UConn Inst for AI Innovation Tech Sol RFP CDR010826 Q&A Addendum 2.pdf

RFP Title: AI Innovation Institute Technology Solutions

RFP#: CDR010826

RE: Question and Answer Addendum 2 Date: January 30, 2026

The following questions and answers will be considered part of the solicitation documents.

Question 1: Is UConn expecting full stack development as the skillset on RFP pages 4-5 suggests?

Answer 1: Refer to RFP Section 3.0 – Scope of Work. The University solicitation reflects a solutions-based model under which the proposer is “expected to deliver state-of-the-art AI resources and supporting software solutions.” The anticipated support “includes set-up and initial configuration, and continuous administration, monitoring, optimization, and support of the resource.”

Question 2: Can the proposal leverage ARM processors such as those contained in the GB300 from NVIDIA, or do you require x86.

Answer 2: Refer to RFP Section 3.0 – Hardware and Software Needs, “the University is seeking the latest generation GPUs and AI accelerator (AIA) nodes." The proposer will propose the architecture that it believes best fulfills UConn’s needs.

Question 3: Computing power of 30pflops as a requirement is low in consideration of the "latest generation" of GPUs. Is this a per GPU metric?

Answer 3: Refer to RFP Section 2.0, the University seeks foundational infrastructure that "addresses the broader AI needs of the University." The proposer will propose the architecture that it believes best fulfills UConn’s needs and addresses cost-effectiveness and feasibility. The University will evaluate proposals in accordance with the Evaluation Criteria in the RFP.

Question 4: For the 15pb of storage, what is the breakdown between high speed and archival storage?

Answer 4: In the competitive process, the University will not advise the proposer on the content of the proposal. It is left to the discretion of the proposer to define a tiered storage strategy to meet UConn’s "overall goal of related storage (approximately 15PB)."

Question 5: Is infiniband an acceptable protocol for the high-speed redundant fabric for intra-cluster networking?

RFP# CDR010826

Question and Answer Addendum 2

Answer 5: In the competitive process, the University will not advise the proposer on the content of the proposal. It is left to the discretion of the proposer to propose a high-speed, redundant data fabric supporting within the cluster aligned to its proposed solution.

Question 6: Is there a ratio of on-premise vs cloud vs hybrid resources for end state?

Answer 6: The solicitation reflects a solutions-based approach under which the proposer is invited to propose the architecture that it believes best fulfills UConn’s needs as stated in the RFP. The University will not pre-evaluate the proposal.

Question 7: Is there an existing monitoring or observability platform to integrate with for the managed services requirement?

Answer 7: It is left to the discretion of the proposer to propose the means and method of administration and monitoring of the resource.

Question 8: Is there a ticket management system to integrate with?

Answer 8: The University utilizes Jira for its ticketing software, as part of the Atlassian suite.

Question 9: With respect to FISMA and NIST - what are the specific controls required for each?

Answer 9: Refer to RFP Section 3.0 – Required Features. Compliance with federal data security frameworks includes at least NIST SP 800-171 and FISMA. The proposer will articulate how the design of the proposed solution supports the ability to meet the requirements in Question

2.2.12 in the HuskyBuy platform.

Question 10: What are the current needs to be accommodated immediately in the cloud including the date required?

Answer 10: The solicitation reflects a solutions-based approach under which the proposer is invited to propose the project plan and timeline that it believes best fulfills UConn’s needs as stated in the RFP. The University will not pre-evaluate the proposal.

Question 11: What is the approximate division in terms of resources for prod/test/uat?

Answer 11: The division has not been determined or defined by UConn. It is left to the discretion of the proposer to propose a breakdown that ensures "stability, security, and performance" across all environments.

Question 12: Is physical network separation required for the different environments (prod/test/uat) or is logical separation allowed?

Answer 12: Refer to RFP Section 3.0 – Required Features, “… (e)ach Enclave must be logically segregated (and physically segregated where appropriate).”

Question 13: What are the physical resource requirements for the different enclaves?

Answer 13: Refer to RFP Section 3.0 – Required Features. It is left to the discretion of the proposer to define and describe an environment that includes general, HIPAA, and federal enclaves.

Question 14: What are the physical isolation requirements for the different enclaves?

Answer 14: Refer to RFP Section 3.0 – Required Features, “… [e]ach Enclave must be logically segregated (and physically segregated where appropriate) with clearly defined security boundaries … .”

Question 15: What is the backup/restore environment, or is this in addition to other requirements in this RFP? If it is in addition, what are the RPO and RTO objectives? Are there DR requirements?

Answer 15: It is left to the discretion of the proposer to articulate the proposed backup/restore processes.

Question 16: Does “periodic refreshes based on technological advances" refer to software, hardware, both?

Answer 16: Both.

Question 17: Does Uconn want the respondent to manage all OEM warranty/incident/breakfix/RMA as part of the managed services requirement?

Answer 17: Refer to RFP Section 3.0 – Technical & Programmatic Support, “[t]he awarded proposer will run the AI resource as a managed service, ensuring stability, security, and performance on a 24/7 basis.” It is left to the discretion of the proposer to propose the managed services and articulate assumptions related to support by UConn personnel, if any.

The proposal will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 18: Does Uconn want the respondent to manage the workflow with the OEMs for any breakfix/callout activity, or hand off to Uconn to manage?

Answer 18: Refer to Answer 17 above.

Question 19: What change management processes/procedures are currently in place that the respondent will need to adhere to and/or interface with?

Answer 19: It is left to the discretion of the proposer to articulate its capabilities with respect to change management processes/procedures.

Question 20: With respect to a possible liquid cooling solution, as well as colo-infra, will the colo provider be responsible for electrical and cool water connections?

Answer 20: The identity of a colo provider, if applicable, has not been determined by the University. The proposer may articulate any assumptions with respect to a colo provider in the proposal. The proposal will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 21: Will colo provider provide CDUs in the event of a liquid cooling solution? If so, who is responsible for support and routine maintenance of the equipment

Answer 21: Refer to Answer 20 above.

Question 22: Is the respondent expected to be "smart hands" on-site or is this part of the colo agreement?

Answer 22: Refer to Answer 20 above.

Question 23: Is equipment outside of an OEM rack (i.e. storage, networking, etc), does the colo provider provide racks, or is there a preference on rack vendors?

Answer 23: Refer to Answer 20 above.

Question 24: Is there a maximum rack density (in terms of KW) per rack supported by the colo?

Answer 24: Refer to Answer 20 above.

Question 25: What compute performance precision is used for the 30 PFLOPS requirement?

e.g. FP8 or FP16/BF16 or FP32, etc.?

Answer 25: In the competitive process, the University will not pre-evaluate the proposal. The proposer will articulate the precision level in the proposal. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 26: What is the priority from a budget perspective? 400 GPUs and 400 AIAs or the compute performance of greater than 30 PFLOPs?

Answer 26: In the competitive process, the University will not pre-evaluate the proposal.

Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 27: Can UConn provide a forecast of how much of the compute capacity will be utilized at a given time?

Answer 27: Refer to RFP Section 2.0, “[t]he AI computing resource will serve as the foundational infrastructure for the AI Innovation Institute and address the broader AI needs of the University. It is contemplated that the University will be able to utilize the computing power delivered by the resource. The proposer will propose the architecture that it believes best fulfills UConn’s needs and to continuously enhance the resource.

Question 28: Organizations are able to scale GPU usage incrementally based on demand - in a pattern of 80, 160, 280, and 400 GPUs, with proportional storage at each stage. Can UConn confirm whether this incremental scaling pattern aligns with the anticipated usage trajectory?

Answer 28: In the competitive process, the University will not pre-evaluate the proposal. The proposer may articulate proposed incremental scaling as part of the system architecture, project plan, and timeline within the proposal. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 29: Does Uconn have a funding source or not to exceed budget identified for this project?

Answer 29: UConn is not disclosing funding sources or budget as part of the solicitation process.

Question 30: What is an AIA and how does it differ from a GPU enabled server?

Answer 30: Refer to RFP Section 3.0 – Hardware and Software Needs, AIA is an "AI accelerator" node.

Question 31: Is CUDA software a requirement for the computing platform and programming model?

Answer 31: It is left to the discretion of the proposer to include proposed software in the proposal. Refer to the RFP Section 2.0, the system will serve as a hub for research, development, education, etc., across areas including "AI/ML development" and "natural language processing.”

Question 32: Is the compute capacity goal of 30 PFLOPS measured with peak theoretical 64 bit double precision performance? 32 Bit single precision.

Answer 32: Refer to Answer 3 and Answer 25 above.

Question 33: The RFP notes install, testing and validation of multiple environments (test, UAT, and production). Are test and UAT workloads included or in addition to the required 400 GPU and 15PB of storage?

Answer 33: The proposer will propose the system architecture that it believes best fulfills UConn’s needs. The University will evaluate proposals in accordance with the Evaluation Criteria in the RFP.

Question 34: The RFP notes "Software and ML models for different applications" - What are some of those applications the University plans to leverage?

Answer 34: Refer to RFP Section 2.0. Seven areas of application are identified on page 3.

Question 35: Is 15PB an initial usable capacity target? What is expected Year over Year Growth?

Answer 35: Yes. Growth of 5PB per year is anticipated.

Question 36: Does UConn have an intial target capacity per enclave (General, HIPAA, Federal)?

Answer 36: The proposer will propose the system architecture that it believes best fulfills UConn’s needs. The University will evaluate proposals in accordance with the Evaluation Criteria in the RFP.

Question 37: Are there critical datasets that must always remain on high‑performance storage, regardless of age or last access time?

Answer 37: Not at the present time. Each proposer may articulate the capabilities of its system design to meet future needs in the proposal.

Question 38: For UConn’s most demanding AI workloads, is there throughput/IOPS targets (per node or aggregate) we should design to for storage performance?

Answer 38: The proposer will propose the system architecture that it believes best fulfills UConn’s needs. The University will evaluate proposals in accordance with the Evaluation Criteria in the RFP.

Question 39: At peak, how many concurrent training/fine‑tuning jobs do you anticipate, and what does a typical “busy season” vs average usage pattern look like (e.g., semester‑based peaks)?

Answer 39: Each proposer may articulate the capabilities of its system design to meet present and future needs in the proposal. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 40: Does UConn expect to run large multi‑node distributed training jobs that heavily stress shared storage (checkpoints, shuffles, data‑parallel reads)?

Answer 40: Yes.

Question 41: Does UConn plan to import any large existing datasets (multi‑PB) into this platform in the early phases, and if so, from which current environments (HPC, other NAS, cloud buckets, removable media)?

Answer 41: The import of existing datasets is to be determined.

Question 42: What are the dominant data modalities UConn expects (e.g., medical images, video, text corpora, tabular, sensor/IoT, model checkpoints)?

Answer 42: A combination of the modalities listed.

Question 43: What is UConn’s expected file size distribution (e.g., mostly large multi‑GB files vs billions of small files), and do any workloads depend on very high file counts in a single directory or dataset?

Answer 43: Unknown.

Question 44: For UConn AI workloads, characterize the typical access patterns.

Answer 44: Unpredictable at this stage.

Question 45: Are there important non‑GPU consumers of this data (e.g., traditional HPC, BI/analytics tools, departmental file shares) that need to share this same platform?

Answer 45: Yes.

Question 46: Is there any need for S3 protocol from PIs?

Answer 46: The available protocols may be articulated by the proposer in the proposal.

Question 47: Is Uconn using any data movement products their environment today that they would want implemented in this footprint?

Answer 47: Each proposer may articulate the capabilities of its system design to meet present and future needs in the proposal. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 48: Has UConn identified/contracted with a data center/colo to house the on-prem infrastructure?

Answer 48: No.

Question 49: What is the power and cooling infrastructure Uconn has planned to support the on prem solution?

Answer 49: Refer to Answer 20 above.

Question 50: If UConn is looking for bidder to provide a Colo as part of their bid, are the geographic restrictions on where that facility can be located?

Answer 50: It is left to the discretion of the proposer to articulate its proposed solution.

Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 51: Will Uconn facilities be able to support direct to chip liquid cooled hardware?

Answer 51: No.

Question 52: Is the provider responsible for system level backups/restores or for data backups/restores as well? Will UConn provide the backup/restore HW and SW with licensing allowing for use by an MSP or is the provider to furnish the necessary HW and SW? If the latter, does UConn require integration into their current backup/restore environment? If the provider must integrate, what is the current backup/restore SW? Please provide the backup frequency and retention requirements.

Answer 52: Refer to RFP Section 3.0, “… [t]he awarded proposer will run the AI resource as a managed service.” Each proposer will articulate its capabilities to address the listed issues in the proposal. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 53: The RFP calls out specific personnel roles. Are these resources expected to be available 40 hours a week / 24x7 / on call for specific project work? Please describe in more detail the expectations for the roles listed in the RFP.

Answer 53: Refer to the RFP. The resource must ensure "stability, security, and performance on a 24/7 basis." Each proposer will articulate its capabilities and method of addressing the listed issues in the proposal.

Question 54: Will the provider take over the current cloud services and provide all functions required to ensure successful operations? Will UCONN provide the licenses for all cloud services or will the provider need to provide cloud licensing?

Answer 54: Refer to the RFP Section 3.0. The AI Innovation Institute resource will operate as a managed service for UConn's use on an ongoing basis. It is left to the discretion of the proposer to articulate its system design and architecture in the proposal.

Question 55: In order to provide configurations and services for test, UAT, and production environments, provide the storage capacity, CPU/GPU performance requirements for each environment within each enclave.

Answer 55: Refer to RFP Section 3.0 – Required Features. It is left to the discretion of the proposer to define and describe an environment that includes general, HIPAA, and federal enclaves.

Question 56: HIPAA - Are there requirements for on shore resources to deliver this service, or are off-shore, non-US resources acceptable?

Answer 56: Refer to RFP Section 3.0. The proposed solution must support "HIPAA compliance, including technical, administrative, and physical safeguards." It is left to the discretion of the proposer to articulate how its system design and architecture address the standards.

Question 57: Are there requirements for on shore resources to deliver this service, or are off-shore, non-US resources acceptable for any positions/roles? If not are US Persons acceptable or do all personnel have to be US Citizen/soil? Are there any ITAR or export control data restrictions? Do any other university security policies, state or Federal regulations apply? E.g.

NIST SP800-53 rev5.

Answer 57: Refer to RFP Section 3.0. It is left to the discretion of the proposer to articulate how its system design and architecture address all required standards and security frameworks.

Question 58: Is a combined monitoring solution (one that crosses the general purpose, HIPAA, and Federal enclaves) acceptable?

Answer 58: The University will not pre-evaluate the proposal. It is left to the discretion of the proposer to articulate how its system design and architecture address all required standards and security frameworks.

Question 59: Does the University require integration with existing observability and monitoring solutions?

Answer 59: Refer to RFP Section 3.0. The anticipated support includes continuous administration and monitoring by the awarded proposer.

Question 60: Does the University require integration with existing ITSM? If so, which ITSM is currently being utilized by UCONN and what level of integration is required? Simple read only access, standard bi-directional, or customized bi-directional integrations?

Answer 60: Refer to Answer 59 above.

Question 61: SIEM/logging is assumed to be a customer responsibility. Is this correct?

Answer 61: Refer to Answer 59 above.

Question 62: Any specific encryption requirements that need to be part of the architectural design? What level of FIPS 140 is required for each of the enclaves?

Answer 62: Refer to RFP Section 3.0. It is left to the discretion of the proposer to articulate how its system design and architecture address all required standards and security frameworks.

Question 63: Will the University utilize their own EDR or should the provider bring in a solution?

Answer 63: Refer to RFP Section 3.0 – Hardware and Software Needs, the proposer is to propose the Software and ML models for the different applications described in the RFP.

Question 64: Will the University handle the certification of the Federal Enclave to meet the FISMA compliance/meeting the NIST SP 800-171 certification requirements?

Answer 64: It is left to the discretion of the proposer to propose the managed services and articulate assumptions related to support by UConn personnel, if any. The proposal will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 65: Who will own the Audit & Review Cycle for the enclaves?

Answer 65: Refer to Answer 64 above.

Question 66: Are all of the following services in scope for the vendor: Hardware architecture;

rack integration; physical hardware deployment; software stack design, installation, and configuration?

Answer 66: Yes.

Question 67: Are there specific AI use cases or domains that should be prioritized when tailoring the initial solution architecture? Or, will AI use cases be identified post-deployment by applicable stakeholders at time of need.

Answer 67: Refer to RFP Section 2.0 for the list of seven primary domains. Priority will be determined by the University at the time of need.

Question 68: Does UConn require the definition of standard AI solution blueprints as part of initial design deliverables? For example, digital twin, chatbot, digital assistant, etc.

Answer 68: The University will not pre-evaluate the proposal. It is left to the discretion of the proposer to articulate its capabilities, system design, value-added considerations, etc., in the proposal.

Question 69: Provide additional context in the tasks and responsibilities of the operational resources listed in the RFP (which are assumed to be different and separate from Managed Services resources).

Answer 69: Application Developers: responsible for the supporting software solutions aligned to the Institute's goals; System Administrators: responsible for continuous administration and OS upgrades; Software Engineers: involved in the design, building, and support of the AI computing resource; Data Scientists: support the research and workforce development initiatives of the Institute; Network and Security Engineers: responsible for the redundant data fabric and security boundaries of the Enclaves.

Question 70: The operational resources, are there expectations in terms of level and location of engagement (part-time, full-time, onshore, offshore, onsite)?

Answer 70: It is left to the discretion of the proposer to articulate its proposed solution to meet the needs of the University.

Question 71: Is UConn receptive to a pooled staffing model whereby the specific resources and skillsets can be determined and onboarded at the time of needs, versus a predefined and fixed staffing model?

Answer 71: It is left to the discretion of the proposer to articulate its proposed solution to meet the needs of the University. The University will not pre-evaluate the proposal.

Question 72: Does UConn expect the managed service provider to sign a BAA as it relates to the HIPAA and FED Compliant Enclaves?

Answer 72: Yes.

Question 73: Is the University looking for a single solution provider or would the university prefer choosing multiple responses and combine them as needed to achieve the University goals?

Answer 73: Refer to RFP Section 3.0, “[i]t is the University’s intent to make a single award as a result of this RFP. However, the University reserves the right to make multiple awards if it is deemed by the University to be in the University’s best interest to do so. The University will not pre-evaluate the proposal.

Question 74: Is the university interested in having a data services layer included in the RFP response that would provide ingest, data transformation capability, metadata/analytics/vector database services, data table services, self service deployment for models, LLMs and agents, in addition to the infrastructure as a service? Or do you intend to have each enclave select their own data service toolsets?

Answer 74: Refer to RFP Section 3.0, the minimum needs include, “Software and ML models for different applications.” It is left to the discretion of the proposer to articulate its capabilities, system design, value-added considerations, etc., in the proposal. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 75: Is there a FedRAMP baseline or impact level target in mind for the federal enclave environment? (Low/Med/High)?

Answer 75: Refer to RFP Section 3.0 – Required Features. Compliance with federal data security frameworks includes at least NIST SP 800-171 and FISMA. The proposer will articulate how the design of the proposed solution supports the ability to meet the requirements in Question 2.2.12 in the HuskyBuy platform.

Question 76: Who is responsible for the security of customer developed applications, and is the vendor expected to provide the environment, tooling, and support for vulnerability management and security scanning of such applications?

Answer 76: Refer to RFP Section 3.0 – Technical & Programmatic Support, “[t]he awarded proposer will run the AI resource as a managed service, ensuring stability, security, and performance on a 24/7 basis.” It is left to the discretion of the proposer to propose the managed services, including ability to ensure security of customer-developed applications, and articulate assumptions related to support by UConn personnel, if any. The proposal will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 77: Are there data residency restrictions, such as PHI or CUI being required to remain on-premises or within the state?

Answer 77: Refer to RFP Section 3.0 – Required Features, compliance with all applicable security frameworks is required. The proposer will articulate how the design of the proposed solution supports the ability to meet applicable security frameworks in Question 2.2.12 in the HuskyBuy platform.

Question 78: Does UConn expect this AI platform to be positioned explicitly in grant proposals as a named institutional capability, and should bidders assume it must be defensible to external reviewers as a differentiator rather than just baseline infrastructure?

Answer 78: Refer to RFP Section 3.0, Value-Added Considerations – Future Collaborations, proposers are invited to include in the proposal any value-added features or approaches that assist the University in realizing the goal of, “ … large grant opportunities involving transformative AI solutions …", among others. The University will not pre-evaluate the proposal.

Question 79: Overall goal of related storage (approximately 15PB) - are we bidding for the frontend network as well as the backend (GPUs/AIAs)?

Answer 79: Refer to the RFP. The University is seeking "state-of-the-art AI resources and supporting software solutions."

Question 80: Does UConn prefer certain GPU, AIA, or network fabric vendors?

Answer 80: The University is conducting an open, competitive solicitation. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP.

Question 81: The RFP lists 3 enclaves and 800 GPUs/AIAs, does the whole network support 800 GPUs/AIAs or 2400 GPUs/AIAs 800 per enclave?

Answer 81: The need is for 800 GPUs/AIAs across all three enclaves.

Question 82: What primary software/models is the university looking towards in the environment?

Answer 82: Refer to RFP Section 2.0. Seven areas of application are identified on page 3.

Question 83: Are there any specific requirements that UConn foresee to drive on-prem consumption of GPU resources?

Answer 83: Not at this point.

Question 84: Is UConn looking for 400 GPUs and 400 AI Accelerator nodes so total of 800 nodes?

Answer 84: Yes.

Question 85: What existing security parameters e.g. identity etc. should the proposed solution federate with for compliance requirements?

Answer 85: Microsoft Entra, with separate Entra Government tenant for NIST 800-171 and FISMA enclave.

Question 86: What is the basis for the pflop and pbyte sizings?

Answer 86: A survey of current users.

Question 87: Do you have incumbent network and/or storage vendors instantiated, that must be supported in the proposal?

Answer 87: No.

Question 88: Does UConn have any requirements relative to workload management, observability, chargebacks, resource allocations, etc.?

Answer 88: The awarded proposer is responsible for continuous administration and monitoring of the resource. It is left to the discretion of the proposer to articulate its proposed managed services and capabilities related to the items listed.

Question 89: Is UConn intending to serve high-availability long-running inference services on the cluster?

Answer 89: Yes. Refer to RFP Section 3.0, Required Features, services are to be architected with high availability in mind.

Question 90: Will a multi-path fabric suffice, or do you require literally redundant network fabrics?

Answer 90: Refer to RFP Section 3.0, Hardware and Software Needs, minimum requirements include, “high-speed, redundant data fabric supporting within the cluster.”

Question 91: Physical segregation may be mandated by certain standards, how do you anticipate carving up the overall cluster in that scenario?

Answer 91: Refer to RFP Section 3.0 – Required Features, “… [e]ach Enclave must be logically segregated (and physically segregated where appropriate) with clearly defined security boundaries … .” It is left to the discretion of the proposer to propose the manner of physical segregation to meet the University’s needs.

Question 92: Will UConn have physical space for the AI Institute (a collaboration space and office space for onsite people)?

Answer 92: When the AI Institute is established, there will be associated physical space.

Question 93: Explain "A high-speed, redundant data fabric supporting within the cluster."

Answer 93: The fabric must be "redundant" to ensure high availability, meaning maintenance or failure of individual components (switches, cables, etc.) must not disrupt the continuity of the overall service.

Question 94: Does UConn want to build 2 clusters?

Answer 94: The proposer will propose the architecture that it believes best fulfills UConn’s needs and addresses cost-effectiveness and feasibility. The University will evaluate proposals in accordance with the Evaluation Criteria in the RFP.

Question 95: Does the University have a strategic implementation roadmap, that could potentially define the solution implementation as this is a multi-year implementation. At a minimum, does the University know what it is looking to achieve in the immediate term and what timeframe?

Answer 95: Refer to the RFP. Proposals will be evaluated in accordance with the Evaluation Criteria in the RFP, including project plan and timeline.

Question 96: Should bidders assume the goal is to deploy the most current generation of GPU at time of award, or should proposals emphasize a structured refresh strategy over the contract term to maintain grant competitiveness?

Answer 96: The University will not pre-evaluate the proposal. The proposer will propose the architecture that it believes best fulfills UConn’s needs and addresses cost-effectiveness and feasibility. The University will evaluate proposals in accordance with the Evaluation Criteria in the RFP.

Question 97: What secure enclaves (HIPAA and FISMA/800-171 compliant) currently exist, on-prem or in the cloud, that will require connectivity to the respective secure enclaves in the proposed solution?

Answer 97: An Azure government tenant exists for existing NIST 800-171 workloads. A similar solution exists for HIPAA, but not in the Azure cloud. The proposer may address connectivity capabilities in the proposal.

Question 98: Are there specific use cases that have already been identified?

Answer 98: Refer to RFP Section 2.0. Seven areas of application are identified on page 3.

Question 99: What are the specific user personas (faculty, students, researchers, external partners) and their expected workloads/use cases?

Answer 99: Personas include the university community of over 40,000 faculty members, students and staff. Workloads range from innovative academic research to HIPAA-compliant medical research to Department of Defense secure research.

Question 100: Is the vendor expected to provide foundational or opensource base models such as Llama, Mistral, or other large language models? If yes, is there a requirement for domain specific models such as medical NLP, genomics, finance, or engineering models?

Answer 100: In the competitive process, the University will not advise the proposer on the content of the proposal. The proposed software or language models is left to the discretion of the proposer.

Question 101: Is the vendor is responsible for model training? If yes , will the UConn provide all datasets and domain expertise required for training or finetuning?

Answer 101: No, the vendor is not responsible for model training.

Question 102: Is there a requirement for model security scanning or detection of model poisoning or tampering?

Answer 102: It is left to the discretion of the proposer to articulate how its system design and architecture address all required standards and security frameworks, which includes

"logging/monitoring, incident response, and access controls."

Question 103: Does the customer require formal compliance attestations such as HIPAA or NIST 800‑171 evidence?

Answer 103: Yes, for the those enclaves.

Question 104: What aspects of the AI environment are most critical to control directly versus consume as a service? particularly around grant competitiveness

Answer 104: The University intends to contract with a provider to "design, build, supply, install, support, and continuously enhance" the resource as a managed service.

Question 105: Are “economic factors” (including Connecticut-based participation, supplier diversity, and in-state economic impact) scored as a discrete subcomponent within the 30% Financial Proposal category, and if so, what approximate weighting or relative importance they carry compared to pricing and cost-effectiveness?

Answer 105: Each proposal will be evaluated in accordance with the Evaluation Criteria as stated in the RFP.

Question 106: Is UConn expecting a solution provider break-fix individual to reside each day in the data center or is the expectation "dispatch" only?

Answer 106: Refer to RFP Section 3.0 – Hardware and Software Needs, the identified minimum is “… 24/7/365 support with a 4-hour onsite response time for physical equipment issues, and a 4-hour response time for cloud service incidents.”

Question 107: Is UConn expecting an onprem, cloud, or hybrid first architecture based on internal constraints or preferences?

Answer 107: It is left to the discretion of the proposer to articulate a solution aligned to its business methods. Proposals will be evaluated in accordance with the Evaluation Criteria in the

RFP.

Question 108: Whether physically onsite or dispatch, is the Solution Provider the lead or will they be working with local UConn datacenter smart hands to perform the remediation?

Answer 108: Refer to the RFP. The awarded proposer is responsible for "supplying the personnel to manage and operate the AI resource."

Question 109: Is there a ServiceNow instance that the solution provider can use to either manual or automated (API) ticket creation for either hardware or software non-conforming events?

Answer 109: The service is to be managed by the awarded proposer.

Question 110: Is UConn expecting the solution provider to lead the data science / machine learning effort for UConn or is the requirement more aligned with ensuring that the environment (Slurm, Kubernetes, etc.) is created and managed (no visibility to data) by the Solution Provider? so that the lead UConn Data Scientists can create jobs that support their research efforts?

Answer 110: Refer to Answer 69 above. The awarded proposer will not be expected to lead the data science / machine learning effort for UConn.

Question 111: Should managed detection and response cover only the infrastructure, or also AI workloads, API gateways, and model execution environments?

Answer 111: The awarded proposer will "run the AI resource as a managed service, ensuring stability, security, and performance" on a 24/7 basis.

Question 112: Should cloud consumption charges be billed directly to the customer or invoiced through the vendor’s managed service?

Answer 112: It is left to the discretion of the proposer to propose a solution aligned to its business methods.

Question 113: Is forensic investigation support required as part of incident response?

Answer 113: Yes, required as part of the NIST/FISMA framework.

Question 114: Will the GPU and AIA capacity need to integrate with an existing on-premise or UConn cloud infrastructure, or will this be greenfield / standalone capacity? (Is this adding to/scaling an existing cluster, or standing up net new capacity?)

Answer 114: The resource will serve as the "foundational infrastructure" for the AI Innovation Institute. It is left to the discretion of the proposer to articulate the proposed system modalities.

Question 115: Is the 15 PB storage requirement inclusive of replication, backups, and snapshots?

Answer 115: No.

Question 116: Will security incidents be handled jointly with Uconn's security team?

Answer 116: The proposer will articulate how the proposed solution supports the ability to meet the requirements of all compliance requirements.

Question 117: Is integration with existing high performance computing clusters required?

Answer 117: The resource is intended as a university-wide asset to address the "broader AI needs of the University". Specific integration with existing clusters is not mandated.

Question 118: Should the overall solution integrate with existing monitoring tools used by UConn?

Answer 118: No.

Question 119: Is the platform to connect to existing data lakes, research databases, or institutional storage systems?

Answer 119: No requirement is currently identified. The proposer will articulate in the proposal how data gets in and out of the service.

Question 120: What onboarding workflows should be in place for new research groups or departments using the platform?

Answer 120: Refer to the RFP. The resource must enable "equitable access for all stakeholders". The proposer will define administration and access management in the proposal.

Question 121: Are there any regulatory, statutory, or organizational requirements regarding country of origin, manufacturing location, or supply chain sourcing for hardware platforms proposed in this solution?

Answer 121: The enclave(s) must meet federal data security frameworks including NIST SP 800- 171 and FISMA. The proposer will articulate how the proposed solution supports the ability to meet the requirements of all compliance requirements.

End of Question and Answer Addendum 2