Start with the schedule, because schedule is what the engineering in this buildout is actually buying. According to a Wall Street Journal survey of suppliers relayed by Wolf Street, 3M's fiber connectors take six months out of a build [11], the modular electrical and mechanical rooms from Clayco and Turner's xPL Offsite take out several months [12], August Robotics' rack-anchoring drill takes out six weeks [13], and Amrize's modelled concrete mixes take out several weeks [14]; sum them generously, ignore the fact that some of them run in parallel, and you get roughly eleven months of compression [6]. New York's moratorium is twelve [3].
The capital has not noticed. July's shell spending ran at a $75 billion seasonally adjusted annual rate [7], which puts the same month a year earlier near $47.8 billion [1] and January 2021 near $9.2 billion [4], a compound rate of about 46% a year across five and a half years [5]; the single step up from June added roughly $4.4 billion to the annual rate [2], and twelve more steps that size would double the line [3]. That series counts the buildings, the improvements around them and integrated systems such as HVAC [9], so at about 7.5% of the announced $1 trillion program per year [7] it is measuring the cheap end of a bill whose expensive end is some multiple of it.
Which is where interconnection stops being a permitting matter and becomes an accounting one. Campuses specified at multiple gigawatts cannot be served by a grid asked to produce that power all at once [5], so speed in the shell buys an earlier depreciation start against a revenue date set by somebody else's queue. This is probably wrong, but the more interesting version of the thesis is that what gets stranded is not the building, it is the schedule, and a year of finished empty concrete is valued very differently by an equity market that has been paying for growth. The industry's own hedging tells you what time costs: the gas turbine bottleneck narrows to blades and vanes [15], retired jet engines are being repurposed into on-site generators [16], Musk bought APR Energy in July for its turbine gensets [17], and electricians are now scarce enough to be a constraint of their own [20]. Wolf Richter, who compiled the construction figures, asks where the trillions of dollars in matching revenue are supposed to come from and answers that no one knows [6].
The counter-thesis is that consent is local rather than binding: pauses and bans across dozens of states [2] move projects to permissive counties, and the aggregate print keeps stepping regardless. That is testable within a few releases. If the national number holds something near a 6% monthly cadence [7] straight through New York's year off, then consent is a siting tax and I am wrong about the ceiling; if the prints flatten while the cost of a delivered megawatt keeps climbing, the adjustment arrives in valuations well before it arrives in concrete.
