Seismic Risk and Data Center Site Selection: Hazard Maps, Design Categories and Liquefaction

An earthquake can harm a data center in four distinct ways, and each one is screened with different data. The first is ground shaking, which loads the structure, the racks, the raised floor and every piece of electrical and mechanical equipment. The second is ground failure: liquefaction, lateral spreading and landslides that move or settle the foundation. The third is surface fault rupture, where a fault offsets the ground directly under a building. The fourth is the loss of lifelines: utility power, water, fuel delivery and road access that the facility depends on for days after an event.

Shaking is regional and is set mostly by the site’s location relative to earthquake sources, then amplified or damped by the soils under it. Ground failure and fault rupture are local and can differ from one end of a large parcel to the other. That is why a seismic screen starts with national maps and ends with borings. The broader multi-hazard view (wind, tornado, wildfire, winter storms) is covered in natural hazard risk for data centers; this guide goes deeper on earthquakes alone.

The economic exposure is large even outside California. FEMA and the USGS estimate the annualized earthquake loss to the national building stock at $14.7 billion per year, against roughly $107.8 trillion of exposed building value.¹⁰ For a campus that may carry several billion dollars of equipment, that exposure is a design and insurance question long before it is a safety one.

The starting point is the USGS National Seismic Hazard Model (NSHM). The 2023 update was produced by a team of more than 50 scientists and engineers and was the first version to include all 50 states.¹ Its headline map shows the likelihood of damaging earthquake shaking over the next 100 years, built from fault sources, historical seismicity and ground motion models.¹

The highest chances fall in southern Alaska, California and Hawaii, which largely exceed 95% over a century, and parts of the Mississippi Valley around the New Madrid zone reach up to 90%.¹¹ Much of the Southeast, the Appalachians and the Northeast sit in a moderate band. Large parts of the Great Plains, the Upper Midwest and Texas show low probabilities, which is one reason those regions screen well on this criterion.

Two cautions apply when using the maps for site selection. First, they describe a regional probability on a reference soil condition, not what a specific parcel will experience; local soils can raise or lower the shaking. Second, hazard maps change. The ASCE 7-22 design values were revised from the prior edition, with most of the changes in the western U.S. and the New Madrid region, so values pulled for an older study should be refreshed.²

Most U.S. jurisdictions adopt the International Building Code, and the 2024 IBC references ASCE 7-22 for structural loads.² Under ASCE 7-22, the mapped seismic parameters are taken directly from the USGS Seismic Design Geodatabase, which the USGS serves through a public web service. A user enters latitude, longitude, the building’s risk category (I to IV) and the site class, and the service returns the design spectral values and the seismic design category (SDC).³

The SDC runs from A (lowest) to F and drives how much seismic detailing, bracing and inspection the building and its equipment need. It depends on three inputs a buyer should track separately:

ASCE 7-22 also replaced the old two-point spectrum with a multi-period spectrum and added intermediate site classes, so the edition matters.⁵ One counterintuitive result: on some soft-soil sites in the central and eastern U.S., the new spectra produce lower accelerations. Engineers have cited Savannah, Georgia, where a Risk Category II building on Site Class E dropped from SDC D under ASCE 7-16 to SDC C under ASCE 7-22.⁵ Some states amend these provisions, so the adopted code and local amendments have to be confirmed with the building department. Code requirements for the building itself are covered in building and fire codes for data centers.

Two parcels a mile apart can land in different seismic design categories because of what lies beneath them. Site class is set from the stiffness of the upper soil profile, measured with borings, cone soundings and shear wave velocity testing in a geotechnical investigation. When soil data are missing, the USGS service offers a “Default” site class option for cases where soil properties are not known.³ Results run on the default are preliminary until field data set the actual class.

Liquefaction is the ground failure most likely to matter on otherwise flat, attractive land. It happens when strong shaking builds up pore-water pressure in saturated soil until the grains lose contact and the soil briefly behaves like a fluid.⁴ The hazard is confined to loose, cohesionless sand and silt below the water table; recently deposited soils and uncompacted, human-made fill are the most vulnerable.⁴ Surface effects include sand boils, ground cracking, settlement, lateral spreading and loss of foundation bearing.⁴

• Where to look first: young river and floodplain deposits, filled former wetlands or water bodies, and sites with shallow groundwater. These often overlap with floodplain screening.
• What the geotechnical report should state: whether liquefiable layers are present, how thick and how deep, and the estimated settlement and lateral displacement.
• Typical responses: ground improvement such as densification or stone columns, deep foundations that bypass the liquefiable layer, or moving buildings to better ground on the same parcel.

Liquefaction is usually a cost and layout question rather than a reason to walk away, but it can consume buildable area and budget. The general ground-condition workflow is described in slope, soils and geotechnical review.

Surface fault rupture is rare at any given site but cannot be engineered away for a building that straddles the fault, so it is screened by avoidance. The national reference is the USGS Quaternary Fault and Fold Database, which catalogs faults and folds with geologic evidence of surface deformation in large earthquakes during roughly the past 1.6 million years and supplies fault sources for the national hazard maps.¹²

California regulates fault rupture directly. Under the Alquist-Priolo Earthquake Fault Zoning Act, the California Geological Survey maps Earthquake Fault Zones, and structures for human occupancy within a zone generally require a site-specific fault study and must be set back from the active trace, typically by 50 feet.¹³ Outside California, few jurisdictions have equivalent rules, which means a mapped Quaternary fault crossing a parcel may raise no permitting flag at all. A buyer should still have the geologist check it, because a fault under a data hall is a risk the building code will not catch.

Faults also matter indirectly. A fault crossing a planned transmission route, gas lateral or water main can sever that lifeline even when the campus itself is clear.

Building codes aim mainly at life safety and, for essential facilities, continued function. A data center owner usually cares about something narrower: whether the facility keeps running after the event. That outcome depends less on the frame than on the equipment inside it.

ASCE 7 Chapter 13 covers nonstructural components. Components assigned a component importance factor of 1.5 include those needed for life safety, those containing hazardous materials and those that must function after the design earthquake; in Seismic Design Categories C through F, active mechanical and electrical equipment in that group needs special seismic certification, commonly by shake-table testing.⁶ Enforcement varies by jurisdiction, so owners often write certification into their own equipment specifications.⁶

Some owners go further. NTT’s SV1 data center in Santa Clara is base-isolated on triple friction pendulum bearings paired with 25 long-stroke fluid viscous dampers, which limit the building’s movement in an earthquake to 32 inches.⁷ In Hillsboro, Oregon, a data center project was built to essential facility standards for seismic, snow and wind, with a structural importance factor of 1.5.¹⁴

Lenders and insurers express earthquake risk as an expected loss. ASTM E2026 is the guide for seismic risk assessment of buildings, and ASTM E2557 is the practice for probable maximum loss evaluations in commercial real estate transactions; both were reissued in 2024.⁹ E2026 defines assessment levels from 0 (screening) through 3 (detailed engineering), and recommends Scenario Expected Loss and Scenario Upper Loss in place of the older “probable maximum loss” label.⁸ For raw land, there is no building yet, so the useful version of this exercise is a desktop seismic screen plus the design values that will feed the cost estimate.

1. 01Pull ASCE 7-22 values for each candidate site with a default site class and the likely risk category, and compare SDCs side by side.³
2. 02Check the Quaternary fault database and any state fault zone maps for traces on or near the parcel and its utility routes.¹²¹³
3. 03Ask the geotechnical engineer to state the site class, liquefaction potential and recommended foundation type in the preliminary report.
4. 04Confirm the adopted code edition and local amendments with the building department.⁵
5. 05If financing or insurance will require it, scope an ASTM E2026 assessment early.⁸

Seismic findings belong in the same comparison as power, fiber and land cost, as laid out in the site selection criteria checklist and the due diligence checklist. Our methodology treats seismic hazard as a cost and design input except where faults or severe ground failure limit where buildings can go. If you want a parcel screened, you can get a site reviewed.