The Deep Dive
For two years, the AI infrastructure narrative centered on semiconductor scarcity. TSMC's 2nm and 3nm capacity sold out through 2027. Chip supply was the chokepoint. That story is now obsolete.
The real constraint is electricity. Goldman Sachs forecasts data center power demand will surge 220% by 2030, with domestic capacity projected to reach 95 gigawatts—a 197% increase from 2025 levels. This is not a gradual ramp. This is exponential load hitting a grid designed for linear growth.
The infrastructure implications are severe. Hyperscale operators are shifting inland away from coastal data centers, chasing cheaper power and grid capacity in regions with available generation. Microsoft's Azure expansion is hitting power bottlenecks that force location strategy reconsideration. The U.S. power grid is struggling to keep pace with rising demand for large-scale facilities requiring consistent, high-capacity energy.
McKinsey and infrastructure consultancies warn that electricity and interconnection delays are becoming binding constraints on hyperscale AI expansion. This is not speculation. This is operational reality.
Enter the Hormuz blockade. Oil prices climbed after the U.S. military announced a shipping blockade against Iranian ports in the Strait of Hormuz, with crude rising above $100/barrel. JPMorgan Chase commodities analysts estimate pre-closure barrels will be exhausted from the global supply chain by April 20. Higher oil prices mean higher natural gas costs, which directly compress data center operating margins and increase the cost of power generation buildout.
The capital intensity of solving this problem is staggering. CoreWeave secured an $8.5 billion delayed draw term loan. Amazon is committing $200 billion in 2026 capex, with a large share supporting AI and cloud infrastructure. But capital alone cannot create generation capacity faster than physics and permitting allow.
This reshapes the entire supply-side calculus. The constraint is no longer manufacturing capacity or chip design. It is kilowatt-hours. Companies that secure long-term power contracts—whether through nuclear PPAs, renewable offtakes, or legacy coal assets—will win. Companies that cannot will be rationed by grid operators or forced to relocate. The economics of AI infrastructure are now fundamentally energy economics.
───
Signal Watch
• Hyperscale Location Arbitrage: Synergy's analysis shows 580 operational hyperscale data centers in the U.S. at end of 2025, with growth shifting inland—a direct response to coastal power constraints. This is capital reallocation in real time, driven by energy availability, not labor or latency.
• Grid Interconnection Bottleneck: Interconnection delays are now a binding constraint alongside electricity availability, meaning even where generation exists, grid connection timelines can stretch projects by years. This is a regulatory and infrastructure problem, not a technology problem.
• Energy Cost Shock from Geopolitics: Oil above $100/barrel due to Hormuz blockade creates upstream pressure on natural gas and power generation costs, directly raising the operational cost of data center expansion and reducing returns on AI infrastructure capex deployed at current pricing.
───
The Bottom Line
Watch power procurement announcements, not chip orders. Companies announcing long-term nuclear or renewable PPAs are positioning for 2027–2030 dominance. Companies relying on spot market power will face margin compression or capacity rationing. The Hormuz blockade is a live tail risk to energy cost assumptions baked into AI infrastructure ROI models—if oil stays above $100, data center operating costs rise materially, and the capex math shifts.
───
Bitcoin Macro
Higher energy costs and grid constraints are deflationary for AI infrastructure returns but inflationary for power prices. Bitcoin mining, which can operate at the margin of available power and relocate to stranded energy, becomes a more rational use of surplus generation capacity. Oil above $100 signals stagflationary pressure; Bitcoin's macro role as a non-correlated store of value strengthens in that regime.
───
The age of energy abundance is over. The age of energy scarcity begins.