By the end of 2025, artificial intelligence (AI) may consume more electricity than Bitcoin mining, a phenomenon that highlights a faster and stealthier energy crisis. The culprit? The explosive growth of large-scale AI models and the data centers that support them.
The Emerging Titans of Energy Demand
Recent research led by Alex de Vries–Gao at the Vrije Universiteit Amsterdam suggests AI may command nearly half of all data-center electricity by late 2025, rivaling and potentially surpassing Bitcoin’s power usage. To assess this, de Vries–Gao tracked AI chip supply through a “triangulation” method, examining industry production data, analyst reports, and earnings calls.
His findings: AI already drives around 20% of data-center electricity; by year-end, demand could hit 23 gigawatts, equivalent to the power consumption of the Netherlands or even the UK.
Why Does AI’s Power Use Matter?
Three big reasons:
A Power Grid Under Strain
Meeting this surge isn’t just a tech problem; it’s a national infrastructure challenge. In the U.S., utilities are scrambling to build new gas plants or revisit nuclear projects to sate AI’s growing appetite.
Daniel Bresette, President of the Environmental and Energy Study Institute, recently touched on this point by emphasizing the urgency of grid upgrades to “accommodate new-generation assets” in the near future.
Environmental Fallout
Unlike Bitcoin, whose network-wide energy use is public, AI’s consumption is opaque. Major tech firms report growing energy and emissions tied to AI, but they rarely break it down. It’s a black box where projections range widely, but the absence of transparency should concern us all.
A Guardian analysis by Vries-Gao estimates AI could account for up to 49 % of data center power use by late 2025, expressing concern over the lack of transparency and sustainability.
Efficiency Paradox
Despite improvements in chip efficiency, such as Nvidia’s Grace Blackwell processors, AI’s total energy consumption keeps climbing. Thanks to the Jevons Paradox, increased efficiency only fuels greater usage.
Understanding the Numbers
Let’s break it down:
2024: AI accounts for about 20% of data-center electricity consumption.
2025 projection: Demand may surge to 23 GW, on par with entire small-to-mid-size countries.
Bitcoin compared: Bitcoin mining uses approximately 10 GW, so AI could more than double that.
Data center growth: Global data-center electricity usage could reach 945–1,050 TWh by 2030, double that of 2022.
Carbon cost: AI alone may consume up to 82 TWh in 2025, similar to Switzerland’s annual power use.
The Drivers Behind This Surge
How did it get so high so fast?
The “Bigger Is Better” Mentality
Competition among tech giants and startups has spurred massive model scaling and bigger neural networks that require vastly more computing. This escalation in size translates directly into exponential energy use.
Generative AI’s Impact
Generative tools, such as ChatGPT and others, have made AI ubiquitous, triggering millions of energy-intensive queries daily. Each query uses far more power than a typical Google search.
Referring to a recent U.S. Department of Energy report, Joanna Stern highlighted in WSJ that such facilities could account for up to 12% of U.S. electricity use by 2028, with OpenAI claiming a 0.34 Wh per ChatGPT query.
Hardware Supply Constraints
TSMC’s output of AI chips has doubled, yet with limited capacity. This bottleneck drives up both chip prices and energy use as data centers rush to run more GPUs for AI workloads.
Ripple Effects Across Infrastructure
What does it mean for the infrastructure?
Stress on Electricity Infrastructure
Regions rich in data centers, such as Northern Virginia or Silicon Valley, are witnessing electricity price spikes, strained grid capacity, and prolonged wait times for new hookups.
MIT scholars Elsa Olivetti & Adam Zewe highlight that generative AI deployment pressures both electricity and water systems, and “you have to have them always on”.
Skeptical Renewables Promise
While efficiency and renewables are part of the solution, de Vries–Gao warns: without concrete measurement, we risk underestimating the scale of energy needed or overestimating clean energy’s ability to keep up.
Environmental and Policy Implications
Safe to say, AI’s energy demand hasn’t come in the best of times, creating a big hurdle towards a green future.
Emissions Trajectory
Emissions from tech firms rose sharply: Google’s carbon footprint jumped by 59% since 2019, largely driven by AI workloads. As Helena Horton notes in a Guardian report, electricity consumption rose by 27% year‑over‑year, driven by AI systems like Gemini and ChatGPT.
Water Use Considerations
Data centers don’t just eat power; they guzzle water. For example, training GPT‑3 consumed ~700,000 liters during one run.
The Disclosure Gap
Policy efforts like the U.S. Artificial Intelligence Environmental Impacts Act of 2024 are pushing for transparency. Still, at present, major AI companies rarely report AI-specific energy figures.
Lessons from Bitcoin’s Energy Pivot
Bitcoin’s network once used more electricity than the Netherlands. When Ethereum transitioned to proof‑of‑stake, its energy use dropped by over 99.9%. AI must learn from that roadmap: structural reforms, not just efficiency gains, are necessary.
Strategies for Mitigation
Now that we understand the roadblocks, what can we do to remove them?
Transparent Reporting
Companies must disclose granular AI energy data, much like Bitcoin’s Open-Source mining statistics, to enable accountability and guide improvements.
Efficiency by Design
Model and hardware innovation must focus on sustainable progress, balancing energy demand with capability. DeepSeek’s more efficient LLM is a promising example.
Policy and Standards
Regulatory bodies should require sustainability assessments and set caps or incentives for green AI operations, especially in sensitive regions.
Renewable Scaling
Major energy stacks must match AI growth. This means broadening renewable deployment, nuclear revival for baseload, and modernization of grids .
The Takeaway
AI’s meteoric rise brings transformative potential and a deepening energy crisis. By 2025, AI may overtake Bitcoin in electricity use, accelerate emissions, strain grids, and impact climate goals. But this isn’t inevitable.
To emerge sustainably, the AI industry must pivot: transparency, efficiency, strategic renewables, and policy coherence. If it does, AI can fulfill its promise without compromising the planet.
The post AI’s Soaring Power Demands: Surpassing Bitcoin Mining By 2025 appeared first on Metaverse Post.
