PANTHERFIELD RESEARCHPower at the point of compute
Flagship thesis · 29 September 2026 · AI infrastructure
The scarce asset in AI infrastructure is not electricity in the abstract. It is reliable power deliverable to a qualified compute site, at an economic price, on the date it is needed.
The investment thesis
AI capacity is being built on a timeline measured in quarters. Grid connections, power equipment and permitting move on different clocks. The research case is selective: assets with verified access to power, a workable cost of service and customers able to pay for that capacity. A broad bet that every generator or data centre wins from AI demand is not the thesis.
Evidence that anchors the view
Where to look
Screen for sites with binding interconnection milestones, traceable equipment orders, buildable cooling and credible off-take. Then examine grid equipment, power distribution, storage and flexible-load systems that solve a defined constraint. A pipeline of announced megawatts is not the same as contracted, energized, billable megawatts.
The economic test is blunt: revenue and utilization must support the full delivered-power cost, construction capital, curtailment risk and a margin of safety. Evaluate each stage of the power chain, not the AI label attached to it.
Underwriting questions
Countercase and what changes the call
Efficiency per AI task is improving quickly, and project pipelines can double-count requests. Capital spending may outrun profitable AI use; local opposition, water constraints, tariffs and financing can impair delivery. The IEA's 2026 update says bottlenecks make the most aggressive near-term demand cases less likely even as the central case remains strong. These are reasons to demand project-level proof, not to ignore the structural shift.
Reduce conviction if successive IEA updates show sustained demand deceleration, energization and contracted utilization fall well short of announced capacity, equipment lead times normalize without a durable service premium, or tariff changes transfer economics away from the project. Increase conviction only when independently verified connections, off-take and unit economics improve together.
Method and sources
Desk research using primary energy-system analysis, a national laboratory scenario report, industry grid modelling and a regulator's order. No site, contract, tariff, off-take or management interview was independently verified for a specific investment. Figures are source estimates or scenarios, not forecasts by Pantherfield.