Procurement, Delivery and Installation Support for AI Data Center Infrastructure.
CustomerALT Infrastructure SA
PartnerAmber AI & Data Science Solutions GmbH
Partner LocationOberhaching, Germany
Date[●]
160 MW
Phase 1 IT Load · Poland
500–1,000
NVL72-Class Racks
Q3 2027
Phased Commissioning Target
60 Days
20/20/20 · +60 Day Extension
01 — Parties
The Customer, the Partner, and the Project Region.
This Letter of Intent (this “LOI”) is entered into between:
Customer
ALT Infrastructure SA
A Swiss société anonyme (CHE-409.502.681) with registered office in Switzerland, acting for itself and its Affiliates.
Partner
Amber AI & Data Science Solutions GmbH
A German GmbH at Grünwalder Weg 32, 82041 Oberhaching, Germany — an NVIDIA Elite Partner with NVIDIA DGX SuperPOD certification.
“Affiliates” means entities controlling, controlled by, or under common control with the Customer, including its current and future operating project companies in the Republic of Poland and elsewhere in Central and Eastern Europe (the “Project Region”).
02 — Purpose
Cooperation on AI factory infrastructure across the Project Region.
The Parties wish to cooperate on the planning, procurement, delivery, integration and installation support of NVIDIA-based AI factory infrastructure to be developed by the Customer in the Project Region (the “Project”). Indicative Phase 1 technical scope is set out in Annex A. Subsequent phases are described directionally in Annex B.
03 — Intended Scope of Cooperation
What the Partner intends to provide, and the Customer intends to procure.
Subject to the Definitive Agreements (defined in §07), the Partner intends to provide, and the Customer intends to procure:
1
Procurement coordination for NVIDIA compute (Vera Rubin NVL72 and, where required for early deployment, Blackwell-generation GB300 NVL72), NVIDIA networking (Quantum-X InfiniBand and Spectrum-X Ethernet), high-performance storage, and liquid-cooled rack infrastructure.
2
Technical planning aligned with NVIDIA Reference Architecture standards (cluster, fabric, storage, density, cooling, power and thermal integration).
3
Deployment and installation support — delivery and logistics, rack integration, staging, commissioning, cluster bring-up and validation — in coordination with the Customer's appointed facility partners.
4
Project coordination during deployment phases, with post-deployment lifecycle support to be specified in the Definitive Agreements.
04 — NVIDIA Allocation
Allocation engagement, status updates, and the absence of any deposit commitment.
The Partner shall use commercially reasonable efforts to engage NVIDIA and applicable OEM/ODM channels promptly and shall provide written status updates at twenty (20) days, forty (40) days, and in any event no later than sixty (60) days of the Effective Date, addressing:
(i)NVIDIA allocation availability for the Phase 1 indicative configuration;
(ii)indicative lead times; and
(iii)any known supply or export licensing constraints.
The Parties may, in the Definitive Agreements, agree on a refundable reservation deposit or other commercial mechanism to support allocation. No such commitment is created by this LOI.
05 — Mutual Project Exclusivity
Material inducement: who can't talk to whom during the Term. Binding
During the Term, as a material inducement to enter into the Definitive Agreements:
Partner exclusivity
No competing Polish AI factory LOI
The Partner shall not enter into any letter of intent, term sheet or comparable commitment with any other data centre developer or operator in the Republic of Poland for AI factory infrastructure of comparable scale and scope to the Project.
Customer exclusivity
No competing German NVIDIA integrator
The Customer shall not engage any other NVIDIA system integrator established in the Federal Republic of Germany for AI factory infrastructure of comparable scale and scope to the Project.
Additionally, during the Term, the Partner shall not, on the basis of information disclosed by the Customer in this LOI (including its Annexes), solicit or pursue discussions with any third party concerning AI factory infrastructure projects of comparable scale and scope in the Central and Eastern European region.
06 — Confidentiality
Non-public information, site identity, allocation status — for five (5) years post-termination. Binding
Each Party shall hold the other's non-public information (“Confidential Information”), including the existence and contents of this LOI, the identity and location of the Customer's sites, technical architectures, indicative pricing and NVIDIA allocation status, in strict confidence, and shall use it solely to evaluate and perform the Project. Disclosure is limited to directors, employees and professional advisers on a need-to-know basis under equivalent obligations.
The Partner shall not disclose the identity or precise location of the Customer's sites to NVIDIA, OEMs or ODMs without the Customer's prior written consent; other disclosures to NVIDIA shall be limited to what is strictly necessary for allocation, quotation and export compliance.
Standard exceptions apply for information that is or becomes public (other than by breach), was lawfully held without obligation, or is independently developed. Disclosure required by law is permitted with prompt notice where lawful. Confidentiality obligations survive for five (5) years from termination. Neither Party shall issue any press release or use the other's name or marks without prior written consent.
07 — Export Controls and Sanctions
EAR, OFAC, EU 2021/821 — and U.S. ECCNs 3A090 / 4A090 for the NVIDIA stack. Binding
Each Party shall comply with all applicable export control, sanctions and trade compliance laws, including the U.S. Export Administration Regulations (EAR), OFAC sanctions, EU Regulation 2021/821 and applicable EU, German, Polish and Swiss measures.
The Parties acknowledge that NVIDIA accelerated computing products under the Project are classified under U.S. ECCNs including 3A090 and 4A090 and may require U.S. export authorisations. The Partner shall procure the licences and authorisations required to deliver to the Customer's designated sites, and the Customer shall provide reasonable end-use and end-user certifications.
Each Party represents it is not, and is not controlled by, a Restricted Party under any applicable sanctions regime.
08 — Binding and Non-Binding; Definitive Agreements
What is binding now, and what arises only upon Definitive Agreements.
This LOI is non-binding, except for the following Sections, which are binding:
§05
Mutual Project Exclusivity
§06
Confidentiality
§07
Export Controls and Sanctions
§08
This Section (Binding and Non-Binding; Definitive Agreements)
§09
Term and Milestones
§10
Costs
§11
Governing Law and Disputes
§12
General
Any obligations relating to the supply, integration or support of Project equipment and services shall arise only upon execution of duly authorised definitive agreements between the Parties, including one or more Master Supply Agreements, Statements of Work, Purchase Orders and Service / Installation / Support Agreements (the “Definitive Agreements”).
09 — Term and Milestones
60 days from Effective Date, extendable +60. 20/20/20 milestone cadence. Binding
This LOI takes effect on the date of last signature (the “Effective Date”) and expires sixty (60) days thereafter, unless extended by mutual written agreement for up to a further sixty (60) days.
The Parties shall use reasonable best endeavours to achieve the following milestones (all targets, non-binding):
T+20
Technical scoping workshop completed; first NVIDIA allocation status update delivered (per §04).
T+40
Indicative pricing and lead-time framework delivered for the Phase 1 configuration in Annex A; second NVIDIA allocation status update.
T+60
Third NVIDIA allocation status update; target execution of Master Supply Agreement (best endeavours); LOI expiry unless extended for up to a further sixty (60) days.
Either Party may terminate this LOI on written notice; binding provisions survive in accordance with their terms.
10 — Costs
Each Party bears its own. Binding
Each Party bears its own costs and expenses incurred in connection with this LOI.
This LOI is governed by the substantive laws of Switzerland, excluding its conflict of laws rules and the United Nations Convention on Contracts for the International Sale of Goods.
Forum
Swiss Arbitration Centre
Arbitration under the Swiss Rules of International Arbitration of the Swiss Arbitration Centre.
Seat & Language
Zurich · English
Seat: Zurich. Language of proceedings: English.
Arbitrators
One — or three above CHF 10M
One arbitrator; three where the amount in dispute exceeds CHF 10,000,000.
12 — General
Entire agreement, amendment in writing, assignment to Affiliates, electronic signature. Binding
This LOI is the entire agreement of the Parties on its subject matter and supersedes prior discussions. Amendments require written signature by both Parties.
Neither Party may assign this LOI without the other's prior written consent, save that the Customer may assign to an Affiliate on written notice.
Executed in counterparts (including by DocuSign or equivalent electronic signature), each an original, and in the English language.
Annex APhase 1 — Indicative Technical Scope
Annex A
Phase 1 — Indicative Technical Scope.
Indicative and non-binding. All figures preliminary; subject to NVIDIA allocation, technical validation, site selection and the Definitive Agreements.
Capacity & Geography
~160 MW · Poland
IT load capacity approximately 160 MW (gross design intent). Target jurisdiction: Republic of Poland. Indicative deployment target: phased commissioning beginning Q3 2027, subject to allocation and site readiness.
Compute
Vera Rubin NVL72 · GB300 optional
Primary platform: NVIDIA Vera Rubin NVL72 rack-scale systems, configured to NVIDIA Reference Architecture. Secondary / early-deployment platform (optional): NVIDIA Blackwell-generation GB300 NVL72 to enable initial commissioning ahead of Vera Rubin H2 2026 partner availability. Indicative rack count: approximately 500–1,000 NVL72-class racks for Phase 1, subject to final architecture, density and cooling parameters.
Networking & Storage
Quantum-X · Spectrum-X · BlueField
Compute fabric: NVIDIA Quantum-X InfiniBand and/or Spectrum-X Ethernet at scale, per the chosen reference architecture. Storage fabric: NVIDIA BlueField-class DPUs with high-performance parallel file system (reference candidates: DDN, VAST Data, WEKA, or NVIDIA AI Data Platform reference designs). Out-of-band management network and DCIM telemetry included.
Power, Cooling & Mechanical
120–150 kW/rack · Direct-to-chip liquid
Medium-voltage utility supply with on-site primary substation; indicative gross capacity of 160 MW for Phase 1. Direct-to-chip liquid cooling for compute racks, with facility water and rejection systems sized accordingly. Rack power density aligned with Vera Rubin NVL72 specifications (approximately 120–150 kW per rack including overhead). Redundancy topology, ASHRAE W-class, floor loading and mechanical layout to be specified during the technical scoping workshop.
Software
AI Enterprise · Mission Control · Run:ai
NVIDIA AI Enterprise, NVIDIA Mission Control (standard with DGX SuperPOD-class deployments), and workload orchestration (e.g. NVIDIA Run:ai), subject to configuration.
Site Information
Withheld · §06 disclosure
Specific site location, parcel details, permitting status and utility supply contracts are intentionally withheld from this LOI and will be disclosed to the Partner on a need-to-know basis under the confidentiality undertakings in §06.
Annex BSubsequent Phases — Directional
Annex B
Subsequent Phases — Directional.
The Parties anticipate further phases of expansion in the Central and Eastern European region beyond Phase 1. The scale, geography, timing and configuration of such subsequent phases will be developed separately and remain at the Customer's discretion. No information in this Annex B creates any commitment, expectation or right on the part of the Partner, and the express scope of §05 (Mutual Project Exclusivity) is not extended by virtue of this Annex.