Four-year grid connection waits make the power date the first question in US site selection

Vicente Garrido, chief financial officer of the Texas power infrastructure developer GoodPeak, described a site that changed category partway through development ^[11]. "Our Houston campus started with greenfield development for battery storage," he said. "As we worked through the site, we realized it also had access to 138 and 345 kV transmission, a major gas pipeline, and dark fiber. That's when we started looking at the land differently." ^[12]

GoodPeak found the transmission after it had picked the site. The high-voltage lines and large pipelines that now sort parcels were routed decades ago to serve refineries, towns and industries that have since changed shape ^[15]. That routing is why two adjacent parcels with the same soil, zoning and highway access can now host different classes of user ^[16].

Data centers used about 176 terawatt-hours in 2023, or 4.4 percent of US electricity ^[2]. The Next Web's analysis argues the binding limit is narrower than total supply: how many specific places can receive a great deal of power on a date somebody will lend against ^[17].

What site buyers actually ask for has four parts. They want a defined quantity of power at one point on the map, by a date, delivered in a way that survives the loss of a piece of equipment ^[6]. JLL's finding that speed to power is "the primary criteria driving site selection" ^[3] says the date is now the part that decides. A wait of more than four years is at least 48 months. Against CBRE's 18-to-36-month window, a buyer in a major-market queue is 12 to 30 months short ^[1].

What teams tell themselves is that this is an AI story. The demand data is broader. In the early-2026 site-selection survey, 82.6 percent rated power at scale very important, the highest intensity score of any infrastructure factor ^[7]. Those respondents were also planning chemical plants, battery factories and distribution centers ^[8]. ERCOT's projected 2030 peak sits at least 59 GW above this July's record, about 65 percent higher ^[2]. Its plan folds data centers in with hydrogen, crypto mining and electrification without splitting them out ^[9].

Garrido has a test for the "powered land" label. "Powered land should mean more than proximity to infrastructure," he said. "It should mean a site with a credible path to a defined amount of power on a reasonably clear timeline. The more certainty around capacity and timing, the more mature the asset." ^[13]

The evidence for the land-price half of the argument is thinner. The Next Web says the gap is repricing American land ^[17], but the article does not include sale prices, premiums, or any lender's view of what delivery date it will finance.

Garrido's two variables make a 2x2 for anyone handed a "powered" listing. One axis asks whether capacity is a defined megawatt figure or a description of nearby lines. The other asks whether delivery lands inside 36 months or out in the four-year queue ^[1]. Defined capacity inside 36 months is powered land in Garrido's sense ^[13]. Defined capacity on a queue date is a queue position, and I'd price it as one. A near date without a defined figure has nothing behind it yet. A listing with neither is proximity, the thing Garrido says the label should mean more than ^[13]. In my view a premium belongs only in the first box, and only after the site passes the fourth test, delivery that survives losing a piece of equipment ^[6]. The tradeoff is location: CBRE's buyers found that box by moving into smaller, less obvious markets ^[5].