Empire_AI_Gamma_RFI.pdf

EMPIRE AI CONSORTIUM, INC.

Empire AI Research Computing Center | University at Buffalo North Campus

REQUEST FOR INFORMATION

GAMMA PHASE

Computing Assets and White Space Infrastructure

Advanced SI/AI and HPC Computing Infrastructure • Up to 5 MW IT Load

Empire AI Research Computing Center — southeast aerial rendering (Gensler, Design Development, 08/13/2026)

RFI No. EAI-2026-102902 Issue Date: October 29, 2026

Responses Due: December 4, 2026, 4:00 PM ET

THIS IS NOT A SOLICITATION

Empire AI Consortium, Inc. | Request for Information — Gamma Phase

This is not a solicitation. Page 2 of 40

TABLE OF CONTENTS

I. PURPOSE

1. Introduction

2. Objectives

3. Nature of this RFI

II. GENERAL INFORMATION

1. The Empire AI Consortium

2. Deployment Roadmap

3. Existing Systems and Relationship to the New Facility

4. New York State Policy Context

III. BUILDING INFRASTRUCTURE DESCRIPTION

1. Site and Building

2. Utility Service and Electrical Distribution

3. Standby and Critical Power

4. Mechanical Systems, Heat Rejection, and Technical Cooling

5. Controls, Monitoring, and Life Safety

6. Network and Telecommunications

7. Point-of-Connection Summary

IV. USER REQUIREMENTS

1. User Community

2. Workload Profile

3. Service and Operational Requirements

4. Storage and Data Requirements

V. SCOPE OF INFORMATION REQUESTED — COMPUTING ASSETS

1. Functional and Performance Requirements

2. Specific Computing Information Requested

3. Future Deployment (Delta)

VI. SCOPE OF INFORMATION REQUESTED — WHITE SPACE INFRASTRUCTURE

1. General

2. Contracting Structure, General Conditions, and Delivery Team

3. Technical Cooling System and Coolant Distribution Units

4. Critical Power Infrastructure

5. Physical Infrastructure

6. Controls, Monitoring, and Integration

7. Design, Installation, Commissioning, and Turnover

VII. BUDGET AND PRICING INFORMATION

1. Program Budget and Tranches

2. Requested Approach

This is not a solicitation. Page 3 of 40

3. Spend Rate and Timing

4. Pricing Format

VIII. QUESTIONS FOR RESPONDENTS

A. Architecture, Roadmap, and Spend Rate B. Facility Fit and Existing Assets C. Procurement and Delivery Model D. Information for a Future RFP

IX. SUBMITTAL PROCESS

1. Schedule

2. Communications and Designated Contact

3. Submission of Responses

4. Costs and Expenses

5. Review of Responses

6. No Binding Obligation

7. Confidentiality

X. SUBMITTAL CONTENT AND FORMAT

XI. CONCLUSION

ATTACHMENT A — INDICATIVE PRICING SUMMARY AND PRICING GUIDE

A-1. Indicative Pricing Summary A-2. Pricing Guide (Checklist)

ATTACHMENT B — COMPUTING EQUIPMENT TECHNICAL DATA SHEET (GAMMA)

ATTACHMENT C — FACILITY INTERFACE DATA SHEET

ATTACHMENT D — PRELIMINARY RESPONSIBILITY MATRIX

ATTACHMENT E — RESPONDENT INFORMATION FORM

ATTACHMENT F — DEFINITIONS AND ABBREVIATIONS

This is not a solicitation. Page 4 of 40

I. PURPOSE

1. Introduction Empire AI Consortium, Inc. ("Empire AI" or the "Consortium") is a public-private artificial intelligence research consortium created to secure New York State's place at the forefront of the artificial intelligence revolution. Launched in April 2024 by Governor Kathy Hochul, the Consortium's flagship infrastructure, including a supercomputing center at the University at Buffalo ("UB"), will promote responsible research and development, create economic opportunity, and advance innovation for the public good.

Empire AI recognizes the federal government's adoption of the terms "Super Intelligence" and "SI" under Executive Order 14434, Inaugurating the Era of Super Intelligence (September 29, 2026). The systems Empire AI deploys are intended to support research at the frontier of both artificial intelligence and Super Intelligence (SI), and Respondents may use the terms "AI" and "SI" interchangeably in their responses.

Empire AI's first system, Alpha, became operational in October 2024 and has since been expanded to support hundreds of users. Its second system, Beta, came online in August 2026, as announced by Governor Hochul on August 5, 2026, and is the most powerful academic research supercomputer operated by any state in the nation. Both systems are hosted in the University at Buffalo's Center for Computational Research (CCR) data center.

Building on this momentum, Empire AI is constructing the Empire AI Research Computing Center (the "Facility"), a purpose-built, liquid-cooled AI data center on the UB North Campus in Amherst, New York, expected to be online at the end of 2027. The Facility is master-planned for 15 MW of computing (IT) load, to be delivered in deployments of up to 5 MW each. This RFI addresses the first two deployments, Gamma and Delta; capacity beyond Delta is part of Empire AI's long-term plan but is outside the scope of this RFI.

Through this Request for Information ("RFI"), Empire AI seeks information from qualified technology providers, system integrators, and their engineering and construction partners ("Respondents") regarding (a) the computing assets for the Gamma deployment, comprising a high-performance, accelerator-based computing infrastructure supporting large-scale AI training, inference, and scientific computing, with up to 5 MW of IT load; and (b) the white space infrastructure required to support those computing assets within the Facility, including coolant distribution units (CDUs), the technical cooling system, flywheel uninterruptible power supply (UPS) systems, and 480 V distribution panelboards. Empire AI has established a combined budget for the Gamma and Delta tranches (Section VII) and invites Respondents to recommend how that budget should be divided between the two tranches and to describe the solution that would deliver the greatest research capability within their recommended Gamma allocation, together with a technology roadmap for subsequent deployments.

2. Objectives The objectives of this RFI are to:

• Obtain high-level architectural concepts and suggested system configurations for the Gamma computing system, including specific information on accelerators, CPUs, memory, scale-up and scale-out networking, storage, software, and rack-level power and cooling requirements.

This is not a solicitation. Page 5 of 40

• Obtain information on the white space infrastructure required to deploy the suggested computing system in the Facility, and the boundary between Respondent-provided and building-provided infrastructure.

• Obtain indicative acquisition, installation, and operating costs in a common format (Attachment A) so that responses can be compared on a consistent basis.

• Obtain Respondents' recommendations on spend rate and timing — how the Program Budget should be divided between the Gamma and Delta tranches — with detailed supporting rationale.

• Understand delivery timelines, supply chain risks, and the principal technical and commercial uncertainties affecting a Gamma deployment in 2027–2028.

• Identify the information, reference workloads, requirements, and evaluation criteria that Empire AI should include in a subsequent Request for Proposals ("RFP").

3. Nature of this RFI This RFI is a formal, non-competitive informational inquiry. This is not a solicitation. No contract will be awarded on the basis of responses to this RFI, and submittal of a response is not a condition for consideration in any future procurement. Upon review of the information received, Empire AI will determine whether and how to proceed. No obligation or commitment of any kind is made by Empire AI to any Respondent.

Procurement Lobbying Law. This RFI is subject to State Finance Law §§139-j and 139-k (the "Procurement Lobbying Law"). Pursuant to these provisions, Respondents are restricted from making contacts with Empire AI regarding this RFI and any subsequent procurement, from the issuance of this RFI through final award and approval of any resulting procurement contract (the "Restricted Period"), to anyone other than the Designated Contact identified in Section IX.2, unless the contact is among the statutory exceptions set forth in State Finance Law §139-j(3)(a). Empire AI is required to record certain contacts during the Restricted Period and to make a determination of responsibility pursuant to these statutes. Certain findings of non-responsibility may result in rejection for contract award, and two such findings within a four-year period will result in debarment from obtaining governmental procurement contracts. Respondents are not required to submit Procurement Lobbying Law forms with RFI responses; such forms will be required with any subsequent RFP.

This is not a solicitation. Page 6 of 40

II. GENERAL INFORMATION

1. The Empire AI Consortium Empire AI is a consortium of leading New York public and private research institutions together with the State of New York. The Consortium provides shared, high-performance AI computing resources to faculty, researchers, and students at member institutions for academic research in the public interest. The Consortium's partner institutions are Columbia University, Cornell University, the City University of New York (CUNY), the State University of New York (SUNY), Rensselaer Polytechnic Institute, New York University, the University of Rochester, Rochester Institute of Technology, the Flatiron Institute, and the Icahn School of Medicine at Mount Sinai. The Facility will be located on the UB North Campus; UB serves as the site owner and host institution.

2. Deployment Roadmap Empire AI's computing program is organized as a sequence of deployments. Alpha and Beta are installed and operating. Gamma is the subject of this RFI. Delta is the planned second tranche of the Program Budget, for which Empire AI requests roadmap and indicative information.

Phase Status / Timing Technology Basis IT Load Power Architecture Alpha Installed (operational

Oct. 2024) NVIDIA H100 HGX nodes; RTX Pro nodes; several PB storage

— CCR data center, CBLS building, UB Downtown Campus

Beta Installed (online Aug.

2026)

NVIDIA DGX GB200 SuperPOD (four GB200 NVL72 racks with SuperPOD networking) and Grace-Grace nodes; 20 PB

VAST; 10 PB DDN

— CCR data center, CBLS building, UB Downtown Campus

Gamma Planned — new Facility online end of 2027; target Gamma power-on approx. July

High-density, accelerator-based AI and HPC computing systems optimized for large-scale research workloads

Up to 5

MW

AC distribution; 480 V AC to the racks preferred over 415 V solutions

Delta Planned — approx. 2– 3 years after Gamma (second tranche of Program Budget)

Next-generation accelerator-based computing systems, informed by emerging technologies and future research requirements

Up to 5

MW

800 V DC distribution (anticipated)

Total (this

RFI)

Gamma and Delta Up to 10

MW

Facility master-planned for 15 MW

The roadmap above reflects Empire AI's current planning intent and is not a commitment to any vendor, product, or schedule. Respondents are encouraged to suggest alternatives where they believe a different phasing, technology basis, or power architecture would better serve Empire AI.

3. Existing Systems and Relationship to the New Facility Alpha and Beta are currently hosted in the University at Buffalo's Center for Computational Research (CCR) data center, located on the ground floor of the New York State Center of Excellence in Bioinformatics and Life Sciences (CBLS) building, 701 Ellicott Street, Buffalo, New York 14203, on UB's Downtown Campus. Empire AI has not determined whether Beta, in whole or in part, will be relocated to the new Facility. If Beta is not relocated, research data and workflows may be split

This is not a solicitation. Page 7 of 40 between locations, and capabilities may need to be replicated or connected by a high-bandwidth campus network link (which can address bandwidth but not latency). Respondents are invited to comment on the implications of relocating or not relocating Beta for the Gamma architecture, storage strategy, and network design (see Section VIII).

4. New York State Policy Context As an entity serving State-sponsored research, Empire AI is mindful of the State's clean energy and climate goals under the Climate Leadership and Community Protection Act and Executive Order 22.

The Facility has been designed for high energy efficiency, including liquid cooling, dry-cooler-based heat rejection, and provisions for recovery of waste heat to support campus heating. Respondents should identify features of their suggested solutions that improve energy efficiency, increase useful work per unit of energy, or support heat recovery.

This is not a solicitation. Page 8 of 40

III. BUILDING INFRASTRUCTURE DESCRIPTION

This Section summarizes the Facility's base building infrastructure as reflected in the Design Development ("DD") documents prepared by Gensler (Architect/Engineer of Record), issued for Design Development on August 13, 2026, and the associated DD cost estimates dated September 2026. The design is still in progress; quantities, ratings, and configurations described below are preliminary and subject to change. Final building documents and interface criteria will be furnished with any RFP. Respondents should identify any building infrastructure characteristic described here that would be incompatible with, or would limit, their suggested solution.

1. Site and Building

Figure 1 — Exterior rendering, building entrance. Source: Gensler, Empire AI Research Computing Center, Design

Development issue, 08/13/2026.

Item Description Location UB North Campus, Town of Amherst, Erie County, New York; latitude 42.9922° N, longitude 78.7972° W (per Whitestone survey) Owner / Site Owner Empire AI Consortium, Inc. / University at Buffalo Project Team Gensler (Architect/Engineer of Record) with consultants IMEG, AZAR Design Co., Encorus Group, P.MAX, and Trinity Consultants; Gilbane Building Company (Construction Manager at Risk); Wendel (Construction Manager as Agent / Owner's Representative)

This is not a solicitation. Page 9 of 40

Item Description Building Size and Organization

Approximately 46,900 gross square feet (46,924 GSF per DD), organized as a compute block and a power plant. Exterior equipment yard: 34.5 kV gas-insulated switchgear and automatic selector switches, pad-mounted 34.5 kV–480 V transformers, and standby generators. Ground level: main electrical room (480 V main switchboards and flywheel UPS units), power and cooling distribution room (CDU positions), electrical distribution line-ups, chiller plant with water-cooled chiller maintenance bays, process cooling water and medium-temperature chilled water distribution, loading dock, freight elevator, main entry, and water and fiber entry.

Second level: data hall, double-height fan-wall gallery (fan-wall units located at ground level, Mechanical Gallery 107A), machine viewing window, staging lab, mechanical and electrical rooms, and operations and administration. Roof:

mechanical dunnage platform with adiabatic dry coolers, served by freight and stair access. Exterior equipment yard with equipment screening.

Data Hall (White Space) Second-level data hall of approximately 7,000 SF of raised access floor (steel-supported, 2'-0" on center, cross-braced) with a suspended strut ceiling system (Tate or equal) for overhead support of distribution; structural floor bracing below the data hall; data hall design condition 75°F DB / 35% RH

Support Spaces Network demarcation room(s), UPS/electrical rooms, staging/build lab, machine room, IDF, offices, conference and support spaces

Logistics Heavy-duty freight elevator and loading provisions sized for power-truck loading;

delivery and rigging pathways to be confirmed with the RFP

Schedule Substantial completion of the base building is targeted for Q4 2027; target Gamma power-on approximately July 2028

Heat Recovery Provisions for future recovery of computing waste heat to support campus heating

Figure 2 — Building axonometric, southeast (operations, entry and lobby, loading dock, chiller plant, water and fiber entry).

Source: Gensler, Empire AI Research Computing Center, Design Development issue, 08/13/2026.

This is not a solicitation. Page 10 of 40

Figure 3 — Building axonometric, northwest (compute block, power plant, exterior equipment yard, rooftop equipment platform). Source: Gensler, Empire AI Research Computing Center, Design Development issue, 08/13/2026.

This is not a solicitation. Page 11 of 40

Figure 4 — Building exploded axonometric: roof (dry coolers), second level (data hall, fan-wall gallery, staging lab), and ground level (34.5 kV switchgear, main electrical room with flywheel UPS units, chiller plant, process cooling and chilled water distribution). Source: Gensler, Empire AI Research Computing Center, Design Development issue, 08/13/2026.

This is not a solicitation. Page 12 of 40

Lobby cutaway

Overlook lounge

Entry lobby

Figure 5 — Interior renderings. Source: Gensler, Empire AI Research Computing Center, Design Development issue, 08/13/2026.

2. Utility Service and Electrical Distribution The Facility will be served at 34.5 kV medium voltage via an underground primary duct bank, with empty conduit provisions for future capacity. Day 1 (Gamma) medium-voltage equipment includes

34.5 kV gas-insulated switchgear, medium-voltage automatic selector switches, and four (4) 3,000 kVA, 34.5 kV–480 V pad-mounted transformers (TR-A1, TR-B1, TR-G1, TR-H1) serving 480 V, 4,000 A main switchboards (MSB-A1, MSB-B1, MSB-G1, MSB-H1). Four (4) additional 3,000 kVA transformers (TR-C1 through TR-F1) are reserved for future deployments and are not part of the Day 1 building scope.

IT power for Gamma is planned on A and B distribution paths. The DD documents show 750 kVA, 480–415/240 V low-voltage IT distribution transformers (TR-A1-1 through A1-4 and TR-B1-1 through B1-4) feeding overhead 415/240 V, three-phase four-wire compute busway with tap-off boxes (BW- A1 through A4 and BW-B1 through B4). Empire AI will give preference to computing solutions that accept 480 V AC at the rack, which would allow the IT transformers to be eliminated or reduced and the busway to be specified at 480 V; Respondents should state whether their suggested platform accepts 480 V AC input. Note that the IT transformers TR-A1 and TR-B1 and main switchboards MSB- A1 and MSB-B1 are 480 V, three-phase, three-wire (no neutral); Respondents should state whether their platform requires a neutral or 480Y/277 V service. The redundancy topology for IT critical power is to be confirmed in the RFP; Respondents should state the redundancy their platform requires and the redundancy they recommend. The scope assignment of the IT transformers and busway between the base building and the white space is under review; see Attachment D.

This is not a solicitation. Page 13 of 40

3. Standby and Critical Power The DD documents include two (2) 750 kW natural gas standby generators with paralleling gear and a generator docking station, and static UPS systems (one 500 kVA and one 1,000 kVA) serving building critical loads. The DD one-line shows static UPS-B feeding distribution panel DP-B (Power and Cooling Distribution Room 105), which serves all fifteen CDU circuits and process cooling pumps P-PC-1 through P-PC-4 (approximately 830 kVA connected). These are building-provided systems.

Ride-through for the IT load is planned to be provided by battery-free rotary flywheel UPS systems (1,500 kVA units; four (4) units for Gamma, two per A/B switchboard, with up to twelve (12) for the full 15 MW build-out), located in the ground-level main electrical room (Figure 4). The flywheel UPS systems are not included in the base building construction scope and are requested as part of the white space infrastructure under this RFI. Respondents should not assume generator backup of the IT load; flywheel UPS ride-through is measured in seconds. Respondents should describe how their computing platform behaves during utility disturbances and outages and what ride-through, power-smoothing, or energy-storage features (rack-level or facility-level) they recommend.

4. Mechanical Systems, Heat Rejection, and Technical Cooling The Facility's base building mechanical systems include three (3) 675-ton water-cooled, magnetic-bearing, variable-speed centrifugal chillers; ten (10) rooftop adiabatic dry coolers on the mechanical dunnage platform (DC-1 through DC-10, installed in phases; Day 1 includes DC-1 through DC-3 on the process cooling loop and DC-8 and DC-9 serving the chiller condensers, with DC-4 through DC-7 and DC-10 future); process cooling water thermal energy storage; water-to-water heat pumps for heat recovery; electric hot water boilers, with natural gas boilers for emergency backup heat; a dedicated outdoor air system providing humidity control to the data hall; seven (7) modular fan-wall air handling units serving the data hall (four Day 1, three future), located in the ground-level mechanical gallery, plus six (6) fan-wall units serving the ground-level main electrical room; and computer room air handlers serving the demarcation and UPS/electrical rooms.

Liquid cooling of the computing equipment will be served by a Technical Cooling System ("TCS").

On the facility side, the CDUs are served by the Process Cooling System (PCS/PCR), a dedicated warm-water glycol loop rejected through rooftop adiabatic dry coolers without chillers. It is separate from the medium-temperature chilled water (MCWS/MCWR) the chillers produce for air-side systems.

In the DD documents, "TCS" denotes the technology-side (secondary) loop downstream of the CDUs (Figure 4). The building provides the facility-side (primary) water loop and heat rejection, terminating in capped branch connections (tap, isolation valve, and cap) in the ground-level Power and Cooling Distribution Room (Room 105), directly below the second-level data hall, for future CDU connections. The PCS and medium-temperature chilled water mains in that room include additional branch taps for future CDUs, rear-door coolers, or other critical cooling components. The DD documents show provisions for up to fifteen (15) CDU electrical connections (CDU-1-1 through CDU- 1-15, 480 V, 38 kVA each, fed from building panel DP-B; see Section III.3). CDUs, secondary (technology-side) piping, rack manifolds, and associated specialties are not included in the base building scope and are requested as part of the white space infrastructure under this RFI.

Design facility water supply and return temperatures, available flow, differential pressure at the branch connections, and fluid type are to be confirmed by the Engineer of Record and furnished with any RFP (the Design Development basis is a process cooling supply of approximately 92°F (33.5°C), a return of approximately 107°F (41.7°C), and 30% glycol). Respondents should state the coolant temperatures, flow rates, pressure drops, water quality, and liquid-to-air heat ratio required by their

This is not a solicitation. Page 14 of 40 suggested platform (Attachment C), and identify the highest facility water supply temperature at which full performance can be maintained, since higher temperatures improve dry-cooler hours and heat recovery.

5. Controls, Monitoring, and Life Safety The building will include a Building Management System ("BMS") controlling base building mechanical and electrical systems, electrical power monitoring, and a fire alarm system with very early smoke detection (VESDA) in the data hall. Respondents will be expected to integrate white space and computing-platform telemetry (CDUs, flywheel UPS, panelboards, leak detection, rack power and thermal data) with the BMS via open protocols, and to describe their data center infrastructure management ("DCIM") and platform management , including provisioning, monitoring, fault management, and lifecycle management, approach.

6. Network and Telecommunications The building will provide exterior telecommunications duct banks, backbone pathways, and conduit or cable tray from the network demarcation room to the data hall. All network infrastructure within the data hall — including cable tray, fiber and copper cabling, and all network equipment — is considered part of the technology deployment and is not included in the base building scope. The wide-area network (WAN) edge is expected to include connectivity to NYSERNet. For reference, Beta currently has three 100 Gbps external connections: one to UB Norton, one to UB Bell, and one to NYSERNet.

Respondents should recommend WAN edge capacity and architecture for Gamma.

7. Point-of-Connection Summary In general, the building provides utility service, medium-voltage distribution, 480 V main switchboards, standby generation for building loads, heat rejection, the facility-side TCS loop to capped branch connections, data hall air handling, the raised floor and ceiling support grid, fire protection and alarm, and the telecommunications pathway to the data hall. Everything downstream of those points of connection is white space or computing scope. A preliminary responsibility matrix is provided in Attachment D; Respondents are asked to comment on it and identify any items they believe should be assigned differently.

This is not a solicitation. Page 15 of 40

IV. USER REQUIREMENTS

1. User Community Gamma will serve faculty, research staff, postdoctoral researchers, and graduate students across Empire AI member institutions. Users range from experienced AI and HPC practitioners running multi-node jobs at scale to domain scientists who are new to accelerated computing. Alpha and Beta currently support over 1,600 active users (1,631 as of October 2026) across 314 approved projects, spanning the consortium’s ten partner institutions. Resource access follows an institution-level allocation model: each member institution receives a fixed Service Unit (SU) budget per allocation period, which it distributes across its own projects, with fairshare scheduling governing day-to-day priority within that budget. Alpha and Beta currently operate at over 80 percent utilization. Of these, only 116 (36 percent) have activated usage to date, indicating substantial additional near-term demand as the onboarding pipeline matures. Empire AI expects the user base and the share of large-scale jobs to grow substantially over the ten-year horizon of this program. Gamma's role relative to Beta and Delta, including support for capability-class ("hero") jobs and for model hosting and inference workloads, is discussed in Section V.1.

2. Workload Profile Empire AI anticipates a mixed research workload spanning large-scale model training, fine-tuning and post-training, inference, scientific AI and HPC, and data-intensive analytics, including capability-class ("hero") jobs that use a large fraction of the system as a single allocation, and model hosting and inference workloads (Section V.1). A detailed workload specification, reference workloads, and representative usage data from Alpha and Beta will be provided with any RFP. For this RFI, Respondents should identify the workload assumptions that drive their recommendations and how changes in demand would alter them.

3. Service and Operational Requirements The Gamma system must support the following at a minimum. Respondents should describe how their suggested solution meets each requirement and note any limitations.

• Multi-tenancy and allocation: secure sharing among many institutions and projects, with fair-share scheduling, quotas, accounting, and reporting by institution and project.

• Scheduling and orchestration: batch scheduling (e.g., Slurm) and container orchestration (e.g., Kubernetes), with support for both traditional HPC jobs and cloud-native AI workflows.

• Software environment: current AI frameworks, libraries, compilers, and containers; a supported path to future software releases; and a portable environment that allows users to move work between Beta and Gamma.

• Data governance and security: support for sensitive and controlled research data, including identity federation across member institutions, data segregation, encryption, audit logging, and compliance with applicable research security requirements for both non-regulated data and regulated data, such as HIPAA, NIST SP 800-171, NIST SP 800-53, and ITAR.

• Availability and serviceability: high availability of the scheduler, storage, and management services; hot-serviceable components where possible; fault isolation such that a single failure does not take down large portions of the system; and clear mean-time-to-repair commitments.

• Monitoring and transparency: system, job, power, and thermal telemetry available to Empire AI operations staff and, where appropriate, to users.

This is not a solicitation. Page 16 of 40

• Operations model: Empire AI is evaluating whether Gamma should be operated by Empire AI/UB staff, by the Respondent as a managed service, or under a hybrid model. Respondents should describe the hosting and operating models they support (see Section VIII).

4. Storage and Data Requirements Gamma will require high-performance parallel storage sized and tuned for the suggested compute, as well as capacity storage and data-protection provisions. Beta currently includes 20 PB of VAST and 10 PB of DDN storage. Respondents should describe their recommended storage architecture for Gamma, the expected usable capacity and sustained throughput, and how Gamma storage would interoperate with Beta storage, whether or not Beta is relocated to the Facility.

This is not a solicitation. Page 17 of 40

V. SCOPE OF INFORMATION REQUESTED — COMPUTING ASSETS

1. Functional and Performance Requirements Empire AI seeks information and suggested system configurations for a high-performance, accelerator-based computing infrastructure capable of supporting a diverse portfolio of advanced AI and scientific computing workloads. The Gamma deployment is planned for up to 5 MW of IT load, using AC power distribution and liquid cooling through the Facility's Technical Cooling System (TCS).

Respondents are encouraged to suggest architectures that maximize useful research capability within the available budget and the Facility's power, cooling, and space constraints. Suggested configurations may incorporate tightly integrated rack-scale systems, modular accelerator-based clusters, or other architectures that meet the Consortium's scientific, operational, and performance objectives.

Respondents should recommend Gamma's architecture, scale, timing, and budget allocation as part of an integrated strategy to maximize Empire AI's useful research capability over time, taking account of Beta's existing capabilities and remaining useful life, the incremental research needs Gamma should address, and the resources and flexibility to preserve for Delta.

Empire AI does not prescribe a particular processor architecture, accelerator technology, interconnect, system configuration, or manufacturer. Respondents should demonstrate the suitability of their suggested configuration for the anticipated research workloads, including large-scale AI model training, fine-tuning, inference, and scientific computing, with particular emphasis on performance, scalability, energy efficiency, software ecosystem, operational flexibility, and total cost of ownership.

Empire AI intends to evaluate suggested configurations primarily on their ability to deliver useful scientific and AI research throughput within defined financial, power, cooling, and operational constraints, rather than on peak theoretical performance or conformity to a particular hardware architecture. Respondents should quantify performance per dollar, performance per watt, and time to scientific solution using representative workloads wherever practical.

Role of Gamma. Empire AI is seeking Respondents' recommendations on how Gamma should complement Beta and position the program for Delta, rather than prescribing Gamma's role in advance. Respondents should explain which research needs Beta can continue to serve effectively, which limitations Gamma should address now, and which investments might reasonably be deferred to Delta. Capability-class ("hero") jobs, meaning very large training, simulation, and AI-for-science runs that use a substantial fraction of the system as a single, tightly coupled allocation, are an important use case, and Respondents should also plan for model hosting and inference workloads. A predominantly capability-focused Gamma is an acceptable outcome if the tradeoffs support it.

Regardless of vendor or platform, Respondents should describe the largest job (in accelerators, aggregate memory, and sustained performance) their configuration could run efficiently as a single allocation, and the scale-up domain, scale-out fabric, and storage characteristics that make this possible.

Rack input voltage: solutions that accept 480 V AC at the rack will be given preference over 415 V solutions. Respondents offering a 415 V solution should state whether a 480 V AC rack input option is available or planned for their platform, and the associated schedule and cost impact.

Respondents should suggest the configuration they believe delivers the greatest useful research capability across Empire AI's existing and planned systems over time, not only within Gamma, given

This is not a solicitation. Page 18 of 40 their recommended Gamma Allocation (Section VII). Respondents should explain the incremental research benefit Gamma would provide alongside Beta, how that benefit would evolve after Delta arrives, and why their suggested balance of architecture, size, and timing represents the best use of the program's resources. Respondents should also identify how much of the 5 MW capacity their suggested configuration would consume. A configuration that uses less than 5 MW is acceptable if it is justified by budget, technology timing, or spend-rate considerations.

2. Specific Computing Information Requested Respondents shall complete the Computing Equipment Technical Data Sheet (Attachment B) and should address the following in their narrative response:

2.1 Accelerators and Compute Nodes

• Accelerator model, generation, and quantity per rack and in total; accelerator memory type, capacity, and bandwidth; performance across numerical precisions relevant to target workloads, with Respondents identifying supported formats, dense versus sparse execution, and comparable benchmark results.

• CPU model, architecture (Arm/x86), core count, memory type and capacity per node, and CPU-to-accelerator interconnect and bandwidth.

• Compute system configuration, including rack-scale integration, accelerator density, number of compute racks, scale-up domain characteristics, and auxiliary nodes (login, management, service, CPU-only, inference-optimized, or visualization nodes).

2.2 Scale-Up and Scale-Out Networking

• Accelerator interconnect architecture, scale-up domain size, topology, and effective communication bandwidth per accelerator.

• Scale-out (compute) fabric: technology (InfiniBand or Ethernet), switch and NIC models, per-accelerator bandwidth, topology, oversubscription ratio, and optics strategy (pluggable vs. co-packaged).

• Front-end, storage, management, and out-of-band networks; DPUs or SmartNICs, if suggested; and the external connectivity to UB, member institutions, and the Beta system.

2.3 Storage

• Parallel file system or storage platform, usable capacity, sustained read/write throughput, metadata performance, tiering, and data-protection approach; integration with existing VAST and DDN systems.

2.4 Software, Management, and Security

• Operating system, cluster management and provisioning, platform management capabilities

(including monitoring, fault management, and system lifecycle management), scheduler and orchestration, AI software stack, and licensing terms (subscription term, renewal pricing, and what happens at expiration).

• Security architecture, firmware and supply chain assurance, identity integration, and vulnerability management.

2.5 Rack Power, Cooling, and Physical Characteristics

• Rack dimensions; static and dynamic weight and point loads; nominal, peak, and maximum power per rack; input voltage, connector type, and power shelf configuration; and any rack-level energy storage or power-smoothing features.

This is not a solicitation. Page 19 of 40

• Liquid-cooled fraction of rack heat; coolant type; required supply temperature range; flow rate, pressure drop, and temperature rise per rack; water quality and filtration requirements; and residual heat to air.

• Electrical behavior relevant to facility design, including load transients and power swings during training, inrush, harmonics, and power factor.

2.6 Performance, Delivery, and Support

• Expected performance on standard benchmarks (e.g., HPL, HPCG, MLPerf Training and

Inference) and on representative research workloads; Respondents should suggest reference workloads Empire AI could use in an RFP.

• Manufacturing and delivery lead times, factory integration and testing (rack-level and cluster-level), on-site installation duration, and acceptance testing approach.

• Warranty, support tiers, on-site staffing, spare parts strategy, and mean-time-to-repair commitments; and expected useful life and refresh considerations.

3. Future Deployment (Delta) Respondents should describe their technology roadmap through the Delta timeframe, including anticipated accelerator generations, rack power densities, coolant temperature trends, and the transition to 800 V DC power distribution anticipated for Delta. Respondents should identify any provisions Empire AI should make now — in the building, the white space, or the Gamma architecture

— to allow Delta to be deployed without major redesign, and how Gamma would interoperate with later deployments (for example, shared storage, fabrics, and management). Because Delta is planned as the second tranche of the Program Budget approximately two to three years after Gamma (Section VII), Respondents should describe how their Gamma configuration would be extended or complemented by Delta.

This is not a solicitation. Page 20 of 40

VI. SCOPE OF INFORMATION REQUESTED — WHITE SPACE

INFRASTRUCTURE

1. General White space infrastructure comprises the mechanical, electrical, controls, and physical systems located downstream of the building points of connection (Section III.7) that are required to deploy and operate the Gamma computing system. Empire AI anticipates that the white space infrastructure will be delivered by the computing Respondent directly or through a team that includes qualified engineering and construction partners, so that a single party is accountable for a working, integrated system. Respondents should describe their suggested delivery structure as described below.

2. Contracting Structure, General Conditions, and Delivery Team

2.1 Construction Manager at Risk Agreement

No contract will result from this RFI. Under any subsequent procurement, the white space infrastructure portion of the work is expected to be subject to a Construction Manager at Risk ("CMaR") agreement. The computing assets portion is expected to be procured under a separate equipment and services agreement. Respondents should describe how they would structure the contracting for the white space work under a CMaR agreement, including which entity would hold that agreement, how cost, schedule, and performance risk would be allocated between the computing and white space portions, and how the two portions would be coordinated so that Empire AI receives a single integrated, working system.

2.2 General Conditions

Not required in RFI responses; will be required at the time of the RFP. The requirements in Sections VI.2.2 through VI.2.4 are provided so that Respondents understand how the white space work is expected to be delivered. At the RFP stage, Respondents' white space pricing and approach will be required to include all general conditions necessary to deliver the work, including at a minimum:

• Permitting and approvals: all permits, inspections, and approvals required for the white space work, coordinated with the authorities having jurisdiction, the University at Buffalo, and the State University Construction Fund as applicable.

• Engineering design documents: complete design documents for the white space work, prepared under the direction of, and stamped and sealed by, professional engineers licensed and registered in New York State, by firms holding the required New York State authorization to provide engineering services; coordination with the Facility's Engineer of Record (Gensler) and the base building CMaR; and BIM coordination.

• Insurance: insurance coverages, limits, and additional-insured requirements as will be specified in the RFP, and payment and performance bonds as required.

• Construction management: CMaR services for the white space work, including preconstruction services, cost estimating and control, scheduling, subcontractor procurement, site supervision, safety, quality control, coordination with the base building CMaR and site logistics, commissioning support, and closeout.

• Labor and compliance: compliance with New York State prevailing wage requirements, including certified payroll, and with all applicable federal, state, and local laws and regulations.

2.3 Diversity Participation Goals and Project Labor Agreement The following participation goals apply to the white space infrastructure portion of the work only;

they do not apply to the computing assets portion:

This is not a solicitation. Page 21 of 40

Category Participation Goal Minority-Owned Business Enterprise (MBE) 16% Women-Owned Business Enterprise (WBE) 14% Service-Disabled Veteran-Owned Business (SDVOB) 6%

Participation shall be by firms certified by New York State. The white space work may be subject to a Project Labor Agreement ("PLA"). A description of the Respondent's approach to meeting the participation goals and its experience working under PLAs will be required at the time of the RFP. For this RFI, Respondents may note any concerns regarding the participation goals or the PLA that Empire AI should consider.

2.4 Vetted Delivery Partners

While not required in this phase, a list of vetted partners that are appropriately licensed in New York State to provide design and construction services for the white space work will be required at the time of the RFP. For each partner, the RFP will request the partner's role, its New York State licensure and authorization, its relevant liquid-cooled and high-density data center experience, any prior work with the Respondent, and its MBE, WBE, or SDVOB certification status. Respondents may identify anticipated partners in their RFI response.

3. Technical Cooling System and Coolant Distribution Units Respondents should describe, and reflect in their indicative white space pricing (Attachment A), the technical cooling infrastructure from the building branch connections to the computing racks, including:

• CDUs: type (in-row or facility-scale liquid-to-liquid), capacity per unit, quantity, redundancy (N+1 or other), approach temperature, pump redundancy, controls, and electrical requirements. The DD documents provide for up to fifteen (15) CDU connections;

Respondents should confirm whether this is adequate for their suggested configuration.

• Secondary (technology-side) piping from CDUs to rack manifolds, including material (e.g., stainless steel), routing (overhead or under-floor) from the ground-level CDU room to racks in the second-level data hall, isolation and control valves, strainers and filtration, balancing devices, air separation, expansion, and drain and fill provisions.

• Rack manifolds, hoses, and quick-disconnect couplings, and the interface to the computing manufacturer's rack plumbing.

• Leak detection and automatic isolation strategy; pressure, temperature, and flow monitoring.

• Coolant: initial fill, fluid type, water treatment and chemistry monitoring, flushing and passivation, and ongoing fluid management.

• Recommended facility-side supply temperature and any requirements the Respondent would place on the building primary loop (flow, differential pressure, and water quality).

4. Critical Power Infrastructure Respondents should describe, and reflect in their indicative white space pricing (Attachment A), the critical power infrastructure from the building 480 V main switchboards to the computing racks, including:

• Flywheel UPS: battery-free rotary flywheel UPS systems (basis of design: 1,500 kVA units, four (4) for Gamma), including input/output and maintenance bypass switchgear, ride-through duration at full load, efficiency, and the integration approach with the building switchboards.

This is not a solicitation. Page 22 of 40

• 480 V distribution panelboards serving CDUs and other white space loads (the DD documents currently show CDU feeders from building panel DP-B on static UPS-B; final assignment per Attachment D), and any 480 V distribution to the computing equipment.

• Overhead compute busway with tap-off boxes (480 V for 480 V AC rack solutions, which are preferred). For 415 V rack solutions only, IT distribution transformers (480–415/240 V, basis of design 750 kVA, eight (8) units) and 415/240 V busway, priced as alternates pending final scope assignment (see Attachment D).

• Rack power connections (busway plug-in units or whips), branch circuit power monitoring and metering, grounding and bonding within the data hall, and emergency power-off provisions as required by code.

• Power quality coordination, including harmonics, transient load behavior of the computing system, and coordination with the flywheel UPS and utility service.

• Recommended provisions for the future transition to 800 V DC distribution for Delta.

5. Physical Infrastructure

• Rack placement and layout, hot- or cold-aisle containment as needed for the residual air-cooled load, rack anchoring and seismic restraint, and floor loading verification.

• Overhead support structure for busway, piping, and cable tray, coordinated with the building ceiling strut grid; cable tray and ladder rack; and fiber and copper pathways within the data hall.

6. Controls, Monitoring, and Integration

• DCIM and/or platform management system, and integration of CDUs, flywheel UPS, panelboards, metering, leak detection, and rack telemetry with the BMS using open protocols (e.g., BACnet, Modbus, SNMP, Redfish), including point lists, alarm priorities, trending, and cybersecurity coordination.

7. Design, Installation, Commissioning, and Turnover

• Engineering design, coordination with the Engineer of Record and Construction Manager at

Risk, BIM coordination, permitting, and submittals.

• Installation of white space infrastructure and computing equipment, phased logistics, and coordination with base building completion in Q4 2027 and a target Gamma power-on of approximately July 2028.

• Commissioning support at all levels (factory testing through integrated systems testing), load-bank testing of power and cooling systems, technology burn-in, operator training, and turnover documentation.

This is not a solicitation. Page 23 of 40

VII. BUDGET AND PRICING INFORMATION

1. Program Budget and Tranches Empire AI anticipates approximately $160 million for computing assets and associated white space infrastructure, to be deployed in two tranches, Gamma and Delta, approximately two to three years apart (the "Program Budget").

Empire AI is not prescribing a fixed Gamma budget. Respondents should recommend how the Program Budget should be divided between the Gamma and Delta tranches. The Gamma portion of that recommendation (the "Gamma Allocation") should be all-inclusive of the following:

• Computing assets (Section V), including compute, networking, storage, software, and management systems;

• White space infrastructure (Section VI), including CDUs and the technical cooling system, flywheel UPS, 480 V panelboards, and related distribution, controls, and physical infrastructure;

• Engineering, installation, integration, commissioning, acceptance testing, training, and documentation; and

• Hardware warranty and support, and software subscriptions, for an initial term of three (3) years. Respondents should also price a five (5)-year term as an option, stating the maximum support term offered by each OEM and how any period beyond it would be covered (e.g., OEM extension, third-party support, or Respondent-provided).

The Program Budget excludes the base building and its systems (gray space) (Section III), utility service and energy costs, Empire AI and UB staff costs, and Owner's representative, design, and commissioning-authority fees. Indicative pricing should exclude New York State and local sales and use taxes; Empire AI is tax-exempt and will furnish its exemption certificate. Pricing should include freight to the Facility and should separately identify any import tariffs or duties included, with the assumptions used.

2. Requested Approach Respondents should recommend Gamma's architecture, scale, timing, and budget allocation as part of an integrated strategy to maximize Empire AI's useful research capability over time, taking account of Beta's existing capabilities and remaining useful life, the incremental research needs Gamma should address, and the resources and flexibility to preserve for Delta. In addition to a base configuration at the recommended Gamma Allocation, Respondents may provide up to two (2) alternative configurations (for example, at Gamma Allocations of approximately $80 million and $100 million) to illustrate how capability scales with investment, and should state which configuration they recommend and why.

3. Spend Rate and Timing A central question for Empire AI is how to divide the Program Budget between the Gamma and Delta tranches. Respondents are asked to recommend a spend rate and tranche split across the program based on their projected technology evolution, cost trends, and supply conditions, and to provide a detailed rationale, including the expected price-performance improvement between generations and the risks of committing too much or too little at Gamma, including market volatility and technology obsolescence.

This is not a solicitation. Page 24 of 40

4. Pricing Format Indicative pricing shall be provided on the…