Forget Nvidia: IREN CEO Says A New AI Factory Built Today May Not Go Live Until 2030

A 1-gigawatt AI factory built today may not bring first compute online until 2030, highlighting that power, transmission, substations and permits are now major bottlenecks in AI infrastructure. The article suggests investors are shifting from chip-centric AI trades toward power and energy infrastructure names such as Bloom Energy as megawatt access becomes as important as GPU supply. The piece is thematic and forward-looking rather than event-driven, so market impact is likely limited but relevant for AI and utility-adjacent stocks.

Analysis

The market is still underestimating how much AI capex is migrating from semiconductors to regulated bottlenecks. Once power, interconnection, and permitting become the gating function, the economic moat shifts toward asset owners with scarce grid access, dispatchable generation, and balance-sheet capacity to pre-fund multi-year projects. That changes the earnings sensitivity: the fastest path to AI monetization may sit in infrastructure names with contractual revenue rather than in the GPU vendor everyone already owns.

This is also a duration story. If meaningful incremental capacity cannot come online until the back half of the decade, then “AI demand” remains structurally ahead of supply for longer than consensus expects, which supports elevated pricing for reliable electricity, behind-the-meter solutions, and transmission equipment for several years. The second-order effect is that compute expansion becomes geographically concentrated around pockets of available power, benefiting incumbents with existing footprints and hurting pure-play data-center developers that rely on greenfield builds.

The contrarian read is that the bottleneck may ultimately compress returns for the whole AI stack: if power adds years to deployment, some hyperscaler and model-development spending gets pushed out, raising the odds of a capex digestion phase before revenue catches up. That matters for NVDA indirectly because a delay in megawatt availability can slow GPU shipment absorption, even if near-term order books remain strong. In other words, the constraint is bullish for infrastructure scarcity trades, but not uniformly bullish for the broader AI complex.

Key Decisions for Investors

• Go long BE or a basket of distributed power / fuel-cell names on 3-6 month horizon; the setup is favorable if the market begins pricing utility-like contracted cash flows rather than cyclical hardware optionality. Risk/reward is attractive, but size modestly because valuations can re-rate violently on any project delay or financing concern.
• Initiate a pair trade: long grid-enabler infrastructure (ETN, PWR, HUBB) vs short a basket of high-beta data-center developers / AI infrastructure leases that depend on rapid greenfield buildout; thesis is that interconnect and transmission spend should precede rack revenue by 12-24 months.
• Use NVDA weakness on any AI capex-delay headlines to express a relative-value hedge, not an outright short; the stock can remain supported by backlog, but the longer the power bottleneck persists, the more likely customers stagger installations and reduce near-term shipment cadence.
• Consider call spreads in utility / power equipment names with 6-12 month maturities, funded by selling upside in the most crowded AI hardware proxies; the trade monetizes the market’s current underweight to the enablers of megawatt delivery rather than compute itself.
• Watch for catalysts in permitting, substation approvals, and utility interconnect queues over the next 2-4 quarters; if there is no improvement, extend duration in power-infrastructure winners, but if regulators accelerate approvals, trim because the scarcity premium will fade quickly.