Qualify AI demand: megawatts or management?
Brokers are being advised to split AI occupiers by whether their constraint is raw power (megawatts) or coordination and talent (management). If a tenant’s bottleneck is compute capacity, the right locations may be out‑of‑market; if it’s hiring, product velocity, or enterprise access, Bay Area locations remain relevant. (arstechnica.com) (motherjones.com)
The first question in AI real estate is no longer “How much space do you need?” It is whether the tenant is short on megawatts or short on managers, engineers, and customers. (jll.com) Brokerage reports in early 2026 describe a market where power delivery, not land price, is setting the map for new data center deals. JLL said power “will be the primary site selection criteria” because grid connections now come with multiyear waits, and Cushman & Wakefield said growth is being shaped by power availability, regulation, and infrastructure readiness. (jll.com) (cushmanwakefield.com) That makes compute-heavy AI tenants different from office-heavy AI tenants. A company trying to secure large blocks of capacity may have to follow substations, transmission lines, and backup generation into secondary markets, while a company trying to hire researchers or meet enterprise customers can still justify paying Bay Area rents. (jll.com) (cushmanwakefield.com) The supply crunch is not theoretical. Ars Technica reported this week that nearly 40 percent of U.S. data centers planned to open in 2026 are facing delays, based on Financial Times reporting that used satellite imagery and project analysis to track construction progress. (arstechnica.com) CBRE’s Silicon Valley data center report already showed what that looks like on the ground. Leasing in the market nearly doubled in 2024, net absorption rose to 45.2 megawatts from 25.7 megawatts a year earlier, 167.8 megawatts were under construction, and 78 percent of that pipeline was pre-leased. (cbre.com) CBRE also said Silicon Valley tenants are enduring power delays of five years or more to stay near the San Francisco Bay Area. The firm said most requirements there are now 10 megawatts or less because larger blocks of new capacity are scarce. (cbre.com) The Bay Area still holds a different kind of advantage: people. CBRE’s 2025 tech talent rankings put the region at about one-fifth of U.S. artificial intelligence workers, and a September 2025 market report said the Bay Area had 76,079 AI-skilled tech workers, up 24 percent from 61,497 a year earlier. (costar.com) (connectcre.com) Office demand has followed that talent base. CBRE said the San Francisco Bay Area accounted for 14 of the 100 largest U.S. office leases in 2025, totaling 4.3 million square feet, and 10 of those 14 deals were signed by tech companies. (cbre.com) The result is a split-screen market. Training clusters and hyperscale facilities are chasing power in places that can actually deliver it, while product teams, sales groups, and startup founders are still clustering in San Francisco and Silicon Valley. (jll.com) (cbre.com) Mother Jones illustrated the power side of that divide with projects like the Stargate-linked site in Abilene, Texas, and Meta’s proposed Hyperion campus in Louisiana, which it reported would require its own plants, lines, roads, water systems, and worker housing. Those are not office searches with bigger electric bills; they are industrial buildouts that can reshape entire counties. (motherjones.com) For landlords and brokers, that means “AI tenant” is now too broad to be useful. The occupier that needs 100 megawatts may belong in Texas, Louisiana, or another out-of-market power node, while the occupier that needs a recruiting pipeline and fast customer meetings may still belong on the Peninsula. (jll.com) (cbre.com)
