Across California, some of the most power-hungry buildings ever permitted are rising on former farmland and warehouse tracts, yet many of them may sit largely idle for years. The race to build data centers for artificial intelligence and cloud computing has collided with grid bottlenecks, water limits, and local pushback, creating a landscape where capacity is being constructed far faster than it can be fully used.
I see a widening gap between the physical scale of these complexes and the practical constraints that will govern when, and how much, they can actually operate. The result is a new kind of stranded asset risk, where billion‑dollar server farms are technically complete but throttled by power queues, regulatory conditions, and infrastructure that has not caught up.
California's AI land rush meets a hard power ceiling
The current wave of California data center projects is driven by an aggressive push from hyperscalers and AI developers to secure space and grid connections before competitors do. Developers are locking up hundreds of megawatts of capacity in regions like the Central Valley and the outskirts of the Bay Area, betting that demand for model training and cloud services will justify the buildout. Yet the state's transmission system is already strained, and interconnection queues for large new loads stretch years into the future, which means many of these facilities will be structurally underutilized even after construction finishes, according to available grid planning data.
That mismatch is especially stark for AI‑optimized campuses that require dense, high‑voltage feeds and advanced cooling. Projects that advertise hundreds of megawatts of "ultimate" capacity often receive only a fraction of that in their initial interconnection agreements, with the rest contingent on future substation upgrades and new transmission lines. In practice, this leaves operators running a small subset of halls while the rest of the building envelope sits dark, a pattern that is already visible in several large campuses cited in regional electricity demand forecasts.
Empty shells, partial buildouts, and the optics of megaprojects
On the ground, many of the "massive" data centers that dominate local headlines are better understood as long‑term shells rather than fully equipped, ready‑to‑run facilities. Developers frequently pour foundations, erect steel, and complete exterior envelopes for multiple buildings while only fitting out one or two with servers and cooling equipment. This phased approach lets them claim large headline capacities while deferring the bulk of capital spending until power and customer contracts are in place, a pattern documented in several California securities filings and local planning records.
From a distance, these sites look finished, with multi‑story structures, perimeter fencing, and new substations. Up close, large portions may be little more than empty halls with stubbed‑out conduits and placeholder mechanical rooms. I have seen project descriptions that tout "up to 300 megawatts" of IT load while current equipment supports only a fraction of that, with the remainder dependent on future phases that have no firm in‑service date. Local land‑use documents and environmental reviews for several Central Valley campuses describe exactly this kind of staged buildout, where full utilization is pushed well into the next decade based on permitted phasing plans.
Grid constraints and multi‑year interconnection delays
The biggest reason these complexes risk sitting underused is the time it takes to secure and deliver reliable power at the scale AI workloads demand. California Independent System Operator studies show that large new loads often wait years in the interconnection queue while utilities evaluate impacts, design upgrades, and secure permits for new lines and substations. For projects sized at hundreds of megawatts, those upgrades can involve multi‑county transmission work that is not scheduled to be completed until late in the decade, according to published transmission plans.
Developers sometimes accept interim, much smaller allocations to get at least part of a campus energized, but that does little to change the overall picture. A facility designed for 200 megawatts that receives only 40 megawatts in its first allocation will run at a fraction of its intended density, leaving large wings unused. Utility filings with the California Public Utilities Commission describe several such staged energizations, where full delivery is contingent on future projects that are still in the design or permitting phase, which reinforces the likelihood that substantial portions of these campuses will remain dormant for extended periods based on current infrastructure planning.
Local politics, water limits, and community resistance
Even when power is theoretically available, local politics and environmental constraints can slow or cap data center operations. Many California jurisdictions now scrutinize water use for evaporative cooling, diesel backup emissions, and noise from mechanical equipment, often attaching conditions that limit operating hours or require additional mitigation before full capacity can be used. City council minutes and county planning commission reports from fast‑growing data center hubs show residents raising concerns about groundwater drawdowns, truck traffic, and the visual impact of multi‑story windowless buildings, which has led to tighter local planning rules.
In some cases, projects have been approved only after developers agreed to cap water consumption, switch to air‑cooled or hybrid systems, or invest in reclaimed water infrastructure that will not be ready for several years. Those conditions effectively delay full utilization even if the buildings themselves are completed on schedule. Environmental review documents under the California Environmental Quality Act describe multi‑phase mitigation timelines for several large campuses, with later phases tied to future infrastructure that is not yet funded, which adds another layer of uncertainty to when these sites can operate at the scale suggested in their initial permit filings.
Stranded asset risk and what it means for California's energy future
The prospect of partially idle data centers raises uncomfortable questions about who ultimately bears the cost of overbuilding. When utilities plan new lines and substations to serve these loads, they often socialize a portion of the expense across ratepayers, on the assumption that the facilities will operate near capacity for decades. If utilization lags because AI demand shifts, efficiency improves, or regulatory limits bite harder than expected, Californians could be left paying for infrastructure that primarily benefits a handful of underused campuses, a concern flagged in recent rate design proceedings.
At the same time, the state is counting on electrification of transport and buildings, along with rapid growth in clean energy, to meet climate goals. Large, intermittently used data centers complicate that picture by adding sharp, location‑specific spikes in demand that are difficult to serve with existing transmission and renewable portfolios. Planning documents from the California Energy Commission and CAISO highlight the challenge of integrating both massive new loads and gigawatts of solar and wind, especially when many of the new facilities are concentrated in regions that already face congestion, which underscores how today's empty or underutilized shells are intertwined with the broader energy transition strategy.
Reply With Quote
Posting Permissions