AI can't outprompt a shortage of power, water, and land

Forrester forecasts that AI datacenter growth will face rising constraints from electricity, water, land and community opposition, with governments increasingly requiring operators to fund grid upgrades, reserve capacity and provide financial guarantees. PwC estimates cumulative global datacenter capex through 2050 at $22 trillion-$50 trillion, with a $31.6 trillion midpoint. Forrester expects at least two countries by 2027 to impose tariffs or equivalent ratepayer-protection obligations, increasing location-specific cloud costs and power-market risk for operators and enterprise AI users.

Analysis

The investable implication is a shift in AI infrastructure economics from semiconductor-led scarcity to power-access scarcity. Grid interconnection rights, contracted generation, and transmission capacity become quasi-regulated inputs; this favors CEG, VST, NRG, PWR, ETN and VRT more than hyperscalers or commodity colocation providers. The first-order capex cycle remains positive, but mandatory deposits, take-or-pay power contracts, and upgrade contributions raise the cost of idle capacity and make speculative datacenter announcements less valuable to REIT multiples.

For EQIX and DLR, location-specific power costs create a widening dispersion between campuses with firm low-cost electricity and those dependent on congested grids or merchant power. Pricing can be passed through to enterprise customers only with a lag, creating a 1-3 quarter margin-risk window; cloud providers have greater negotiating leverage but will face lower incremental AI ROI if power and transmission costs remain unrecovered. The second-order beneficiary is electrical equipment and EPC capacity: interconnection requirements pull spend forward into transformers, switchgear, substations and transmission before servers are installed.

Consensus is likely too focused on whether permitting eases and not on who bears the financing obligation. Faster permits may accelerate projects, but do not solve transformer shortages, transmission lead times, or utility balance-sheet constraints; indeed, reduced procedural friction could increase competing load requests and worsen congestion. Over 6-18 months, the key differentiator will be disclosed contracted power versus merely announced MW capacity; a rising share of projects with nonrefundable power commitments would validate the grid-equipment and independent-power thesis, while cancellation of large AI campuses or weaker utility load forecasts would falsify it.

FORR is not a clean vehicle for this theme: its research forecast does not create material operating leverage without evidence of incremental subscription demand. Treat the report as a policy-risk alert rather than a catalyst for the stock.

Key Decisions for Investors

• Initiate a 6-12 month pair: long PWR and ETN / short DLR and EQIX in equal dollar amounts. Grid-upgrade obligations and interconnection bottlenecks should improve EPC/electrical backlog visibility while raising colocation capital intensity; target 15-20% relative return, cut if DLR/EQIX demonstrate sustained pricing pass-through with stable adjusted EBITDA margins.
• Accumulate CEG or VST on 5-10% pullbacks over the next 3 months, preferably where new datacenter load is matched by long-duration contracted generation rather than merchant exposure. Target 15% upside from incremental capacity-contract repricing; exit if major utility load forecasts are revised down or nuclear/gas contract realizations fail to improve.
• Use VRT as the higher-beta AI-power derivative only after backlog disclosures show orders converting without extended working-capital deterioration. A 6-month call spread can express upside, but cap premium at 1% of NAV given valuation sensitivity and the risk that datacenter projects are delayed before equipment delivery.
• Create an event watchlist around hyperscaler earnings: flag any disclosure of nonrefundable utility deposits, self-funded transmission, or power-curtailment commitments at AMZN, MSFT, GOOGL and META. Material commitments without corresponding AI revenue monetization would support reducing hyperscaler exposure and adding to the PWR/ETN leg.