Transformative AI/ML High-Performance Computing

Overview

The U.S. Army Engineer Research and Development Center (ERDC) Information Technology Laboratory (ITL) High Performance Computing Modernization Program (HPCMP) is seeking commercial solutions capable of delivering secure, scalable, and real-time artificial intelligence (AI)/machine learning (ML) computations in complex and operationally relevant military scenarios. This initiative aims to support the development, prototyping, and demonstration of cutting-edge AI/ML capabilities tailored for modern, high-performance computing environments to be located at the Army Research Laboratory (ARL) Department of Defense (DoD) Supercomputing Resource Center (DSRC).

Background

As the DoD prepares for the complexities of next-generation conflict, it is imperative that computing architecture and software infrastructures evolve to meet the demands of modern warfare. Traditional computing systems, while effective for traditional High-Performance Computing (HPC) workflows, are not fully equipped to address the unique and rapidly growing requirements of the modern DoD. The increasing reliance on data-driven decision-making, real-time analysis, ML based surrogate modeling, generative AI, scene reconstruction, computer vision, sensor fusion, and autonomous systems necessitate specialized computing solutions tailored to AI/ML workloads.

AI/ML technologies are critical enablers for the DoD’s modernization and Critical Technology Area priorities, including predictive analytics, autonomous operations and enhanced situational awareness. However, these capabilities require computing systems that are not only powerful but also scalable, adaptive, and optimized for the unique challenges of military applications. Specialized AI/ML computing platforms are essential to process vast amounts of data, execute real-time calculations, and support distributed operations in contested and dynamic environments.

Project Manager

Information Technology Laboratory (ITL), U.S. Army Engineer Research and Development Center (ERDC)

Requirements

ERDC ITL invites pre-proposals that introduce groundbreaking AI/ML advancements in emerging and future computing paradigms. Offerors shall propose solutions that address one or more of the following objectives:

• Develop an AI/ML system with a modern software environment that can scale-up to an HPCMP Technology Insertion (TI) system size (no greater than approximately 1.2MW) and a test and development system (TDS) with identical architecture
• Support a wide variety of ML scenarios ranging from relatively large-scale training to inference workloads
• There will be a test and development system (TDS) deployed in an unclassified CUI environment and a larger system installed at the SECRET classification level (TI 26 Base System 1)
  • While the larger system (TI 26 Base System 1) will be ultimately deployed at SECRET classification, offerors are encouraged to propose solutions that will result in accelerated timelines to achieve full production environments.
• Solutions/approaches that result in speed from award to system delivery and acceptance
• If successful, options to procure additional systems at classifications up to TS/SCI would be desired.

Proposed solutions should articulate clear technical architecture, maturity level (e.g., TRL), compliance with DoD security requirements and expected outcomes. 

Constraints and Considerations:

• Facilities description
• Personnel should be able to operate in a SECRET environment
• Environment (e.g. computer plus software on computer) should include hardware, software and storage that will be located at ARL
• All contracted systems must meet all physical and information security requirements set by the DoD, the Government worksite, and the Information Systems Security Manager (ISSM), primarily detailed by the DISA Risk Management Framework (RMF) NIST SP 800-53r4. Additional detail may be found in DoD cybersecurity policies (e.g. DoDI 8500.01 and USCYBERCOM TASKORD 17-0019), CNSSI 1253, and applicable DISA Security Technical Implementation Guides (STIGs). The ISSM will provide further guidance, where needed, regarding documented security requirements.
• Applicable STIGs may include, but are not limited to, those for any, network devices, internal databases and internal web servers. For Unix-type operating system distributions for which there is not a distribution-specific STIG available, the general Unix STIG will be used as appropriate. All deviations from STIG requirements, as described by the RMF, must be documented in a written Plan Of Action and Milestones (POAM), in a format provided by the ISSM. (https://www.cyber.mil/stigs)
• All provided nonvolatile storage must support full disk encryption utilizing a cipher approved for national security information.
• Initial solution proposed should be of sufficient size to validate a larger TI class system
• ARL will provide primary system administration support for this system, but will require TIER 2&3 technical support
• Full training regimen will be required for system administration and application support
• Bi-Annual training will be provided for customer focused application development system workflow development.
  • Training must be directly relevant to (and consistent with) the technologies in the contracted systems.
  • Training will be onsite or virtual and support up to 20 developers/customers.
• Hardware maintenance will be required for contracted systems for a 5-year term after acceptance.
  • 24×7 support, with a 4-hour response time
  • Support personnel will require SECRET clearances
  • All personnel requiring administrative system access must comply with the latest versions of DoD Directive 8140.01 and DoD 8570.01-M and must have Information Assurance Technician (IAT) Level II or IAT Level III certification. Per DoD 8570.01-M, DFARS 525.239.7001 and AR 25-2, the Contractor employees supporting IA/IT functions shall be appropriately certified prior to be granted administrative access to systems on the Government’s network.
• Hardware maintenance fees will be prorated on a monthly basis based on system availability to customers.
• Software maintenance will be required for contracted systems for a 5-year term after acceptance. All packages delivered with initial system will be fully supported and maintained via periodic updates/upgrades.
• Acceptance testing for systems will include full functionality tests for system and software packages over a multi-week availability and effective usage period. Details will be negotiated fully after selection of solution.

Applicants must be registered on SAM.gov. The Purpose of Registration on SAM.gov should reflect all awards.

Note: Submissions should NOT include confidential or proprietary details. Submissions may be shared with other ERDC teams if there is an apparent fit with other ERDC projects. 

Estimated Government Funding Profile 

Funding is available for this effort. A minimum and/or maximum amount will not be provided. The government would like to see a variety of scalable technical solutions not bound by a minimum or maximum funding level.

Estimated Period of Performance

While the government prefers quick delivery, the period of performance will be based on the vendor’s solution and may vary. The government may elect to award single and/or multi-year periods of performance.

Desired End State

The anticipated outcome of this effort is the deployment and full acceptance of a state-of-the-art AI/ML supercomputer at Aberdeen Proving Ground, MD. This solution shall encompass not only advanced AI/ML hardware but also a fully integrated and optimized AI/ML software stack. The system must demonstrate both cutting-edge technical innovation and practical deployment in real-world mission environments, ensuring its relevance and utility for the DoD.

A successful solution must deliver a fully operational and secure AI/ML supercomputer on the SECRET network, meeting all required security specifications, regulations, and accreditation standards. The system shall include a comprehensive software stack tailored for AI/ML workloads and must meet or exceed defined functional benchmarks prior to acceptance.

In addition, the system shall impact the DoD will be transformative, as this capability will empower DoD researchers to conduct mission-critical AI/ML research and development at TI-class scales.

Evaluation Criteria

Submissions will be evaluated based on the criteria described in the CSO Solicitation document. Additional criteria specific to this project are listed below:

• Technical requirements will assess how innovative the solution is (as defined in this announcement) and the feasibility of the solution solving the agency’s challenges.
• Importance to agency programs will assess the solution’s potential to enhance the mission effectiveness of the agency.
• Funds availability will assess the availability of funding to procure the solution.

Additional evaluation criteria will be considered:

• Value Proposition – Perceived value of the solution (price/performance)
• Schedule/Delivery – Timeframe to field initial solution up to full scale system
• Support/Training – Ability to support AI/ML workloads in an DoD HPC context
• Security – Ability to support SECRET and above SECRET deployments

All resultant contracts will be firm-fixed price. All items, technologies, and services (including research and development) procured via this CSO are treated as commercial. Applicants from universities and/or non-profit organizations should be aware that commercial clauses will be integrated into the award and should coordinate proposals with associated legal counsel prior to submission.

Notional Project Schedule

Proposed project milestones include:

September 18, 2025	Project Announced, Submissions Open
September 29, 2025	Question Period Ends, FAQ Document Finalized
October 9, 2025	Submissions Close
October 10-24, 2025	ERDC Conducts Reviews
October 2025	ERDCWERX Facilitates Virtual Pitches, if needed
November 2025	Downselected Vendors Notified
Winter 2025	Building 120 – TDS Development/Demonstration
2026	Main System Delivered

*If needed; dates may vary to accommodate project team and participant availability.

Project Security Classification

This project announcement is unclassified.