Is AI helping the climate or hurting it?

The central tension is between the potential of AI to speed up scientific discovery and the massive amount of electricity required to build it. To train the large models that power AI, companies need huge data centers, which are essentially warehouses filled with thousands of computers running 24/7. These facilities require so much energy that major tech companies are currently buying up all the electricity they can find. This has led to a spike in investment for renewable energy, such as nuclear and geothermal, but it has also caused companies like Google, Meta, and Microsoft to see their total greenhouse-gas emissions rise for the first time in years.

To understand why AI is such a power-hungry beast, consider the process of training. When a company builds an AI system, they show it massive amounts of data—billions of books, articles, and images—so the system can learn patterns. This is not a one-time event; it is an intensive, months-long computational process that requires thousands of specialized computer chips working in unison. Once a model is trained, it doesn't stop working. Every time you ask a question, the model has to perform a series of rapid calculations to generate a response. This process, called inference, happens every second of every day for millions of users at once. These systems are never truly off, and the more popular they become, the more energy their data centers must consume to keep the processors running.

The core issue is timing. While AI research might one day help us discover new materials for batteries or better carbon-capture technology, the energy demand is happening right now. To meet this immediate need, companies are often turning to natural gas power plants because they are quick to build compared to solar or wind farms. A natural gas plant, once built, will likely run for decades, effectively locking in years of carbon emissions. We are currently witnessing a massive, messy experiment: we are racing to build the future of intelligence, while simultaneously struggling to figure out how to keep the lights on without further warming the planet. The question is no longer just what AI can do for us, but what we are willing to sacrifice to power it.