AI Model Hosting and Inference

The Government intends to procure AI model hosting and inference services to support the Center for AI Standards and Innovation in order to support CAISI's tastings under President Trump’s AI Action Plan to evaluate frontier PRC models for security vulnerabilities, censorship, and risks to national security. In order to deliver on these taskings at an acceptable level, CAISI requires access to AI model hosting and inference services. These services must be quick-turn (newly-released models are available within a few business days), have accurate inferences, support automatic and instantaneous scale-up, support many models, require a low amount of CAISI staff time for support, allow customization including raw model prompting and direct compute accelerator access, and meet NIST security requirements. Current internal CAISI AI model hosting and inference infrastructure (on AWS) is quick-turn, accurate, and secure, but does not fulfill the