An expert perspective by Dan Scarbrough, CCO – Stelia.
AI’s efficiency gains are accelerating — but so is its insatiable demand for power. Is the AI revolution on a collision course with reality?
The numbers are staggering: McKinsey forecast global AI-driven power demand could surge 240GW by 2030 — the equivalent of adding up to 6 new UKs worth of power demand. And that’s before we factor in EVs, electrification of heavy industries such as steel and ammonia production, and AI-driven industrial automation. 
Efficiency vs. Reality: Koomey vs. Jevons
Koomey’s Law tracks relentless efficiency gains: Compute power doubles every 18 months. AI chips get smarter, leaner, faster. Nvidia’s latest chips are claimed to be 25x more power-efficient than the previous generation.
But Jevons Paradox reveals the twist: efficiency fuels exponential usage. The more powerful and cost-effective AI becomes, the more aggressively we deploy it.
📈 Current status? Jevons is winning. AI and compute appetite is growing faster than our ability to optimize. The digital economy’s ultimate challenge is sustainable scaling.
AI’s Hunger for Power, Water, and Data is Multiplying.
🚨 Meta’s new $10B Louisiana AI campus will demand 1,500MW — more than some small countries.
🚨 Microsoft & OpenAI’s Stargate supercluster could need 5GW — enough to power all of London.
🚨 Google, Microsoft, Meta, & Amazon have already signed PPAs for 70GW of clean energy (~40GW in the US alone).
And yet, despite these investments in clean energy:
❌ Google’s carbon emissions are up 48% since 2019.
❌ Microsoft’s are up 29% since 2020.
Will AI Accelerate or Delay Sustainability Outcomes?
As well as reshaping technology AI is colliding with geopolitics, energy strategy, and climate policy. The U.S. under President Trump is doubling down on fossil fuels, revoking clean energy policies, and exiting the Paris Agreement.
Meanwhile, the EU is accelerating decarbonization, with nations like the UK, Finland, and Germany setting ambitious 2035–2045 carbon neutrality goals.
What does this mean for AI infrastructure?
Under Trump’s approach: AI data centers may become more reliant on fossil fuels, benefiting from cheaper, deregulated energy but risking higher emissions and regulatory uncertainty in future administrations.
Under the EU’s strategy: AI will be expected to integrate renewables, energy efficiency mandates, and carbon pricing, increasing sustainability but also infrastructure costs due to grid constraints and regulatory hurdles.
AI’s climate impact is about direct energy use and also about how it reshapes industries. Two possible futures present themselves:
1️ — AI accelerates energy transition: AI optimizes energy grids, accelerates battery breakthroughs, and enhances industrial efficiency.
2️ — AI delays the transition: AI drives higher fossil fuel use, resource demand, and grid strain, slowing decarbonization.
Which path are we on? It’s still too early to tell. 🤔
AI’s Bottleneck Isn’t Just Power — It’s Data Networks.
New grid capacity and generation won’t solve AI’s infrastructure crisis. AI workloads are also pushing data networks beyond their limits — from IoT requiring real time data mobility to distributed inference, the classic Internet architecture is not designed to cope with these new loads and traffic patterns.
It’s clear the AI revolution needs smarter infrastructure. At Stelia, we’re building Hyperband — a suite of intelligent networking solutions to scale AI faster and more responsibly:
🔹 Adaptive AI traffic routing to eliminate bottlenecks.
🔹 Ultra-low-latency interconnects for simple, efficient, scalable AI training & inference.
🔹 Energy-aware networking that maximizes performance per watt.
AI is revolutionizing everything — from industry to energy demand. Can it also fix the climate? Relax — AI is on the case… just don’t ask how much power it needs to do it. ⚡🌍😏
So, are we heading for an AI sustainability crisis — or can intelligent networks balance growth and efficiency?
#AI #Energy #Sustainability #EnterpriseAI #DataCenters
