AI power: Top 10 tech titans unleash gas, nuke & solar power plants to fuel the bot boom
Grid meltdown amid data centre buildup sees billions poured into power self-generation
Every like or heart ❤️ you tap on your smartphone, every binge-watched video 🎥,, every meme, text, image, or animation you share — powers up from a data centre humming somewhere.
These giant high-security warehouses crossed with sci-fi server farms are the present-day equivalent of mainframe computer rooms from the 1940s-1970s.
Power hogs
To say they are energy-intensive is an understatement: the AI boom is bigger than the internet itself.
The vast, humming halls of precision-engineered chaos that power AI systems under strict climate control guzzle electricity due to stacks energy-intensive AI and "crypto-mining" servers.
And it's just the beginning.
S&P Global estimates a massive 11.3-Gigawatt (GW) data centre demand rise in 2025 alone; a report by Coeli, which runs energy funds, calls this bring-your-own power rewiring an "all-of-the-above" necessity.
Demand explosion
In general, servers guzzles up to 60%, while cooling them and the data warehouse takes 7-30%. AI models like GPT-3 training alone used 1,287 MWh; inference scales massively.
Amid AI's explosion, a key challenge emerges: where to get the power to cover 100%+ annual compute growth?
The AI power crunch hit like a plot twist no one saw coming — a sort of tech "black swan" moment — but one that serves up mega opportunities.
According to International Energy Agency (IEA) estimates, the 945 TWh additional electricity demand from data centres by 2030 is a realistic figure.
At the higher end of the IEA estimate, power infrastructure demand could go up to 1,000+ TWh by 2030.
How on earth can it be ramped in such a short span of time?
For scale, the US generated around 4,178 TWh (Terawatt-hours) of electricity in 2023, a capacity that took 141+ years to build (since 1882), as per the US Energy Information Administration (EIA).
Data centre energy demand
US projections
2024: 183 TWh (4-4.4% national electricity)
2026: ~300-400 TWh (rising demand)
2028: 6.7-12% of US total; 426 TWh by 2030 (+133%)
Capacity: 75.8 GW (2026), 108 GW (2028)
Global projections
2022: 460 TWh (1% global)
2026: ~1,050 TWh
2030: 945 TWh
AI share: 5-15% now, 35-50% by 2030.
Due to potential grid meltdowns from the spike in data centre demand, tech giants are racing to generate their own power.
As AI training and inference devour massive energy — data centres could consume 8-10% of US power by 2030 — while grids face delays in connections (3-5 years).
Self-generation via gas turbines, nuclear restarts, and renewables ensures 24/7 uptime without utility bottlenecks.
Top 10 tech companies
AI scales, this hybrid on-site/grid trend will continue. Following are the top 10 companies, ranked by announced investments/projects (2024-2026 data).
| Rank | Company | Key Projects/Details |
|---|---|---|
| 1 | Microsoft | Restarting Three Mile Island nuclear (835MW); nuclear deals for 10GW+ |
| 2 | Amazon | X-energy SMRs (320MW+); Meta Hyperion gas (2.2GW via Entergy) |
| 3 | Geothermal, SMR RFPs (1-4GW nuclear); small modular reactors | |
| 4 | Meta | Louisiana gas plants (2.2GW); nuclear partnerships |
| 5 | OpenAI | 29 gas turbines (986MW total) for Stargate DC in Texas |
| 6 | Oracle | Gas turbines for AI campuses (scale undisclosed) |
| 7 | xAI (Musk) | Gas turbines similar to OpenAI (early adopter) |
| 8 | Nvidia | Power-dense AI chips driving partner DCs to on-site gen |
| 9 | Apple | Renewables + storage for facilities (hyperscale push) |
| 10 | Equinix | On-site nat gas turbines w/ carbon capture |
Sign up for the Daily Briefing
Get the latest news and updates straight to your inbox
