Continuity Research Facility Alpha

Overview

Continuity Research Facility Alpha (CRF-A) is the Department's principal high-performance computing site. It is managed by the Continuity Infrastructure Administration (CIA-DSC) and serves as a shared research resource for all five divisions. The facility occupies a desert site in southern Nevada chosen for its available transmission capacity, low cost of land, high solar resource and low humidity, which together favor dry heat rejection over evaporative cooling.

The site and a partially completed data hall were developed during the period of the Interagency Task Force on Civilizational Continuity (ITFCC) as a federal research computing project. Custody transferred to the Department in April 2026. CIA-DSC completed the installation of the primary system, ALPHA-1, over the following four months, and the facility reached initial operating capability (IOC) on 18 August 2026. Full operating capability (FOC), defined below, is planned for the third quarter of FY2027.

Facility Alpha is the computational anchor of the Continuity Infrastructure Initiative. It produced the 40,000-run ensemble behind the Civilizational Systems Outlook: 2026–2035 and hosts the controlled evaluation enclaves used by the Frontier System Evaluation Program. It is a research facility: it does not operate or control any external infrastructure, and its systems are not connected to the operational networks of any utility, carrier or agency.

Definition of full operating capability

• Data Hall B power and cooling infrastructure commissioned, bringing installed capacity to the 30 MW design IT load.
• Tape archive populated and verified to its 1.6 EB rated capacity, with a second copy of designated collections replicated to Facility Bravo.
• Continuous 24-hour operations staffing and completed security and privacy accreditation for all research enclaves.
• Operational acceptance of the site energy management system that will dispatch the Phase 2 solar array and battery plant.

Research focus

Allocation of ALPHA-1 time is set each quarter by an allocation committee chaired by the Office of the Chief Scientist, with representation from each division. For FY2027 the allocation is weighted toward three workloads, which together account for about 85 percent of scheduled node-hours.

Integrated systems modeling

Facility Alpha runs production versions of the Integrated Civilizational Systems Model (ICSM), currently version 4.2, for the Integrated Systems Modeling Program. A full Outlook ensemble of 40,000 runs across 1,280 scenario configurations occupies roughly one-fifth of ALPHA-1 for about eleven days, including post-processing. The model description is published as DSC-TR-2026-003. Every production run is archived with its configuration, input data hashes and code version, so that any published result can be regenerated.

Neural emulation benchmarks

The Neural Continuity Research Program uses Facility Alpha for its emulation benchmark suite, which measures how well neuron-level models of small nervous systems reproduce recorded behavior and physiology. Current benchmarks operate at Emulation Fidelity Level 3 (EFL-3) for C. elegans and for partial Drosophila circuits. The facility also hosts reconstruction pipelines for synaptic-resolution imaging data, whose acquisition requirements are set out in DSC-TR-2026-009. Projections of the compute and energy required for mammalian- and human-scale emulation, which inform the facility’s long-range planning, appear in DSC-TR-2026-008. Those projections are requirements estimates, not commitments, and nothing at Facility Alpha emulates a human or mammalian brain.

Frontier system evaluation

The Office of Advanced Artificial Intelligence conducts structured evaluations of frontier computational systems in physically and logically isolated enclaves at Facility Alpha, following the methodology in DSC-TR-2026-007. Enclaves have no outbound network path; results leave the enclave only through a reviewed export process.

Technical capabilities

ALPHA-1 is a liquid-cooled system of 2,304 compute nodes, each with four accelerators and two host processors, installed in 72 cabinets in Data Hall A. Nodes are connected by a three-level fat-tree interconnect with 400 Gb/s per-accelerator injection bandwidth. At full load the system draws approximately 13.5 MW.

Cooling and water

Compute nodes are cooled by direct-to-chip cold plates on a warm-water loop with a design supply temperature of 32 °C. Heat is rejected to the atmosphere through closed-loop dry coolers. On the hottest days of the year, when ambient temperatures exceed the dry coolers’ approach limit, an adiabatic pre-cooling stage engages; it is expected to operate for fewer than 300 hours per year. The facility uses no cooling towers, and annual makeup water is a small fraction of what an evaporatively cooled facility of the same load would consume. The annual power usage effectiveness (PUE) target is 1.15.

Power

Facility Alpha has a 42 MW contracted grid supply delivered over two independent high-voltage circuits. Storage, networking, cooling controls and building systems are carried by uninterruptible power supplies and standby generation sized at approximately 6 MW. Compute partitions are not on standby generation; during an extended grid outage they are checkpointed and shed in a defined order. Phase 2 adds an on-site 60 MW solar array with a 240 MWh battery plant, now under construction, which will supply a substantial share of daytime load and reduce peak grid draw.

Safety and environmental compliance

Facility Alpha was sited and built under the applicable federal environmental review process, including surveys of desert habitat and cultural resources before the solar array footprint was cleared. Construction and operations follow applicable federal occupational safety, fire protection and hazardous materials requirements. The battery plant is being built with separated enclosures, thermal runaway detection and setbacks agreed with the regional fire authority.

Specifications