The AI infrastructure boom is colliding with a stark physical constraint: power. According to new projections, global data center electricity consumption is expected to quadruple by 2035, with new facilities built between now and 2033 poised to consume as much electricity as India — the world's third-largest energy consumer — uses today.
Why Demand Is Surging So Fast
The primary driver is no mystery. AI training and inference workloads are far more energy-intensive than traditional cloud computing tasks. A single large language model training run can consume hundreds of megawatt-hours, and as model sizes grow and inference scales to hundreds of millions of users, the cumulative load compounds rapidly.
But it's not just AI. The broader digitization of enterprise infrastructure, the proliferation of edge computing, and continued cloud migration from legacy on-premise systems are all adding baseline load on top of AI's spike.
The India Comparison Is More Than a Talking Point
Framing the projected consumption against India's national electricity use is a deliberate signal of magnitude. India powers roughly 1.4 billion people across industrial, residential, and agricultural use — placing new data center construction in that context underscores that this isn't incremental growth. It's a civilizational-scale energy commitment baked into infrastructure decisions being made right now.
This creates compounding pressure across several dimensions:
- Grid infrastructure: Many regions simply lack the transmission capacity to support planned data center clusters, especially in markets like Northern Virginia, Dublin, and Singapore that are already heavily saturated.
- Renewable energy procurement: Hyperscalers have made aggressive net-zero commitments, but sourcing sufficient clean power at this scale — and ensuring it's available on-demand rather than intermittently — remains unresolved.
- Water usage: Cooling systems tied to this electricity demand also carry significant water footprints, adding another regulatory and environmental variable.
What This Means for the Broader Tech Ecosystem
For hyperscalers like Microsoft, Google, Amazon, and Meta, the implications are already being felt in capex. Microsoft alone has signaled over $80 billion in data center investment for fiscal 2025. These companies are increasingly functioning as quasi-utilities, and their ability to secure power agreements is becoming a core competitive moat.
For AI startups and SaaS founders, this matters in less obvious but equally important ways. Cloud compute costs are unlikely to fall as quickly as they historically have if power constraints slow new capacity additions. Founders building inference-heavy products should model compute costs conservatively and consider architectures that optimize for energy efficiency — not just latency or throughput.
There's also a geographic dimension opening up. Data center development is accelerating in emerging markets — parts of Southeast Asia, the Middle East, and Latin America — where power infrastructure is being built fresh rather than retrofitted. This could shift where AI compute is cheapest to run within the next decade.
The Policy and Investment Landscape
Governments are beginning to treat data center power demand as a national strategic issue. The EU has proposed stricter energy efficiency reporting requirements for large facilities. In the US, several states have introduced legislation to scrutinize data center tax incentives in light of grid strain.
For investors, energy access is increasingly a diligence item — not just for infrastructure funds, but for anyone backing companies whose unit economics depend on affordable compute. The 4x projection isn't a warning to slow down; it's a signal that whoever controls the power supply controls the next decade of AI infrastructure.
