NOVA builds and operates modular GPU data centers across the Gulf, South Asia and Europe, and runs the full stack on top: bare metal, GPU cloud and managed inference. Reserved capacity on long-term contracts; from ready site to live racks in six months.
five-year target for our global AI infrastructure portfolio
contracted energized sites across 6 countries
GPUs on a single fabric, under long-term contracts
uptime SLA on every cluster, backed by service credits
NOVA is an AI compute infrastructure operator: we build and run AI data centers and the full stack on top, spanning colocation, bare-metal GPU clusters, GPU-as-a-Service and managed inference, for enterprise, sovereign and institutional customers.
We take sites where power already exists, deploy prefabricated GPU modules, and operate them as production infrastructure. Customers buy compute on long-term contracts. The complexity stays on our side: construction, hardware integration, cooling, uptime.
The financing arrives with us: NOVA structures the capital for each build, procures GPUs and equipment at industrial volume, and delivers the project turnkey: siting, construction, integration, licensing, operations. You contract the capacity; everything between the power line and your workload is on us.
Dedicated GPU clusters on take-or-pay contracts, 3 to 7 years. Single-fabric clusters from 1,000 to 10,000+ GPUs, NVIDIA Blackwell and Rubin generations, liquid-cooled. A predictable price per GPU-hour, a delivery date committed in the contract, and capacity that grows in 3 MW steps.
Have your own GPU allocation, or need capacity in a specific region? We deliver powered, operated halls for your hardware: on our modular campuses or qualified partner facilities, with our 24/7 team and your security perimeter.
A managed layer on top of our bare metal: virtualization, orchestration and inference endpoints, white-label and tenant-isolated, with true pay-as-you-go billing and economics transparent down to the unit of inference.
GPU-hours are your cost of goods. We give you dedicated, contiguous clusters: thousands of GPUs on one fabric, at a predictable price per GPU-hour, on contracts shaped to your economics. Capacity sits close to fast-growing markets in the Gulf and South Asia, with European sites in development for workloads that must stay in the EU.
Your allocation has a home. If you hold GPU quotas without powered sites, workloads that need their own environment, or a national AI program that must run in your jurisdiction, we provide the missing half: powered land, prefab modules, grid contracts, licensed entities, and an operating team that runs it 24/7.
NS-1 is the 1 MW liquid-cooled unit the whole portfolio is built from. Power distribution, switching, networking and cooling arrive factory-integrated, pre-cabled and pre-tested, so on-site work is limited to foundations, utility tie-in, assembly and commissioning: that is why the schedule holds.
Three NS-1 containers combine into one NS-3 module. The repeatable 3 MW unit is built around NVIDIA Blackwell and Rubin generation systems, following NVIDIA reference architectures (DGX SuperPOD class).
Four NS-3 modules combine into an NS-12 block. Capacity then grows block by block toward a 100+ MW campus: the same design, the same BOM and the same contract stack at every scale.
Factory-integrated, liquid-cooled and tested before it reaches the site.
Scalable campus capacity, deployed in phases
Modular deployment from a ready site
Planned NVIDIA platform for rack-scale deployments
of on-site module installation
Integration, cabling and burn-in happen on a production line. Every module leaves the factory tested.
Modules are manufactured while the site is being prepared, and the two streams meet at assembly: that is where the six months come from. Switchgear alone carries a year of lead time industry-wide; ours ships inside the module, ordered with it.
See the detailed deployment roadmapCapacity lands in 3 MW steps. Start with one module and add capacity module by module: contract each phase as your demand grows.
Securing powered land is the hardest problem in AI infrastructure, so that is where we start: operating energy campuses, industrial sites with live grid connections, partner facilities with room to grow. Capacity is serving load today, construction is underway, and the pipeline is moving to contract through 2028.
| Region | Site type | Phase 1 | Full build-out | Status | RFS |
|---|---|---|---|---|---|
| India | Colocation campus, partner-operated | 20 MW | 100 MW | Operating | Live |
| OmanSite I | Anchor energy campus | 30 MW | 400 MW | Contracting | Q2 2027 |
| OmanSite II | Free-zone campus | 12 MW | 50 MW | Contracting | Q1 2028 |
| Romania | Industrial site, with site partner | 24 MW | 120 MW | Contracting | Q4 2027 |
| Azerbaijan | Grid-secured industrial site | 30 MW | 150 MW | Contracting | Q4 2027 |
| Greece | Industrial site, with site partner | 24 MW | 150 MW | Sold-out | Q2 2027 |
| Bulgaria | Industrial site, with site partner | 12 MW | 100 MW | Contracting | Q3 2027 |
Operating: live and serving load · Under construction: build in progress · Contracting: capacity on this site is being contracted to customers now. Site types: Energy campus: operating power infrastructure on the site · Industrial site: grid-connected industrial land · Colocation campus: partner-operated facility we deploy into.
Every NOVA site is built to one standard: 2N power and N+1 cooling, design PUE below 1.3, and diverse fiber routes with at least two independent carriers.
For every site we operate, we will publish our commitments to the people who live next to it: noise measured at the nearest residence, water use in plain numbers, local jobs, and what the grid gains. Site by site, as each project becomes public.
For twenty years compute was abundant and demand was the question. AI inverted it: chips get allocated in a spreadsheet; powered sites have to be built. Whoever holds powered, operated capacity holds the market.
Cluster prices keep rising and delivery windows keep stretching. Reserving early gets you two things nobody can sell you later: a delivery slot and a defined price. And it keeps the slow, irreversible part of the project moving while everything reversible stays open.
Sign this quarter, and your modules enter the production queue now. Wait six months, and the queue moves ahead of you: RFS slips a quarter. Wait a year, and you are behind the next buyer, at next year's prices. An early signature costs nothing extra; waiting is priced by the queue.
Our team has spent years running power-hungry, high-density compute: substations, transformers, liquid cooling and round-the-clock operations at megawatt scale. That muscle is what AI data centers are built on.
Every cluster earns its way into production: firmware and fabric validation, stress testing under full load, performance benchmarks against the contracted spec. Clusters are designed to NVIDIA reference architectures and validated against them at acceptance. The protocol is published, your engineers can run every test themselves, and the definitions are strict: a node with one unhealthy GPU is a failed node. Billing starts when the cluster passes acceptance. The same gate re-runs whenever a node returns to the pool.
Continuous passive and active health checks across GPUs, fabric, storage, power and cooling. Automated fault isolation and remediation, node swaps from a dedicated spare pool, and a defined repair path from reboot to RMA. Contracted capacity comes back: node failure here is a rehearsed procedure.
Nothing we monitor is hidden from you: full-stack telemetry and audit logs, delivered to your dashboards and your SIEM. Support is a direct channel to the engineers who run the site, under a severity matrix committed in the contract: Sev-1 answered in minutes, 24/7, with response times written in as numbers. Every reserved-capacity customer gets a named engineer and regular capacity reviews.
Dedicated clusters and networks with strict segmentation: other workloads never touch your perimeter. BYOK, KMS and HSM support, role-based access, enhanced administrator auditing. Data residency and use are contractual commitments: your data is not used to train third-party models and stays within the jurisdiction defined in your agreement. And the structure behind it is as clean as the racks: real, disclosed entities built to be audited, structured for vendor KYC and banking compliance from day one. To be, not to seem.
NDA, then a workload and eligibility review, then technical design and indicative terms, then a reservation and contract. From first conversation to a signed reservation: a matter of weeks. Want to start smaller: the NOVA Pilot runs your workload on a dedicated cluster while the contract is prepared in parallel, with the cost credited into your reservation.
When your cluster passes acceptance testing, and no earlier.
Pricing is contract-specific: a predictable per-GPU-hour formula on take-or-pay terms, falling with term length and volume. Tell us your workload and scale, and you will get indicative terms at the design stage.
An evidence pack under NDA: site control and utility documentation, the bill of materials and factory slots, the construction schedule, the financing structure and the acceptance criteria. Full electrical schedules and thermal parameters live in the engineering package, shared under NDA.
The delivery date is a contractual commitment with remedies written into the agreement. The clock, its start, its inclusions and its exclusions are all defined in the contract.
The GPU platform and its refresh path are fixed in the contract. The modular architecture is built for generational change: liquid cooling and power headroom sized for what comes next, and refresh options priced at the renewal window. You contract compute; keeping the platform current is our problem.
Bare metal by default: dedicated clusters with no virtualization overhead. Managed Kubernetes and Slurm are optional layers on top.
Ours by default: reserved capacity runs on NOVA-owned clusters. If you bring your own GPU allocation, we deliver powered, operated halls for it. Both run under the same 24/7 operations and SLA.
Non-blocking, rail-optimized fabrics built to NVIDIA reference architectures, with full east-west bandwidth per node committed in the contract spec. From 1,000 to 10,000+ GPUs on a single fabric, growing in 3 MW increments.
Local NVMe per node plus a shared parallel filesystem tier, sized to your workload and fixed in the design.
Dedicated clusters and networks with L2/L3 segmentation and fabric partitioning: other workloads never touch your perimeter. Your data lives in the jurisdiction defined in your agreement, remains fully yours, and is never used to train third-party models.
Every project is structured for vendor KYC and banking compliance in Western jurisdictions. Eligibility is confirmed during the review stage, before terms.
India and Oman first, with a pipeline across Europe and the Caspian: see the portfolio above. Exact sites, specs and site visits are part of diligence, under NDA.
Reserving capacity: tell us your workload, scale (GPUs or MW), target region and timeline.
The pilot exists so you do not have to believe anything we say: run your production workload on a dedicated cluster for a few weeks, and let the measurements decide.
The pilot runs on capacity that already exists, under a short agreement you can sign in days. The reservation is prepared in parallel: while your workload runs, the MSA and diligence move, with your price, delivery slot and acceptance protocol fixed at pilot start and held while you decide. When the report lands, the decision is one signature: sign the reservation and the pilot fee credits into it, or walk away and keep the report.
What it is: from 128 GPUs on a single fabric, dedicated to you, running on our live campus today: under the same SLA, severity matrix and data terms as a full reservation. Larger blocks and longer runs are sized by agreement.
| Pilot | Reserved capacity | BYO hardware | GPU cloud | |
|---|---|---|---|---|
| What it is | A paid pilot on a live campus | Dedicated clusters, take-or-pay | Powered, operated halls for your GPUs | Managed layer on our bare metal |
| Commitment | 2-4 weeks, short pilot agreement | 3, 5 or 7 years | Multi-year | Early access, pay-as-you-go |
| Scale | From 128 GPUs | 10-30 MW typical, 3 MW steps | Per hall | Per workload |
| Who owns the hardware | NOVA | NOVA | You | NOVA |
| When billing starts | At acceptance | At acceptance | At hall handover | At first token |
| Price shape | Fixed pilot fee, credited on conversion | Per-GPU-hour formula, falls with term and volume | Per-kW and services | Per use |
| Expansion | Credits into a reservation; bridge until RFS | Slots reserved at signing | By hall | Elastic |
Every option runs on the same operations: acceptance gate, severity matrix, full-stack telemetry, contractual data residency.
One NS-3 module (3 MW) is the floor; most reserved contracts start at 10 to 30 MW on 3, 5 or 7 year take-or-pay terms, with rates falling as term and volume grow. Shorter runs start with the Pilot.
A prepayment at signing, credited against invoices over the term; milestone payments tied to real events (site, power, modules, fabric, acceptance); then per-GPU-hour billing from acceptance onward. No setup fees, no ingress or egress fees, no support surcharges.
Price and delivery slot can be frozen for a defined window while you decide: automatically during a pilot, as a paid hold otherwise. The hold has two parts: a reservation fee that pays for taking the capacity off the market, and a prepayment that converts into compute in full. Extending the window is a priced line item, and hold fees count toward the contract when you sign.
Phased: first modules go live at RFS, then capacity grows module by module on a schedule fixed in the contract. RFS per site is in the portfolio. The delivery date carries remedies, and a missed milestone is your clause to invoke.
Capacity expands in 3 MW steps under the same contract, into slots reserved for you at signing. A pre-priced burst reserve covers demand spikes: activated on days' notice, at a rate agreed before you needed it.
Flexibility is written in as terms, priced at signing: step-down checkpoints at fixed dates, SKU flexibility inside a GPU generation, and a technology-refresh path at renewal. A strategy change is handled by a clause you already hold.
A renewal window with right of first refusal on your capacity: extension and expansion are agreed before expiry.
We handle the site, power, construction, hardware, firmware, fabric, cooling, 24/7 monitoring and vendor warranty. You run your workloads and your software stack. Managed Kubernetes and Slurm are available on top.
NOVA funds each build through a dedicated project company. A signed take-or-pay contract anchors the structure: equity and project debt are sized against contracted revenue, following the discipline that finances power plants and fiber networks. Capital is drawn against verified milestones, from site control through module production to acceptance. Buyers sign for compute and keep construction off their balance sheet. Site owners bring powered land and leave with a financeable project.
The customer contract, the site rights, the modules and the debt live in one dedicated project company (SPV) per site. The offtake anchors it, sponsor equity and project debt fund it, and the module supply chain and operations sit under contracts a lender can read.
Reservation, then diligence on the evidence pack, then financial close, and the contractual delivery clock starts: notice to proceed, module production, energization, acceptance. Billing starts at acceptance. Each step is a defined event, with obligations on both sides.
Capital is released step by step: site control, grid connection, module production, energization, acceptance. Prepayments, where agreed, are credited against contract years and released against build milestones.
Lenders fund contracted revenue from creditworthy counterparties, delivery obligations they can verify, and structures they can audit. Our contracts, sites and entities are assembled against those requirements from day one. That is also why the evidence pack exists.
Who this is for: offtakers who want capacity without construction on their balance sheet, and site or program owners who need a build financed.
What you leave with: an indicative structure for your project: sources and uses, the milestone schedule, and the delivery clock it produces.
Per-project terms are set in documents. Bring your project, and we will structure it with you.
Bring a projectWe build AI compute infrastructure from the ground up: module by module, campus by campus, delivered in months and run around the clock, across the Gulf, South Asia and Europe. Engineers and operators first.
Own things end to end. Scope here grows faster than titles.
Challenge every assumption. No egos: the best idea wins.
Megawatts forgive nothing. Preparation, forecasting and mastery of your tools are the job itself.
Delivery in months, not years, is a promise our customers build on. We keep it.
Our favorite kind of engineer is equally at home in a switchyard and in a pull request.
Competitive pay with equity and transparent salary ranges in every posting; healthcare for you and your family; relocation support: visa, family, temporary housing; paid professional certifications: data center, electrical safety, project management; benefits in line with local norms in each region.
Write to
career@novascaler.aiThe site, module, GPU supply, commissioning and cluster do not move one after another. They advance in parallel and converge at customer acceptance. Every gate closes on evidence; every material delay activates a named contingency path; billing begins only after customer SAT.
Before the contracted date is fixed, NOVA verifies the power route, site scope, jurisdiction, workload design and hardware supply route. Factory work and GPU procurement may begin earlier; the final schedule is issued when the workstreams can be joined into one executable plan.
Planning model, not a promise: the dates are recalculated for the contracted site, jurisdiction, workload and hardware allocation.
Each gate names the condition, the evidence that closes it and the path used when it does not close on the first attempt.
Re-sequence the site or activate the contracted alternative supply architecture before module design is released.
Move to the qualified alternate supplier or delivery channel before customer capital is committed downstream.
Corrections remain at the factory. Nothing ships until the gate closes.
Activate the weather or civil-work contingency and, where required, use responsible storage without exposing the module.
Close authority or commissioning findings, repeat only the affected tests and re-submit the evidence pack.
NOVA closes the SAT punch list. Billing does not begin until the contracted acceptance criteria pass.
Share your workload, target scale, preferred region and required live date. We’ll turn them into a project-specific site, module and GPU delivery plan.
NovaScaler public legal and service documentation.
NovaScaler · novascaler.ai/legal/privacy Version: 1.1 · 2026-09-07
This website, novascaler.ai, is operated by Maximalist Ventures L.L.C-FZ, trading as NovaScaler (Meydan Free Zone, Dubai, UAE, registration No. 2644784) ("NovaScaler", "we"). NovaScaler is the controller of the personal data described in this notice. Full company details: see the Legal Entity and Imprint page.
Contact for privacy matters: compute@novascaler.ai, subject "Privacy".
Contact and capacity inquiries. When you submit the contact form or write to us, we collect: name, company, work email, what you need (reserved capacity, powered space, cloud early access, capacity pack, other), workload type, GPU type or allocation, scale (GPUs or MW), target region, target RFS, contract term, and your message.
Presentation access requests. When you request the NovaScaler presentation, we collect your name, work email, how you heard about us, the referring page and campaign parameters. We also record the versions of the presentation, Terms of Use and Privacy Notice shown when access was granted. We do not collect your IP address or browser fingerprint for this download.
Eligibility and diligence. If your inquiry moves toward a contract, we collect the information needed for counterparty due diligence: the identity of your organization, the names and roles of its representatives, and end-use information. This can include screening against sanctions, export-control and watch lists, performed by us or by specialized screening providers.
Job applications. If you apply to work with us: name, contact details, CV, and anything you choose to include. See also our Recruitment Fraud Notice.
Technical data. Standard server logs: IP address, browser type, pages requested, timestamps. Analytics as described in the Cookie and Analytics Notice.
Correspondence. Emails you exchange with us, including under NDA processes.
We do not collect payment data through this site, and we ask you not to send credentials, secrets or sensitive personal data through the form.
| Purpose | Data | Legal basis |
|---|---|---|
| Answering inquiries, preparing indicative terms | Inquiry data, correspondence | Steps toward a contract; legitimate interest |
| Providing requested presentations and following up about the request | Presentation access and attribution data | Your request and acknowledgement; legitimate interest in responding to business inquiries |
| Eligibility review, KYC, sanctions and export-control screening | Diligence data | Legal obligations; legitimate interest in lawful trade |
| Recruitment | Application data | Steps toward an employment contract |
| Site security and operation | Technical data | Legitimate interest |
| Understanding site usage | Analytics data | See Cookie and Analytics Notice |
We do not sell personal data. We do not use your data for third-party advertising.
Service providers who run our infrastructure for us: hosting, email, CRM, and screening providers, each bound by contract. Professional advisers: lawyers, auditors, insurers. Entities in the NovaScaler group, including project companies, where an inquiry concerns their site. Authorities, where the law requires it. Counterparties in a transaction see only what diligence requires.
The current list of processors is available on request, and will be published as the Sub-processors List.
Our systems are hosted in the European Union and the United Arab Emirates. Where data moves across borders, we use recognized safeguards: contractual clauses and, where available, adequacy decisions.
Depending on where you are, you have the right to access your data, correct it, delete it, restrict or object to its use, and receive a copy in a portable format. EU and UK visitors have these rights under GDPR; UAE residents under the UAE Personal Data Protection Law. To exercise them, write to compute@novascaler.ai with "Privacy" in the subject. We answer within 30 days. You can also complain to your data protection authority.
We do not make automated decisions with legal effect about you. Screening results are always reviewed by a person before any decision.
We version this notice. Each published version carries an effective date, and prior versions remain available in an archive. Material changes are flagged on this page.
NovaScaler public legal and service documentation.
NovaScaler · novascaler.ai/legal/terms Version: 1.1 · 2026-09-07
This website, novascaler.ai, including status.novascaler.ai and other subdomains that link to these terms (together, the "Site"), is operated by Maximalist Ventures L.L.C-FZ, trading as NovaScaler (Meydan Free Zone, Dubai, UAE, registration No. 2644784) ("NovaScaler", "we"). Full company details: see the Legal Entity and Imprint page. By using the Site you accept these terms. If you do not accept them, do not use the Site.
These terms govern the Site only. Compute capacity, colocation, cloud services and pilots are governed by separately signed agreements, not by this page.
The Site describes what we build and how we work. It is information, not a contract:
The Site and its content: text, graphics, diagrams, layouts, and the NovaScaler name and logo, belong to NovaScaler or its licensors. You may view the Site and print or save extracts for your own evaluation of our services. You may not republish, sell or modify Site content, or use our name or logo, without our written permission. Third-party marks, including NVIDIA and Lenovo, belong to their owners; their appearance here does not imply their endorsement of this Site.
A presentation made available through a gated download is NovaScaler confidential information and is provided solely for internal evaluation. You may share it only within your organization and with professional advisers who are bound to protect confidential information. You must use reasonable care to protect it and may not publish it, post it online, or distribute it outside that evaluation group without our written permission. If a separate non-disclosure agreement applies, the stricter obligation controls.
Do not: probe or disrupt the Site outside our Responsible Disclosure Policy; misrepresent your identity or affiliation when contacting us; use the Site to send unlawful, deceptive or malicious content; scrape the Site at volumes that degrade it. Automated access for indexing and AI-assisted reading of public pages is fine.
The Site links to third-party resources, including the status page infrastructure and social profiles. We do not control them and are not responsible for their content or practices.
The Site is provided as is, without warranties of any kind, to the extent the law allows. We do not warrant that the Site is uninterrupted, error-free or complete. To the extent the law allows, NovaScaler is not liable for damages arising from use of, or reliance on, the Site. Nothing in these terms limits liability that cannot be limited by law.
How we handle personal data collected through the Site is described in the Privacy Notice and the Cookie and Analytics Notice.
We can update the Site and these terms at any time. Each published version of these terms carries an effective date; prior versions remain available in an archive. Continued use after a change means acceptance of the current version.
These terms are governed by the laws of the Dubai International Financial Centre (DIFC), and disputes about the Site belong to the courts of the DIFC. This clause covers the Site only; signed commercial agreements carry their own law and forum.
compute@novascaler.ai for capacity and partnerships. security@novascaler.ai for vulnerabilities. kyc@novascaler.ai for compliance and diligence. compute@novascaler.ai, subject "Privacy", for personal data.