How to Judge a Power Strategy

In September 2026, the Nuclear Regulatory Commission was still holding public meetings on the restart of the Crane Clean Energy Center, two years after Microsoft signed its purchase agreement. That timeline is not unusual, and it is a fitting place to end a course about data center power. The most important variable in almost every power strategy is the date on which electricity actually flows, and that date is set by institutions and equipment, not by announcements.

This course has covered four layers. The first was demand: the Lawrence Berkeley National Laboratory and International Energy Agency estimates, whose wide ranges reflect real uncertainty about how quickly artificial intelligence computing will grow. The second was the power chain inside the building, from substation to rack, and the redundancy designs, described by the Uptime Institute's tiers, that protect it. The third was supply: utility tariffs, power purchase agreements, and on-site or co-located generation. The fourth was the rack itself, where rising power density is pushing the industry toward higher-voltage direct current. A sound judgment about any project touches all four.

Begin with the connection. Ask whether the project holds a signed agreement with the utility or grid operator, what capacity it covers, and when. Ask what the customer has committed to pay, and for how long. The AEP Ohio tariff approved in 2025, with its minimum payment on reserved capacity, is a model of the terms regulators increasingly expect. A developer who cannot describe these terms plainly probably does not have them.

Next, look at supply. If the project relies on a power purchase agreement, establish whether it is physical or financial, and whether it adds new generation or claims existing output. If it relies on on-site generation, ask about fuel supply, air permits, spare units and what happens during maintenance. The Memphis turbines and the Susquehanna restructuring both show that the permitting and regulatory path can matter as much as the engineering.

Then consider cost across all four of its forms: energy, capacity, delivery and time. A cheap energy price paired with a five-year wait for connection may be worth less than a higher price available next year. A project that has shaved redundancy to save money may be sensible for training workloads that save their progress often, and imprudent for services that must never stop.

Look at the building's design as well. What is its reported power usage effectiveness, how was it measured, and over what period? How much of the facility is concurrently maintainable, and is that certified or asserted? Is the electrical system designed for the rack densities the tenant plans to install, or for those of five years ago? Can it be adapted to higher-voltage direct current without rebuilding, or will it be stranded as racks move toward the megawatt designs described in the Open Compute Project's Mount Diablo specification?

Finally, watch the rules. The Federal Energy Regulatory Commission's December 2025 order on co-location in the territory of PJM Interconnection, the eastern grid operator, state tariffs such as Ohio's, electrical codes that will need to address 800-volt distribution, and nuclear licensing timelines all change the value of a power strategy after it has been chosen. A strategy that works only under today's rules carries a risk worth naming.

None of these questions requires an engineering degree. They require a habit of asking for documents rather than descriptions, and of separating what has been signed from what has been hoped for.

Faraday's read is that the best power strategies are rarely the most novel. They secure a firm connection date early, match their redundancy to their workload, keep more than one path to supply open, and leave room in the electrical design for densities that have not yet arrived. Deeper study of each area builds on exactly these questions, and on the documents that answer them.