Aruba for Xference

Xference is an ambitious startup founded with the goal of solving the paradox of contemporary AI: relying on global, opaque and energy-hungry infrastructures for every inference need, even the simplest and most targeted one. An inefficient approach that leads to an oversized information perimeter, dependence on global connections and privacy risks, in exchange for needlessly excessive costs and consumption. "Our idea, instead, is to provide an infrastructure for the inference of open source LLM models that guarantees complete digital autonomy and maximum privacy", says Andrea Pili, co-founder and CEO of Xference.

"Unlike traditional providers, in fact, we don't train models — we optimize their execution, ensuring that customer data stays confined within its own environment and never needs to be duplicated or transferred externally". The goal is to facilitate AI adoption through no-code solutions that are energy-efficient and compliant with European regulations such as GDPR and the AI Act. By installing open source LLMs on dedicated machines connected via an end-to-end, site-to-site encrypted connection, the inference process remains under the customer's exclusive control, with no external access possible. Having ruled out the cloud due to inadequate levels of transparency and control, Xference also chose to avoid building its own physical sites, opting instead to partner with an external provider able to supply a secure, scalable, high-performance data center suited to the needs of a growing infrastructure. "In particular, we need to guarantee all the power our systems require — we're talking about a need for 24kW per rack, when the typical offering in the sector is around a quarter of that — together with linear scalability that lets us grow from a small startup to a leading European Inference Provider with hundreds of active nodes, without ever running into physical space saturation issues", Pili pointed out.