Acquisition, delivery, installation and hardware and software maintenance of NLAIF Supercomputer for the European High Performance Computing Joint Und...

EuroHPC JU has launched a competitive dialogue tender (EUROHPC/2026/CD/0003) to procure, install and maintain the NLAIF AI‑optimised supercomputer to be owned by EuroHPC JU and hosted by SURF in Groningen, Netherlands. The estimated contract value is €64,100,000 (excl. VAT) with a maximum duration of 72 months including up to five years of operations and maintenance, a 10% performance guarantee and optional pre‑financing up to 30% subject to a guarantee. The procedure is two‑step: Requests to Participate (deadline 30 April 2026) will prequalify candidates for a dialogue phase (up to three shortlisted) and final tenders, which will be evaluated on technical value (40%), cost‑performance/TCO (30%), services quality (20%) and EU added value (10%). Key technical requirements include a GPU‑centric accelerated partition, multi‑tier storage (NVMe, large warm tier, object and tape), RDMA‑capable networking, direct liquid cooling targeting PUE<1.2, specified AI/HPC benchmarks for acceptance, and at least 4% of nodes using European technology.

Procurement title

Acquisition, delivery, installation and hardware and software maintenance of NLAIF Supercomputer for the EuroHPC JU (EUROHPC/2026/CD/0003)

What is procured:supply, delivery, assembly, installation, commissioning and hardware/software maintenance of one AI-ready high-end EuroHPC supercomputer (to establish the NLAIF AI Factory). The system must be AI-optimised (GPU-centric partitions, tiered storage, networking, software stack) and meet mandatory performance, energy and usability targets.

Estimated contract value:Approximately €64 100 000 (contract estimated value) ¹.

1. 1Who may apply: economic operators (single suppliers or consortia). Subcontractors and group members must meet exclusion and selection rules.
2. 2Procedure: competitive dialogue with two-stage process — request to participate (selection), dialogue sessions, then invited final tender; award by best price-quality ratio; maximum contract duration 72 months.

Eligibility and requirements:applicants must demonstrate proven experience in acquiring, delivering, installing and maintaining supercomputers (references, financial and technical documentation, staffing). Security, supply‑chain resilience and inclusion of European technology (minimum share requirement) are mandatory parts of the tender evaluation.

Support and documents:tender specifications, financial offer templates, exclusion/selection declaration forms and questionnaire templates are published on the F&T Portal; submissions and official Q&A are handled via the portal. ¹

Footnotes

1. 1Official tender page and downloadable documents: F&T Portal – Tender details.

The European High-Performance Computing Joint Undertaking (EuroHPC JU) is running a Competitive Dialogue procurement to select one economic operator to acquire, deliver, assemble, install (hardware and software), and maintain a world-class AI-optimised supercomputer, together with training and support services. The machine will be owned by EuroHPC JU and hosted by the Dutch Hosting Entity (SURF) to underpin the Dutch AI Factory (NLAIF) in the province of Groningen, the Netherlands. The contract is classified as supplies, awarded on the basis of best price-quality ratio, with an estimated total value of €64,100,000 and a maximum duration of 72 months including up to 5 years of operation. Submission is electronic via the EU Funding & Tenders Portal eSubmission system. TED reference: 49/2026 169488-2026.

Key deadlines and identifiers:Procedure identifier: EUROHPC/2026/CD/0003. TED publication date: 11/03/2026. Deadline for receipt of requests to participate: 30/04/2026 16:00 (Europe/Luxembourg). Submission method: Electronic (eSubmission). CPV: 30211100 Super computer. Nature of contract: Supplies. Award method: Best price-quality ratio. Estimated total value: €64,100,000. Maximum contract duration: 72 months. Lead contracting authority: EuroHPC JU, Luxembourg.

Procurement structure and stages:This is a Competitive Dialogue with three phases: 1) Selection Stage: Request to Participate and pre-qualification against exclusion and selection criteria; 2) Dialogue Stage: at least two interactive sessions per candidate, optional site visits, iterative technical and commercial refinement, distribution of benchmarks and datasets; 3) Tendering Stage: submission and evaluation of Final Tenders solely against the published award criteria. Only pre-selected candidates from Stage 1 will be invited to the Dialogue and to submit a Final Tender.

Purpose and scope of supply

EuroHPC JU seeks a contractor with proven experience delivering supercomputers in comparable environments to procure, install, and maintain an AI-optimised supercomputer for the NLAIF. The solution must outperform existing systems across a range of dimensions, including performance per total cost of ownership, usability for AI model development and deployment, versatility, system stability, power and energy efficiency, and computing density. The system must include EU technologies and components, with a minimum of 4% of total node count comprising European technology (e.g., CPUs, accelerators, NICs). The contractor must also provide training and knowledge transfer.

Target architecture and technical requirements

Overall system design and partitions

GPU-centric, heterogeneous architecture supporting large-scale AI training, high-throughput analytics, and inference. The design includes distinct partitions and shared services, enabling both HPC-AI and cloud-native workflows, and enforcing security, multi-tenancy, and data integrity.

• Accelerated partition: High-density AI accelerators delivering multiple Exaflop/s theoretical FP16 (dense) performance. Accelerators organised into High-Bandwidth Domains (HBDs) of at least 4, preferably 8, devices with uniform, all-to-all high-speed fabric. Host subsystem x86 or ARM with sufficient CPU-accelerator I/O, memory sized at roughly 1.5x aggregate accelerator memory per node unless justified alternative; offload via SmartNIC/DPU-class functions or equivalent software design; scale-out network sized to avoid structural bottlenecks and sustain predictable multi-user concurrency.
• Scalar partition: CPU-only nodes for interactive use, data pre/post-processing, hybrid AI-HPC coupling, and CPU-centric inference; at least 48 compute nodes expected; heterogeneous CPU profiles (high-core/high-frequency) and high-speed interconnect; support for composability/disaggregation for memory-intensive tasks.
• Specialized partition (optional): One coherent accelerator-based solution optimised for a declared focus such as cost-efficient inference (tokens per euro/watt), low-latency serving, or secure/confidential computing. Must fully integrate into the operational environment and show measurable benefit relative to an economically equivalent share of the accelerated partition.
• Service and management nodes: Architecturally equivalent to scalar nodes; at least one node per flavour outside production for maintenance, validation, and testing.

Storage solution and data management

Centralised, multi-tier storage aligned with AI data lifecycles, accessible from all worker nodes including batch and containerised workloads. The design balances performance, scalability, security, and cost-efficiency, with governance for GDPR and sectoral compliance.

• Hot tier: Shared parallel filesystem, NVMe SSD, POSIX-like mount, RDMA and GPU-Direct Storage-class capabilities, container-mountable; at least 15 PiB NVMe-backed capacity guaranteed effective logical capacity if data reduction is used.
• Warm tier: Shared parallel filesystem, HDD and/or SSD, POSIX-like mount; optimised for sustained throughput and sequential access; at least 60 PiB; supports home directories, staff-managed apps, and 15–25 PiB of curated datasets; periodic incremental backups to cold storage with efficient restore.
• Lukewarm tier: S3-compatible object storage (≥ 5 PiB) for federation and exchange with peer AI Factories and external platforms; supports versioning, lifecycle, and hybrid workflows.
• Cold storage: Tape archive for protected backups of selected warm-tier data.
• Data lakes: Central metadata catalogues, federated access to national/European repositories and EOSC platforms, and in-place analytics to minimise data movement.
• Security and governance: Encryption at rest/in transit, role-based access control, multi-tenancy and isolation (enclaves and segmentation), comprehensive audit trails.

Networking

• Front-end datacentre network: Ethernet, 200–400 GbE per server typical; user/API/control-plane and non-latency-critical data flows; optional SmartNIC/DPU for security and policy.
• Front-end storage network: High-speed Ethernet or InfiniBand depending on storage backend; supports NVMe-oF, parallel filesystems, NFS, and S3; engineered for sustained read/write throughput, including checkpoint bursts.
• Scale-up accelerator fabric: NVLink/UALink or similar high-bandwidth, low-latency connectivity within HBDs (and possibly across nodes), critical to tensor/pipeline parallelism efficiency and low oversubscription.
• Scale-out cluster network (east–west): RDMA-capable inter-node fabric such as InfiniBand, UEC, or UALink/UALoE-class solutions; carries collectives (all-reduce/all-to-all) for multi-node training with predictable, congestion-managed performance.
• Out-of-band management network: Dedicated, physically isolated 1/10/25 GbE for BMC/IPMI/Redfish, power control, console access, continuous health monitoring, and secure firmware management.

Software stack and computing environment

• Software layering: Hardware/network drivers; OS and system libraries; compilers, MPI, device runtimes; toolchain libraries (BLAS, FFT, I/O); AI lifecycle tools (e.g., MLFlow, NeMo, Neptune). EESSI stack available to support EuroHPC Federation Platform interoperability.
• Access patterns: Traditional HPC via login and batch scheduler (e.g., Slurm); cloud-native via orchestrator (e.g., Kubernetes) for containerised AI apps; and Open OnDemand for a user-friendly, web-based portal integrated with both HPC and orchestration platforms.
• Hybrid HPC-cloud operation: Dynamic repartitioning and interoperability between batch-scheduled and containerised AI workloads.

Cooling, power, and sustainability

Target PUE below 1.2 through a state-of-the-art datacentre and direct liquid cooling at high temperatures allowing near-continuous free cooling. Integrate real-time power telemetry, dynamic power management (voltage-frequency scaling, workload-aware power capping), renewable energy procurement, and extensive energy monitoring dashboards at infrastructure and application levels.

Testing, benchmarks, and acceptance

Acceptance emphasises AI-centric performance, scalability, energy efficiency, and resilience. Benchmarks include synthetic, training, and inference workloads, executed as both performance-optimised and energy-optimised runs. Certain benchmarks have minimum required performance; failure on any such test results in rejection. Weighted scoring applies, and sustained performance over system lifetime is required.

• Key tests: HPL-MxP (mixed-precision FLOPS); IO500 (throughput, metadata, latency); GEMM performance at FP4/FP8/FP16/FP32; STREAM and BabelStream (memory bandwidth); OSU Microbenchmarks (latency/bandwidth).
• Key end-to-end benchmarks: MLPerf Training (time-to-train to target quality across model classes); LLM inference using vLLM (tokens/sec per user and per GPU, with area-under-curve focus reflecting concurrency/throughput trade-offs).
• Energy evaluation: Joules-to-solution and average watts over runtime using tools such as PowerAPI, Energy Aware Runtime, and/or Slurm integrations.

Hosting site and indicative schedule

Hosting entity:SURF, the Dutch education and research cooperative. The datacentre will be in the province of Groningen, Netherlands. The facility will provide at least €200 M² of IT space, ≥ 3 MW peak power (expandable), support for high-density racks and multiple cooling technologies (air, DLC, rear-door heat exchangers), tiered redundancy, ISO 27001 certification, 24/7 security and CCTV, heavy-load logistics (≥ 3000 kg pre-cabled racks/CDUs at 220 cm x 150 cm x 230 cm), spare parts space, and remote visitor access management.

• Q3 2026: Datacentre provider contract signature.
• Q2 2027: Datacentre preparations ready.
• Q2–Q3 2027: System installation and acceptance.
• Q3 2027: Start of operations of main partition.
• Default assumption: Single-stage installation in Q3 2027 (CPU, storage, GPU). Phased delivery may be considered during the procedure.

Contracting, guarantees, support, and service levels

• Contract language: English. Owner: EuroHPC JU. One contract with the successful tenderer.
• Financial guarantees: Performance guarantee of 10% of contract value; optional pre-financing up to 30% requires equivalent pre-financing guarantee; contract can proceed without pre-financing if guarantee not provided.
• Maintenance and support: At least 5 years; 5x10h support (Mon–Fri, 08:00–18:00 CET); critical component response within 1 hour during prime hours; qualified partner personnel on site within 2 hours for critical issues; next-business-day model for non-critical single-node issues; planned maintenance in scheduled windows; all software updates included; support in English and/or Dutch.
• Lifecycle assurances: No diminished support of offered HW/SW for first 6 years post-install; immediate critical security patches; full English documentation for admins and users.
• Metrics and monitoring: Granular, low-overhead power/energy/temperature measurements per node type and infrastructure elements; export via CSV or API; job-level energy metering; separate measurement of infrastructure functions (e.g., cooling, power distribution).

Eligibility, access to procurement, and restrictions

• Access to procurement: Open on equal terms to natural and legal persons within the scope of the EU Treaties and to international organisations; also open to operators established in third countries with special agreements with the Union in the field of public procurement, and to countries that have ratified the WTO GPA, under applicable conditions.
• Restrictive measures: All involved entities and subcontractors must not be subject to EU restrictive measures (asset freezes and prohibitions on making funds available).
• Next Generation European AI partition: Inclusion of a partition based on next-generation European technology (accelerated processors) is required; ownership control assessment applies to suppliers providing European technology chips (CPUs, GPUs, accelerators, or other computational elements). The assessment obligation concerns the European chip provider; the contracting consortium itself is not subject to this specific assessment if it does not supply the chips.
• High-risk suppliers: Entities assessed as high-risk suppliers of 5G/6G mobile network equipment (and entities they own/control) are not eligible in any capacity. Equipment/services related to European communication networks must not be subject to third-country security requirements that could affect execution, must comply with Commission cybersecurity guidance (5G toolbox), apply lifecycle cybersecurity requirements, and ensure availability, authenticity, integrity, and confidentiality.

Exclusion and selection criteria (Selection Stage)

Administrative documents

• Cover letter (max 1 page, authorised signature, contact details).
• List of documents to be provided.
• Form for identification of the candidate (Annex 3).
• Questionnaire for joint requests and subcontracting (Annex 4A; complete per subcontractor).
• Power of Attorney for groups of economic operators (Annex €4B).
• Letter(s) of intent from subcontractor(s) (Annex 4C).
• Mutual Confidentiality Agreement signed by all parties involved (Annex 4D).
• Financial identification form (bank stamp/signature or recent bank statement).
• Legal entity form with annexes: proof of registration; VAT registration (if not shown); appointment and authorisation evidence of legal representatives; instruments of incorporation/statutes; relevant powers of attorney where applicable.

Exclusion criteria and evidence

• Declaration on honour on exclusion and selection criteria signed by an authorised representative (Annex 6).
• Before contract signature for the awarded tenderer: recent extract from judicial record; recent certificates on payment of taxes and social security contributions (or sworn/solemn statements where not issued in the country of establishment).
• Applies to each member in case of joint request and to subcontractors above 15% share or whose capacity is needed for selection criteria. Evidence may be waived if already provided to EuroHPC JU within the last year and unchanged, or accessible free of charge on a national database.

Economic and financial capacity

• Declaration on honour (Annex 6).
• Economic and financial capacity questionnaire (Annex 7).
• Balance sheets for 2023 and 2024.
• Profit and loss accounts for 2023 and 2024.

Technical and professional capacity

• Short description of the candidate’s economic activity (max 10 pages overall) focused on R&D and building supercomputers; show alignment with EuroHPC JU objectives.
• Installation Project Activity Reference Forms (IPARFs, Annex 10A): three references executed within the preceding 4 years; indicate amount, date, location, recipient; demonstrate fit to similar environments; include quality assessment with respect to systems listed in any Top500 list since June 2018; references evidenced by recipient certificates or operator declaration. Framework contracts acceptable; use one IPARF per project.
• Educational and professional qualifications statement: demonstrate staffing with at least three years’ experience per role and coverage by different persons across systems architect, systems engineer, data centre infrastructure specialist, application/benchmarking specialist, and project manager.
• Professional qualifications: relevant certificates by independent bodies.
• Technical certifications: relevant conformity certificates by official quality control institutes or authorised departments.

Award criteria (Final Tender evaluation)

• Criterion 1: Technical value of the system design.
• Criterion 2: Cost-performance analysis based on performance evaluation and total cost of ownership.
• Criterion 3: Quality of services.
• Criterion 4: EU Added Value, including inclusion of EU technologies and contribution to EuroHPC JU objectives.

Submission, communication, and procedural rules

• Requests to Participate must be submitted exclusively via eSubmission available from the Funding & Tenders Portal opportunity page. Each organisation must have a valid PIC (9-digit Participant Identification Code).
• Only one Request to Participate per candidate; the latest timely submission counts if multiple versions are submitted. Submissions received after the deadline are rejected.
• Queries: via the F&T Portal Questions & answers section only, by the specified deadline; answers are published for all interested parties; no individual guidance on eligibility.
• Opening sessions (both stages) are not public; confidentiality applies to all submissions.
• Pricing: all-inclusive in euros, VAT-exempt for EU institutions; no decimals; unambiguous and unconditional; conditional offers are not permitted.
• Data protection: Processing under Regulation (EU) 2018/1725; EDES registration may apply in exclusion cases.
• Applicable law and jurisdiction: EU law complemented, where necessary, by Luxembourg law. Environmental, social and labour obligations per Directive 2014/24/EU Annex X apply.
• EuroHPC JU may cancel the procedure before signature without compensation; tender documents are non-binding until contract signature.

Financial and contractual specifics

• Estimated total value: €64,100,000.
• Guarantees: 10% performance guarantee; optional pre-financing up to 30% against equivalent guarantee.
• Payments: Contractual payments against deliverables/services; VAT-exempt regime for EU institutions.
• Liquidated damages: May apply proportionally in case of delays or failures, subject to contract terms.
• Framework agreement: Not indicated; single contract with the successful tenderer.

Templates and required forms (application package)

Candidates must structure their Selection Stage submissions using the provided templates and checklists. Completion quality and compliance directly affect pre-qualification.

Categorisation and structured extraction

Eligible Applicant Types:Economic operators capable of supplying, integrating, and maintaining supercomputers: large enterprises, SMEs, systems integrators, hardware manufacturers, software vendors, and consortia of such entities. Universities and research institutes may participate as members of a consortium or as subcontractors where appropriate to meet technical capacity and service obligations. Public-private partnerships and international organisations may participate if within access to procurement rules. Individual natural persons are not expected to meet capacity requirements alone but may be involved within a consortium or as subcontractors.

Funding Type:Procurement — supplies contract with associated services (installation, maintenance, training). The beneficiary receives remuneration under a service and supply contract, not a grant.

Consortium Requirement:Single contractor will be awarded one contract. Joint requests/joint tenders by groups of economic operators are permitted, with a designated representative. Subcontracting is allowed. The group composition should remain stable through to contract signature; any restructuring requires authorisation and must continue to meet eligibility and selection criteria.

Beneficiary Scope (Geographic Eligibility):Open to natural and legal persons within the scope of the EU Treaties and to international organisations; open to third-country operators under special agreements with the EU and to countries that have ratified the WTO GPA, subject to agreement conditions. Restrictions apply to entities under EU sanctions and to high-risk 5G/6G suppliers. Ownership control assessment applies to European chip technology providers for the Next Generation European AI partition.

Target Sector:High-performance computing and artificial intelligence: supercomputing systems, AI accelerators, advanced interconnects, scalable storage, orchestration platforms, security/compliance for sensitive data, and energy-efficient data centre engineering. Downstream supported application areas include healthcare, mobility, finance, agriculture, and other sectors requiring large-scale AI training and inference.

Mentioned Countries:Luxembourg (contracting authority location), Netherlands (hosting site in the province of Groningen). Region: European Union.

Project Stage:Implementation, deployment, and operation of a production AI supercomputing system with high-TRL components; includes installation, validation/acceptance testing, training, and multi-year maintenance.

Funding Amount:Estimated total contract value is €64,100,000. Pre-financing up to 30% may be requested, subject to an equivalent pre-financing guarantee. A performance guarantee of 10% of the contract value is required.

Application Type:Open call for tenders using a two-step Competitive Dialogue: 1) open Request to Participate; 2) invitation-only Dialogue and Final Tender.

Nature of Support:Money via a supply and services contract; payments against deliverables and service levels. No grant funding is provided.

Application Stages:Three stages in total: Selection Stage (Request to Participate), Dialogue Stage (interactive technical/legal/financial refinement with at least two dialogue sessions), Tendering Stage (Final Tender submission and evaluation).

Success Rates:Not disclosed. The number of candidates invited to Dialogue is up to three, subject to the number of operators meeting exclusion and selection criteria.

Co-funding Requirement:No co-funding by the contractor is required. Financial guarantees are mandatory: 10% performance guarantee and, if pre-financing is requested, a pre-financing guarantee equal to the pre-financing amount.

How to prepare and submit

1. 1Register and obtain or reuse your PIC; ensure EU Login access and eSubmission readiness.
2. 2Assemble the Request to Participate using the Questionnaire template (Annex 10D) to track completeness of administrative, exclusion, financial, and technical documents.
3. 3Complete and sign all required forms (cover letter, identification form, declarations, confidentiality agreement, legal and financial identity forms, and—if applicable—joint tender and subcontracting forms).
4. 4Prepare technical capacity evidence: three IPARFs per Annex 10A, qualifications statements, professional and technical certifications, and the short description of economic activity aligned with EuroHPC objectives.
5. 5Upload the complete Request to Participate via the F&T Portal before the deadline. Monitor the portal Q&A and notifications for updates.
6. 6If selected, participate in the Dialogue Stage, run required benchmarks/datasets, refine architecture, SLAs, pricing, and EU added value components.
7. 7Submit the Final Tender via eSubmission in line with the final Invitation to Tender (ITT) and the Financial Response template, covering acquisition, installation, training, multi-year maintenance, and dismantling.

Contacts and official resources

• Opportunity page and eSubmission: EU Funding & Tenders Portal — Acquisition, delivery, installation and hardware and software maintenance of NLAIF Supercomputer F&T Portal Opportunity.
• Contracting authority: European High-Performance Computing Joint Undertaking (EuroHPC JU), 12, Rue Guillaume Kroll, Technopolis Building - Wing E, L-2920 Luxembourg, Luxembourg.
• General EuroHPC JU information: EuroHPC JU website.

Summary: What this opportunity is about

EuroHPC JU is procuring, via a Competitive Dialogue, a turnkey, large-scale AI-optimised supercomputing system for the Dutch AI Factory (NLAIF). The contractor will deliver a GPU-centric, heterogeneous platform optimised for training and inference of state-of-the-art AI models, backed by high-performance, multi-tier storage, layered networking fabrics, and a hybrid HPC-cloud software environment, integrated with Open OnDemand. The system must include at least 4% European technology by node count and comply with stringent supply chain security and cybersecurity restrictions, including the exclusion of high-risk 5G/6G suppliers. Acceptance focuses on AI-centric benchmarks (e.g., HPL-MxP, IO500, MLPerf Training, vLLM inference) and energy efficiency, with performance sustained over the system lifetime. The contract covers acquisition, delivery, installation, training, and a minimum of five years of hardware and software maintenance with demanding service levels. Financially, the tender value is approximately €64.1 million, with a 10% performance guarantee required and optional pre-financing up to 30% backed by a guarantee. The process is open to international economic operators within EU procurement access rules, with joint tenders and subcontracting allowed. Applications proceed in three steps: submit a Request to Participate; engage in at least two Dialogue sessions with EuroHPC JU and the Hosting Entity (SURF); then submit a Final Tender evaluated on technical merit, TCO-based cost-performance, service quality, and EU added value. The hosting site will be in the province of Groningen, with datacentre readiness in Q2 2027 and operations of the main partition targeted for Q3 2027. This is a major opportunity for capable HPC/AI integrators and technology providers to deploy a sovereign, energy-efficient, production-grade AI supercomputing capability serving European scientific, industrial, and public sector users.

Footnotes

1. 1All details, timelines, documents, and forms are provided on the EU Funding & Tenders Portal opportunity page. Always consult the portal for authoritative updates and the latest versions of templates and specifications. See: F&T Portal Opportunity.

Overview

This is a competitive dialogue tender (EUROHPC/2026/CD/0003) issued by the European High-Performance Computing Joint Undertaking (EuroHPC JU) to select one economic operator for the acquisition, delivery, assembly, hardware and software installation, and maintenance of an AI-optimised supercomputer, designated NLAIF, to establish an AI Factory. The supercomputer will be owned by EuroHPC JU and hosted by SURF in the Netherlands. Solutions must outperform existing systems in performance/TCO, programmability/usability, versatility, system stability, power/energy efficiency, and computing density.

The tender follows a two-step process:first, submission of requests to participate for selection; second, dialogue with selected candidates leading to final tenders. It is classified under CPV 30211100 (Super computer) as a supplies contract.

Key Dates and Deadlines

• TED publication date: 11 March 2026
• Deadline for requests to participate: 30 April 2026 at 16:00 Europe/Luxembourg time
• Estimated contract signature: January 2027
• System installation start: Q2 2027
• Acceptance: Q3 2027
• Operations start: Q3 2027

Financial Details

Estimated total value:€64,100,000 (excluding VAT).

Contract duration:Maximum 72 months, including up to 5 years of operations and maintenance.

Financial guarantees:Performance guarantee of 10% of contract value; pre-financing guarantee (up to 30% of contract price) if requested.

Eligibility and Who Can Apply

Open to natural and legal persons within the scope of EU Treaties, international organisations, and entities from third countries with relevant agreements (e.g., WTO GPA signatories). Joint tenders and subcontracting allowed, but group composition cannot change post-request to participate without approval. Subcontractors with >15% share or key to selection criteria must meet exclusion/selection rules.

Restrictions:No EU restrictive measures; no high-risk suppliers for 5G/6G equipment; Next Generation European AI partition requires ownership control assessment for European tech providers (min. 4% of nodes). No conflicting interests.

Exclusion Criteria

Candidates excluded for bankruptcy, tax/social security breaches, grave professional misconduct, fraud/corruption, criminal offences, significant contract deficiencies, irregularities, shell entities, or resisting audits. Declaration on honour required (Annex 6); supporting evidence post-selection.

Selection Criteria

• Economic/financial capacity: Balance sheets/profit-loss for 2023-2024; questionnaire (Annex 7).
• Technical/professional capacity: Proven experience with 3 recent supercomputer projects (IPARFs, Annex 10A) in Top500 since June 2018, executed within 4 years; staff qualifications for key roles (systems architect, engineer, data centre specialist, application/benchmarking specialist, project manager) with >=3 years experience; max 10-page economic activity description.

Up to 3 candidates selected based on ranking for dialogue stage.

Technical Requirements

GPU-centric AI supercomputer for training/inference, with accelerated (multi-Exaflop FP16), scalar (min. 48 CPU nodes), optional specialised partitions, multi-tier storage (hot >=15 PiB NVMe, warm >=60 PiB, lukewarm >=5 PiB object, cold tape), layered networking (Ethernet front-end, InfiniBand/UALink scale-out/up, OOB management), hybrid HPC-cloud (Slurm/Kubernetes), Open OnDemand, direct liquid cooling (PUE<1.2), benchmarks (MLPerf, IO500, HPL-MxP, etc.). Min. 4% European tech nodes. Hosted in Groningen, NL datacentre (ready Q2 2027).

Procedure and Submission

Electronic submission via EU Funding & Tenders Portal eSubmission. Requires PIC registration. Documents: cover letter, identification form (Annex 3), declaration on honour (Annex 6), financial/legal forms, IPARFs, staff statements, financial response template (Annex 2). Max 200 files, <50MB each. English preferred.

Dialogue and Award:Selected candidates enter dialogue (Jun-Sep 2026) for solution refinement; final tenders evaluated on technical value (40%), cost-performance/TCO (30%), services quality (20%), EU added value (10%). Best price-quality ratio.

Maintenance and Support

• 5x10h support (Mon-Fri 08:00-18:00 CET); 1h response for critical, on-site within 2h.
• Next business day for non-critical single nodes.
• All updates/releases included; 6-year support guarantee; English/Dutch.
• Metrics/logging for power/energy; dismantling price required.

Key Documents and Portal

All documents available at EU Funding & Tenders Portal. TED reference: 49/2026 169488-2026. Contact via Q&A section; subscribe for updates.

Detailed specs in Descriptive Document (01a), Invitation (03), templates (04,06,07). Review system requirements for eSubmission.

Footnotes

1. 1Data extracted from official tender portal and documents as of March 2026. Verify latest at primary URL for updates.