New supercomputers and AI Factory: Italy invests in Artificial Intelligence.

Last update: July 21, 2026

AI Factory Italy is not "just" a supercomputer: it is an ecosystem that combines GPU computing, data, software, services, and skills to train and deploy Artificial Intelligence models into production. In Italy, public initiatives (IT4LIA, CINECA in Bologna) and private initiatives (Fastweb's NeXXt AI Factory) coexist. The decisive challenge concerns access, security, governance, and operational roadmap.

What is an AI Factory and why supercomputers are central to AI

An AI Factory (an operational ecosystem for creating and deploying AI) is a coordinated set of High-Performance Computing (HPC), data, software tools, services, and specialized support: the supercomputer is the "engine," but it is not enough without processes and people. A supercomputer "for AI" differs from traditional HPC because it is designed for GPUs (parallel accelerators), Large Language Models (LLMs), distributed training, and large-scale inference, featuring high-performance storage and low-latency networking.

Within the European framework, the EuroHPC JU (European High Performance Computing Joint Undertaking) supports infrastructure and services to increase "sovereign" computing and AI capacity. In parallel, the European Commission frames AI as an "ecosystem of trust" oriented toward innovation and rights protection (European approach to artificial intelligence).

16 Member States have submitted over 60 proposals to host new European AI Gigafactories (01net, 2026). In this scenario, Italian investments in AI Factories and supercomputers aim to make training and deployment capabilities available within the country.

What investments Italy is making between IT4LIA, CINECA, and private AI Factories

Italy is investing in AI Factories through a public-private model: on one hand, national infrastructures and programs linked to the EuroHPC JU ecosystem; on the other, industrial initiatives like NeXXt AI Factory (Fastweb's AI supercomputer). The strategic hub is CINECA (interuniversity consortium and HPC center) and the Bologna Technopole (national hub for supercomputing and data), where Leonardo (a EuroHPC supercomputer) also operates.

To understand what an AI Factory operationally "offers," the service portfolio published by IT4LIA AI Factory (national hub with data services, cybersecurity, and compliance) is a useful reference: IT4LIA AI Factory – Portfolio of Services.

How NeXXt AI Factory compares to IT4LIA, Leonardo, and European AI Factories

NeXXt AI Factory, IT4LIA, and Leonardo address different needs: a credible comparison must focus on governance, access models, target users, and strategic roles, rather than "marketing" performance metrics. In Europe, European AI Factories (initiatives connected to EU programs) aim to simplify AI adoption for research, public administration, and SMEs, offering guided pathways and services alongside computing power.

SMEs and universities benefit most from "hub" models that offer structured services and guidance like IT4LIA. Companies with strict integration needs and time-to-market priorities often choose private AI Factories like NeXXt. For a broader view of the national landscape, see the comparison between NeXXt AI Factory and Vitruvian 1, the Italian AI.

What NVIDIA DGX SuperPOD is and what capabilities it enables for industrial AI

NVIDIA DGX SuperPOD (reference architecture for AI data centers) is an integrated platform for training and serving AI models at scale, combining NVIDIA DGX systems (GPU servers), high-performance networking (often InfiniBand, low-latency interconnects), software stacks, and orchestration. In practice, it is designed to reduce the time needed to move from prototype to production while maintaining predictable performance for both training and inference.

AI factories are locally owned, operated and governed AI clouds for training and inference.

— NVIDIA, official company blog

When an Italian AI Factory claims the use of DGX SuperPOD, the key aspect is not the brand name, but the enabled capabilities: distributed workload management, data pipelines, and optimized tools for CUDA (NVIDIA's GPU computing platform). NVIDIA describes AI factories as local AI clouds for training and inference (How Nations Are Deploying AI for Strategic Priorities), a useful framework for evaluating national projects.

• Distributed training: training across multiple GPUs and nodes
• Fine-tuning: adapting LLMs to enterprise data
• Inference serving: low-latency model delivery
• Rapid deployment: optimized, repeatable stack

For organizations handling sensitive data, dataset governance and techniques like synthetic data also make a critical difference: read our deep dive on the importance of synthetic data in AI training according to Nvidia.

What these infrastructures will be used for by Public Admin, startups, research, and enterprises

AI Factories matter because they lower entry barriers to compute power, data management, and model deployment: without GPUs and dedicated services, many organizations remain stuck at the proof-of-concept (POC) stage. Actual usage varies depending on the user type: Public Administration (PA), startups, universities, and regulated enterprises each face different constraints and requirements.

For Public Administration, typical use cases include document intelligence (document classification, data extraction), controlled internal assistants, and multilingual translation. For startups and SMEs, the availability of scalable environments accelerates experimentation and time-to-market: see also tools for AI adoption in Italian SMEs.

In banking (a heavily regulated sector), workloads include risk analytics, fraud monitoring, and internal knowledge base assistants, all subject to strict auditing and data segregation rules. In healthcare, secure environments for clinical data and decision-support models are crucial. Across all cases, "guided" access to services is often more decisive than raw computing power alone.

Security, governance, and roadmap: what a credible AI Factory must prove in Italy

Computing power alone does not automatically make an AI Factory credible: it requires security, governance, compliance, and a verifiable roadmap. A Security Operation Center (SOC) in an AI context must cover identity, networking, GPU workloads, and data pipelines, with explicit responsibilities for incident response and business continuity. Minimum requirements include IAM (Identity and Access Management), project/tenant segregation, logging, and end-to-end observability.

The regulatory landscape is taking shape: the Department for Digital Transformation describes the Italian law as consistent with the AI Act, centered on transparency, security, and coordinated governance (innovazione.gov.it). Additionally, Italian AI Law (132/2025) was officially passed by Parliament on September 17, 2025 (Fondazione Leonardo, 2025).

Two risk signals should not be ignored: only 18% of Italian companies define their AI governance as "industry-leading" (Adnkronos, 2026), and 42% fear reputational damage from data incidents compared to 35% across the EU (Adnkronos, 2026). Related deep dives: managing AI governance and EU AI Act compliance in enterprises and privacy and security in local LLMs: the DS4 revolution.

The development of a trustworthy AI ecosystem is essential to foster trust in AI and provide legal certainty so that businesses can innovate.

— European Commission, EU policy page

FAQ on AI Factories, Italian supercomputers, and European competitiveness

Who can access IT4LIA AI Factory and its associated services?

IT4LIA AI Factory is designed to serve research institutes, Public Administrations, SMEs, startups, and enterprises through a portfolio of services and expert support. Access is typically provided through structured programs and service requests (rather than raw "compute hours"), with onboarding procedures and compliance/security criteria defined by the program.

Is an AI Factory the same as a traditional data center?

No, an AI Factory is not a traditional data center because it integrates GPU compute, tools for training and inference, data management pipelines, operational security, and expert skills required to deploy models into production. A traditional data center can host AI hardware, but without integrated services and governance, it risks remaining an "empty" infrastructure.

When will Italy achieve European-scale AI capacity beyond current active systems?

Italy is building capacity across multiple initiatives, some already operational and others expanding. Across Europe, demand is growing rapidly: 16 Member States have submitted over 60 proposals for new AI Gigafactories (01net, 2026). Scalability will depend on roadmaps, access models, and service execution, not just on acquiring new hardware.

Is NeXXt AI Factory better suited for enterprises than public initiatives like IT4LIA?

NeXXt AI Factory tends to be a more natural fit for enterprise projects and industrial partnerships where direct integration, strict SLAs, and fast time-to-market are paramount. IT4LIA is structured for broader access (research, Public Admin, SMEs) with standardized service models. The right choice depends on data sensitivities, security requirements, and procurement preferences.

Why are security and SOC capabilities critical in an AI Factory?

Security and SOC operational capabilities are critical because AI workloads handle sensitive data and business-critical models. An effective SOC monitors access, network traffic, and data pipelines, handles incidents, and ensures complete auditability. Without robust IAM, logging, and segregation, even a powerful supercomputer becomes an operational and reputational risk.

For a global perspective on the strategic role of supercomputers, useful when comparing EU and non-EU approaches, read our analysis on advanced supercomputers like Japan's project to build the world's most powerful machine. For European regulatory context, see also the European AI Pact signed by major tech firms.