AI Model Hosting and Inference

BACKGROUND

President Trump’s AI Action Plan tasks CAISI with evaluating frontier AI systems, including open weight AI models, for national security risks. In order to deliver on this tasking, CAISI requires access to open-weight AI model hosting and inference services.

DESCRIPTION OF REQUIRED SERVICES

NIST is seeking information from contractors capable of providing open-weight AI model hosting and inference services.

Requirements include, but are not limited to, the following:

1. Quick-turn hosting: Support newly-released models within the first few business days after they become available.
2. Accurate inference: Configured for accurate inference, including but not limited to reasoning and tool calling performance. The service must have a validation checklist in place to test that models are served correctly.
3. Automatic and rapid inference scaling: Support rapid scaling of inference usage, without requiring that CAISI reserve all needed inference capacity in advance including ability to support high amounts of throughput (input and output tokens) and high concurrency (number of inference requests at a time).
4. Hosting access to many models: Host a large selection of models by default and be able to quickly host open-weight models on-demand for CAISI.
5. Data retention: Not retain inference data.
6. Reliability: High uptime for its inference services.
7. Cost: Hosting and inference for models at a price that is in line with typical industry pricing.
8. Raw model prompting: Provide CAISI with unfiltered API access for raw model prompting (model access without a chat template).
9. Direct Compute Accelerator Access: The ability to access GPUs directly either in a bare-metal or passthrough configuration. This access should allow the execution of arbitrary workloads that require GPU acceleration.
10. Security: NIST SP 800-171 Rev 3 (“Protecting Controlled Unclassified Information in Nonfederal Systems and Organizations”) security controls implemented for all system components involved in storing, processing, or transmitting Controlled Unclassified Information (CUI).

• Cryptographic operations executed by FIPS 140-3 validated cryptographic modules.
• Authentication capabilities implement phishing-resistant MFA.
• All physical infrastructure, data storage, and processing activities restricted exclusively to the United States.
• Sources with the following certifications are preferred:
  1. FedRAMP, SOC 2 Type 2, CMMC Level 2

MARKET RESEARCH OBJECTIVES

The purpose of this Sources Sought Notice is to:

• Identify responsible sources capable of performing the required services
• Determine the availability of small businesses, including socio-economic categories
• Assess industry experience with AI model hosting and inference services

Information received may be used to support acquisition strategy decisions, including potential set-aside determinations.

ANTICIPATED ACQUISITION

NIST anticipates issuing a Request for Quotation (RFQ) in the second quarter of FY 2026, with contract award anticipated no later than the third quarter of FY 2026.

RESPONSE INSTRUCTIONS

Interested parties shall submit a written capability statement addressing the following:

1. Company name, address, Unique Entity Identifier (UEI) number, CAGE code, and point-of-contact information
2. Business size and socio-economic status (if applicable)
3. Description of company capabilities relevant to the services described above
4. Description of prior experience performing similar services
5. Identification of applicable contract vehicles (e.g., GSA FSS, GWACs), including contract numbers
6. Any other information the Government should consider for market research purposes

SUBMISSION REQUIREMENTS

• Submission Method: Email
• Email Address: [email protected]
• Format: Microsoft Word
• Page Limit: 12 pages maximum
• Font: Times New Roman, 11-point
• Paper Size: 8.5 x 11 inches
• Margins: Minimum 1 inch on all sides

Header and footer information that does not contain evaluative content may be included within the margin space.