Breakingviews - AI struggles to model for angry Main Street mobs

Maine advanced a proposal to ban new data centers larger than 20 megawatts until late 2027, underscoring escalating regulatory resistance to AI infrastructure. More than 300 state bills have been introduced in the first six weeks of 2026, while pushback reportedly blocked or stalled about $160 billion of projects last year. The article highlights growing ratepayer concerns and rising bottlenecks for developers as demand shifts to states like Texas, Indiana and Ohio.

Analysis

The market is underpricing how quickly siting risk can turn into capital intensity risk for hyperscalers. The immediate issue is not demand for AI compute; it is that every incremental megawatt is now facing a higher approval hurdle, which elongates project timelines and pushes more spend into pre-construction, interconnection, and local concessions before revenue is visible. That mechanically lowers near-term free cash flow quality even if headline capex remains unchanged.

The second-order winner is the power-delivery ecosystem: regulated utilities, grid equipment, and transmission-linked contractors should see a longer runway as the bottleneck shifts from server procurement to electrons and wires. But the more interesting effect is that utilities may be incentivized to overbuild under guaranteed returns, transferring option value from developers to ratepayers and regulators. That raises the probability of future legislative crackdowns on allowed returns or cost recovery, which is a medium-term headwind for utility multiples if this becomes politically salient.

For the large-cap platforms, the equity risk is less about near-term earnings and more about narrative compression. If investors start discounting slower incremental data-center deployment, then AI monetization assumptions get pushed out even if model adoption stays strong. That creates a gap between capex intensity and revenue realization, which is exactly the kind of mismatch that can compress growth-multiple stocks for 6-12 months without any earnings miss.

The contrarian view is that the backlash may actually improve project economics by killing speculative capacity and forcing demand discipline. If AEP-style take-or-pay rules spread, the market could ultimately reward the survivors with fewer wasted dollars and tighter supply, while punishing only the weakly committed projects. So this is not a blanket bearish AI call; it is a dispersion trade between firms with secured power and those still shopping for permits.

Key Decisions for Investors

• Short AMZN/META/MSFT basket on rallies over the next 1-3 months; thesis is multiple compression from delayed AI capacity rollout rather than earnings risk. Use a defined-risk structure such as 3-6 month put spreads to avoid fighting any near-term AI sentiment squeeze.
• Long utility/grid enablers versus hyperscalers: buy a basket of regulated utilities with explicit load growth exposure and grid buildout beneficiaries, funded by a short in the most capex-sensitive large-cap AI names. Best expressed as a pair over 6-12 months.
• Buy call spreads on electrical equipment and transmission beneficiaries for the next 2-4 quarters; the key trade is that even stalled data centers still force grid spend, so the wire-and-transformer chain captures the spend shift before the compute stack does.
• Avoid naked shorting hyperscalers; instead, hedge with index calls or long semis if AI enthusiasm re-accelerates. The squeeze risk is high because any large-cap AI capex guide-up can quickly reverse the trade.
• Watch for state-level adoption of take-or-pay or high-load deposit rules over the next 90 days; that would be a positive catalyst for utilities and a negative catalyst for speculative data-center developers, making the short/basket pair more attractive.