At AI Week

Aruba presented the Aruba AI Platform, a fully European full-stack platform for developing, orchestrating, and deploying artificial intelligence solutions. The platform is explicitly open, multi-model, and lock-in free: the company does not propose a single model but rather the capability to combine models, data, and services while maintaining control of the architecture over time. This choice is based on a precise understanding of the state of AI in businesses.

In an interview with Edge9, Luca Luperini, Head of Digital & AI at Aruba, explained the logic behind the project, starting from a data point that downscales much of the public narrative surrounding AI. As Luperini recalls, Aruba is a heterogeneous technology company that ranges from certified email services to data centers, and it has approached artificial intelligence by first asking how to govern it to build a long-term investment consciously.

Why AI Stops Before Production

The starting point is the gap between market enthusiasm and the operational reality of businesses. After years of proof of concepts, pilot projects, and limited adoptions, the processes truly entrusted to AI remain few.

"The companies that today have core processes and services with AI on board are still very few compared to how many discuss it. There is an asynchrony between the noise around investments and what we actually see at the level of processes and tools," observes Luperini. It is an unusual slowdown: typically, when a new technology arrives, the transition from the lab to production has reasonable timing, while here the leap is stuck.

According to Luperini, the slowdown stems from questions more mature than those regarding the simple choice of model: how governable is a solution, what is the return on investment, how much does one bind to third parties by relinquishing data and know-how that may not entirely belong to them anymore. These are questions related to data governance and regulatory compliance, from GDPR to NIS2 and the AI Act, which signal a growth in awareness rather than resistance. It is in this space, made of infrastructure, data control, and economic sustainability, that Aruba has decided to position its offering.

A Composable Platform, from Data Center to Service

The technical response is a platform built to be composable at every layer.

"We start from the building block of the data center, from the GPU we place on top, from the open-source model and the service built on that model. Everything is governable, and that's why we talk about a full-stack platform," explains Luperini. The goal is for the customer to drive the investment, starting from hardware and use cases rather than from flat-rate plans imposed on business processes.

At the model level, the platform adopts a multi-model and agnostic approach, with an automatic routing that directs each request to the most suitable model.

"Not all AIs are the same, and not all are suitable for everything. If I need to do two plus two, it’s pointless to bother the biggest model: automatic routing takes each use case to the right model, and so the cost also becomes much more understandable and scalable," summarizes Luperini. For simple tasks, small models are sufficient, while advanced reasoning or code generation justifies larger and costlier models, with the advantage of knowing the cost for each call in advance.

The offering is structured across multiple consumption levels. The first is the model delivered as a service, where an open model is called in a private version, such as Mistral or DeepSeek, paying transparently based on the number of calls instead of a flat fee. The second level consists of services already optimized for specific functions, such as text extraction from documents via Optical Character Recognition, where the customer pays for a single call to a service already fine-tuned. The third is customized vertical design, for example, an internal skills analysis based on sensitive data that the company does not want to entrust to a public AI. On this basis, Aruba will progressively add new application services over the year.

Even the degree of isolation is modulable. It starts from a Private AI shared but localized in Italy, moving to a dedicated AI with reserved hardware for the single customer, up to colocation, where Aruba provides space in data centers, and the customer hosts their machines. The installation at the customer’s site is excluded: this is a policy the company already applies to the cloud and extends to AI, as only in their structures can they guarantee the expected levels of security, certification, and support. Everything relies on proprietary data centers in Italy and Europe, with state-of-the-art NVIDIA DGX B300 systems, ISO/IEC 27001 certifications, CISPE Code of Conduct, compliance designed for GDPR, DORA, NIS2, and AI Act, and energy that comes 100% from renewable sources, partially self-produced from hydroelectric plants.

The thread that holds this architecture together is economic as well as technical.

"The parallel with the cloud makes sense: by relying processes and services on public models, one risks losing control and lacks awareness of medium and long-term sustainability," warns Luperini, who cites the case of those who exhausted their annual AI budget in a single quarter. As already happened with the cloud, where it became clear that not all workloads are suitable for consumption, cost predictability becomes a criterion for choice. Opening the box of the platform and seeing how it is made helps to make the cost of AI readable, avoiding flat-rate plans tied to the number of tokens that obscure sustainability.

It is here that Aruba’s positioning compared to large international providers is measured, built not on the single model but on the control of the entire supply chain and on data sovereignty, with proprietary assets from Italy and Europe. The platform was publicly announced at AI Week and already has the service model for enterprise customers active, with the launch on online channels expected by summer and a roadmap that will continue throughout the year. For businesses considering how to incorporate AI into processes without sacrificing control and compliance, Aruba's bet is that the competitive advantage first lies in the infrastructural foundations and only later in the models.