type: insight tags: [ai-infrastructure, moat, power, data-center, neocloud, capacity, competitive-advantage] confidence: medium created: 2026-02-25 source: IREN earnings-review Q2_FY26 persona: gaucho provenance: legacy source_analysis_path: null source_paragraph_quote: null source_transcript_span: null source_loss_log_path: null

Power Capacity Is the Scarce Moat in AI Infrastructure — Not GPUs or Capital

In the AI infrastructure buildout, physical grid power (measured in GW) is the binding constraint that creates durable competitive moats. GPUs are financeable commodities; capital is available at scale once a hyperscaler contract is in place. Grid power access is neither — it requires years of site development, regulatory approvals, utility relationships, and infrastructure construction that cannot be accelerated with money alone.

Evidence

• IREN: 4.5 GW of secured power capacity accumulated over 7 years. Microsoft selected IREN — not CoreWeave, not DataBank — for its largest HPC outsourcing contract ($9.7B TCV). The differentiator cited by management: power scale that hyperscalers cannot source independently at this speed.
• CoreWeave has GPUs at scale but not grid power at comparable scale. The inverse asymmetry (GPU-rich, power-constrained vs power-rich, GPU-financeable) matters for competitive positioning.
• IREN's $9.2B in committed funding was unlocked by the power asset, not the reverse — power enabled the hyperscaler relationship, which enabled the financing.

Implication

When evaluating AI infrastructure companies, rank moat sources: (1) secured grid power (GW, location, timeline to energize) — hardest to replicate, (2) hyperscaler anchor contracts — durable but achievable, (3) GPU access — financeable, not a moat. A company with 2+ GW of energized or near-term power has a multi-year head start that capital cannot close quickly. Track GW secured, GW energized, and GW pipeline as primary moat metrics ahead of GPU count. For competitive analysis, ask: who has the power, not who has the GPUs.