieve "carbon negative" status by 2030. Signing the nuclear power deal is a way to expand AI computing capacity without contradicting that pledge.
The future of AI depends not only on algorithms, but also on whether we can find clean energy that's cheap enough.
References:
Comments
Comments are closed.