Artificial Intelligence Center of Excellence

The Texas A&M University System is seeking qualified vendors to collaborate on the design, development, and implementation of an Artificial Intelligence (AI) Center of Excellence through a Request for Proposal (RFP) for an NVIDIA DGX SuperPOD solution. The procurement offers two configuration options: a 3-node system with 760 H200 GPUs or a 4-node system with 1,016 H200 GPUs, accompanied by high-speed networking infrastructure and 8PB of usable storage capacity. Vendors must provide comprehensive project management, site preparation, infrastructure assessment, hardware installation, network integration, technical assistance, user training, and documentation support. The system must be capable of running specific AI models like Grok-1 and Llama 3.1 405B and support both Kubernetes and Slurm scheduling. Proposals are due December 27, 2024, at 2:00 PM Central Time, with implementation expected to begin in June 2025 and last 9-12 months. Evaluation criteria will include demonstrated competence, knowledge, qualifications, implementation timeline, and cost reasonableness. Technical requirements mandate compliance with Texas A&M Information Security Controls Catalog based on NIST standards and integration capabilities with cloud services from AWS, Azure, GCP, and OCI.

The RFP includes a 15% Historically Underutilized Business (HUB) participation goal for subcontracting opportunities, encouraging diversity in vendor selection. Only NVIDIA products will be accepted, with DDN storage alternatives potentially considered. Vendors must maintain specific insurance coverages, including worker's compensation, automobile liability ($1M), commercial general liability ($1M/$2M), and cyber liability ($2M). The project provides 30 NVIDIA Deep Learning Institute training credits and requires TX-RAMP cloud computing certification. The data center infrastructure features 6MW power availability with current rack power limited to 35kW and 14 24-ton CRACs for cooling. While no specific incumbent vendors are mentioned, the infrastructure must have a minimum useful life of 5-7 years and be delivered to Texas A&M University in College Station. Although a precise budget is not explicitly stated, the comprehensive hardware requirements, including 95-127 DGX H200 systems, extensive networking equipment, and advanced storage solutions, suggest a multi-million dollar investment in cutting-edge AI infrastructure.