When a search-and-advertising company starts investing in fusion reactors, it's worth asking what changed. In early July 2026, Google joined a €411 million ($468 million) funding round for the German startup Proxima Fusion — a Max Planck Institute spin-out targeting what it hopes will be Europe's first commercial fusion power plant. It's the latest and most striking move in a broader trend: the largest technology companies are no longer just buying electricity, they're buying the means to produce it.
Why AI made energy the bottleneck
For most of the cloud era, the scarce resource was compute — chips and the money to buy them. That's shifting. Training and running large AI models consumes electricity on a scale that's straining existing grids. Google's own 2025 environmental report disclosed a 37% year-over-year jump in electricity consumption, the largest in its history, driven mainly by AI data-centre expansion — with its data centres consuming over 42 million megawatt-hours, comparable to the annual usage of a mid-sized country.
When your growth plans are capped not by how many servers you can buy but by how many megawatts you can secure, energy stops being an operating expense and becomes a strategic asset. Hence the scramble.
What the hyperscalers are actually doing
The energy playbook has several plays, and the big providers are running all of them:
- Nuclear fission, via long-term power-purchase agreements with existing plants and, increasingly, commitments to small modular reactors (SMRs) — compact, factory-built units designed to sit near a data centre.
- Nuclear fusion, the longer-horizon bet Google's Proxima investment represents — still pre-commercial (Proxima targets a demonstrator in the early 2030s and a commercial plant later in the decade), but with enough capital now flowing that the timelines are being taken seriously. Google has separately backed US fusion firm Commonwealth Fusion Systems and agreed to buy power from its first commercial plant.
- Renewables at scale, with round-the-clock clean-energy matching rather than simple annual offsets.
- Grid and siting deals — securing land, transmission access, and power contracts before local capacity is exhausted, often in emerging hubs like India and the Nordics.
Why this matters even if you'll never run a data centre
You don't need a reactor to feel the effects. Three consequences reach ordinary companies.
Cloud pricing tracks energy. When compute is gated by power, and power gets more expensive or more contested, that pressure eventually reaches the price of the cloud instances you rent — especially in the hottest regions.
Region and capacity planning matters more. As providers chase power, capacity grows unevenly across regions. Where you deploy — and whether your preferred region has headroom — becomes a real architectural decision, not an afterthought.
Efficiency is now a cost lever, not just good practice. Every unit of compute you don't waste is one you don't pay energy-inflated prices for. Right-sizing, sensible caching, and appropriately scaled models translate directly into lower bills as the energy squeeze tightens.
The bigger picture
This is the clearest sign yet that AI's story has moved from software to physics. The companies buying power plants are betting that whoever controls energy controls the pace of AI — and that bet reshapes the ground everyone else builds on. For most businesses the takeaway isn't to follow them into nuclear; it's to recognise that cloud capacity and cost are becoming energy-driven, and to plan accordingly.
Where Educatifu fits
We help companies build cloud strategies that stay resilient as the economics shift — right-sizing spend, choosing regions wisely, and designing efficient architecture that isn't at the mercy of energy-driven price rises. If you want your infrastructure planned for where the market is going, get in touch.