The AI race fuelling the future of energy

Last week, news broke that Google's carbon emissions were up 50% due to AI energy demands, putting their 2030 net-zero target in question.

As AI systems grow more sophisticated, they're creating increased demand for energy. This surge is pushing tech companies to develop new power strategies, initiating a race to secure sustainable energy sources that can support AI growth while maintaining their climate goals.

The energy challenge

Electricity use for data centres driven by demand from AI workloads, is projected to double between 2022 and 2026. This growth drives major AI players to invest in clean energy, with companies like major companies making large investments in renewable power.

Diversification strategy for energy sources by major AI players

Tech companies are pursuing various strategies to meet their energy needs:

1. Nuclear Fission: Amazon Web Services is investing $650 million in a data center powered by a nuclear plant, accessing 960 megawatts of carbon-free energy. OpenAI’s Sam Altman has invested in nuclear fission startups like Oklo. Bill Gates has invested $1 billion into a nuclear fission startup TerraPower that broke ground on its first nuclear plant in Wyoming.

2. Nuclear Fusion: Microsoft has signed a power purchase agreement with fusion startup Helion Energy, despite commercial fusion not yet being available. Sam Altman personally invested $375 million in Helion.

3. Diversifying energy sources: Companies are looking at a mix of solar, wind, nuclear, and potential breakthrough technologies like fusion to meet their growing energy needs.

4. Global impact: Companies are securing clean energy deals across multiple regions, including the U.S., Europe, Asia, and Latin America.

Looking to the future

Energy is one of three major constraints for scaling AI. The energy challenge represents one of three significant constraints currently facing the scaling of AI technologies:

1. Energy : AI systems' increasing energy demands are pushing companies to seek innovative and sustainable power solutions.

2. Chip hardware: Another limiting factor is the development and production of specialised AI chips. Companies are investing in advanced semiconductor technologies to meet AI's computational needs.

3. Model architecture limitations: Current transformer-based models are approaching their performance ceiling. A new class of models that can surpass transformer limitations is needed, potentially advancing AI capabilities. e.g. Bill Gates says scaling AI systems will work for two more iterations and after that the next big frontier is meta-cognition where AI can reason about its tasks

These three constraints are interlinked. More advanced chip hardware and new model architectures could increase energy efficiency, while more abundant and cheaper energy could allow for larger and more complex models. Addressing these challenges simultaneously is important for the continued advancement of AI technology.

AI's growing energy needs are driving clean power innovation. Tech industry investments are accelerating sustainable energy adoption. This trend links AI growth with climate action, creating a race for sustainable AI power sources. This pursuit advances renewables, nuclear tech, and energy storage. Companies effectively managing their energy needs may significantly influence future energy systems.