Request for Proposal: ALCF Equinox AI System

The Department of Energy Office of Science, through UChicago Argonne LLC as operator of Argonne National Laboratory, seeks proposals for the ALCF Equinox AI System—a sixteen-rack, NVIDIA GB300 NVL72-based artificial intelligence computing platform with associated software, networking, installation, and five-year maintenance—with proposals due October 16 and questions due October 2. Addendum No. 2 was issued October 12, 2026.

Place of performance is Building 240 at Argonne's Lemont, Illinois campus, northeast corner with preferred layout of two contiguous rows of compute racks. The NAICS code is 334111, Electronic Computer Manufacturing; PSC is 7B20, IT and Telecom High Performance Compute.

The Base System is mandatory; offerors may propose optional Alternative Commercial Structure (ACS) approaches, but no ACS is eligible for award unless accompanied by a compliant Base System and the Laboratory determines both represent best value. The system must operate independently as a production resource without dependence on future expansion infrastructure and fit within constraints of 10 megawatt Design Maximum IT Electrical Load, 10 megawatt Heat-Rejection Requirement, and under 6,000 square feet of occupied data-center floor space.

Technical requirements include NVIDIA's rail-optimized network topology for east-west networking (alternative topology must be justified); integration with ALCF's north-south Ethernet (192 nonblocking 400 Gigabit Ethernet ports on HPE/Juniper QFX5241-64QD-AO leaf switches with Juniper equipment preferred but not required if interoperability is demonstrated); 480V five-wire power with up to six power whips per compute rack (non-redundant configurations accepted); 64 degrees Fahrenheit supply temperature with 18 degrees Fahrenheit differential cooling; and support for AI training, inference, interactive, service-oriented, agent-based, and batch workloads. Contractor provides all software as separately line-itemized costs (perpetual or subscription acceptable); ALCF maintains a bulk NVIDIA AI Enterprise license for the GPU base system.

Contractor must arrange lease-to-own financing through an acceptable lessor at a fixed interest rate over a 48-month term, with an affiliate lessor identified as having a Master Lease to Ownership Agreement already in effect. Hardware and software line items are invoiced post-acceptance; maintenance and support is invoiced annually in advance. Offerors must provide delivery timeline in calendar days from order receipt on F.O.B. Destination basis to Building 240, with delivery coordination required at least two weeks in advance and all truck drivers U.S. citizens with names and contact information provided five days prior.

Installation and integration includes delivery coordination, staging, rack placement, cabling, power and cooling connections, software installation, system bring-up, and acceptance testing. Contractor and ALCF shall develop an Acceptance Test Plan no later than fifteen calendar days before delivery, addressing functional, performance, and stability testing using benchmarks including GEMM, STREAM, Llama inference, OSU collective communications, HPL, vLLM, SGLang, and AIPerf.

Evaluation employs best-value tradeoff methodology across three proposal parts: Technical Proposal (rated Outstanding, Good, Acceptable, Marginal, or Unacceptable, weighted as more important than Price), Business/Management Proposal (pass/fail determination on contractor responsibility including financial condition, organizational capability, past performance, delivery schedule capability, and proposed exceptions to terms), and Price Proposal (evaluated for total evaluated cost/price, price reasonableness, price realism, and lifecycle-related price implications without adjectival rating). Technical evaluation addresses ten principal areas in descending order of importance: Overall Technical Solution and System Architecture; Delivery and Production Readiness; Facilities Efficiency and Data-Center Integration; High-Speed East-West Network Architecture; North-South Ethernet, Storage, and Laboratory Integration; Software, Resource Management, and Operational Model; Installation, Implementation, and Demarcation Approach; Operations, Support, and Lifecycle Management; Security, Management, Telemetry, and Cloud Independence; and Future Expansion, Technology Evolution, and Operational Flexibility. Proposal risk is independently assessed as Low Risk, Moderate Risk, or High Risk.

Contractors must satisfy on-site supplemental conditions including $2 million general liability insurance, $1 million automobile liability, $2 million professional liability, and statutory workers' compensation naming UChicago Argonne LLC, University of Chicago, and U.S. Government as additional insureds. Subcontractors at Building 240 face additional requirements: $2 million commercial general liability with $10 million umbrella coverage on follow-form basis, comprehensive crime insurance at $10,000 for monies and $100,000 for employee dishonesty, and A.M. Best A-rated or higher policies with thirty-day cancellation notice. Contractors must submit a Job Safety Analysis per 10 CFR 851 before on-site work commences and maintain a qualified Foreperson at all times. Compliance with NEC Article 645 for computer room electrical safety is mandatory.

Deployment-related deliverables including installation documentation, configuration records, and acceptance test reports are classified as data first produced in performance and subject to FAR 52.227-14. Offerors may identify pre-existing intellectual property in proposals, though acceptance is not assured and subject to evaluation and negotiation. The procurement carries a DO rating under DPAS pending approval. Existing master lease agreements with Argonne may be used to fulfill lease requirements if valid and unexpired. Proposal validity is seventy-five days following the closing date.