Xference, the Italian inference provider specializing in private AI infrastructure, raises €700,000 in a Preseed round and announces the opening of its Seed round to institutional investors. One year after its founding, the company is building its distributed infrastructure across Europe starting from one of Italy's most important data centers and is preparing to open its invite-only Beta service to the first 100 users, who will become the early adopters of technology developed in Italy.
Co-founded in 2025 in Cagliari by Andrea Pili (CEO), Michele Fadda (CTO) and Francesca Audino (CSO), Xference was built with a clear positioning: to offer a secure, high-performance, flexible and reliable private AI inference infrastructure, built and operated entirely within certified national data centers and also available in an on-premise version. The goal is to allow anyone to harness the power of generative AI while guaranteeing full data control, digital sovereignty and the highest privacy standards, optimized for reduced energy consumption.
"This is a response to a paradox of contemporary AI", says Andrea Pili, Founder & CEO. "To achieve targeted inference on internal data such as documents, images and databases, companies today rely on global infrastructures that are oversized relative to their actual needs, with significant energy and cost expenditure. But contextual inference does not require consulting such a vast environment. Inference is a mathematical computation on an already-trained model. If the information perimeter is internal to the company, it makes sense for the processing to stay there too. Xference was created to bring AI back within the customer's domain, with no compromises on performance or ease of use".
Xference does not train proprietary models: it optimizes and governs the operation of open-source Large Language Models, including Qwen, Llama and Mistral, installed and "frozen" locally on dedicated servers. The models do not connect to the Internet, do not send data to remote servers, and operate in an isolated manner.
The solution pairs an intelligent reasoning engine with no-code tools and plug-and-play API connectors that allow multiple data sources to be integrated into a single interface. Users can upload documents, connect corporate drives and query their information bases via prompts, without any technical expertise.
The architecture is designed to ensure linear scalability from a single node to thousands, minimal latency, optimized energy consumption and compatibility with various hardware infrastructures, from standard processors to AI accelerators.
"We have built a complete infrastructure, orchestrating hardware and software, making our technology independent from any single vendor", explains Michele Fadda, Founder & CTO. "The goal is to make private inference measurable, efficient and under the full control of those who own the data, reducing dependence on global infrastructures".
As a European company, Xference's infrastructure is natively designed in compliance with the GDPR and the AI Act. Privacy is not an afterthought but an architectural principle: data is neither transferred nor stored on external clouds, and processing takes place exclusively within the user's secure environment.
In a landscape dominated by major international providers, Xference positions itself as a European alternative built on a different infrastructure: less hyper-centralization, lower infrastructure costs, greater energy efficiency and full data sovereignty.
"Privacy is the most complex challenge of contemporary AI", adds Pili. "We believe that Europe can and must build its own AI infrastructure: more secure, more controlled, more aware. It is not just a technological choice but a cultural and strategic one. Privacy is non-negotiable".
The team, led by its founders, brings together expertise in Internet infrastructure, data centers, cloud, open source and machine learning, with the ambition of building the fastest, most secure and easiest-to-deploy private inference platform on the European market.
Andrea Pili, serial entrepreneur with over 25 years of experience, has founded seven startups and achieved a notable exit with Sardegna.com, Italy's first online booking site, acquired by Alpitour in 2016. Following experience in Silicon Valley and in the fields of data centers, GPU and Big Data, he has led the growth of highly scalable digital companies. Michele Fadda, senior technical manager with over 30 years of experience in IT infrastructure and Internet services, has held leadership roles at Video On Line and Tiscali, where he served as IT Director for over 10 years. An expert in cloud architectures, distributed data centers and open-source solutions, he has led teams managing international mission-critical infrastructures. Francesca Audino, aerospace engineer with a strong background in Business Administration, Francesca boasts over 20 years of international career spanning Milan, London and New York. She has held senior roles as Chief Strategy Officer in leading global groups, where she managed complex M&A operations and strategic planning for major Private Equity funds, including Carlyle and InvestIndustrial.
The PreSeed round involved strategic partners and high-profile angel investors who contributed not only capital but also expertise and international networks, including Bastiano Sanna, an investor and entrepreneur with international experience, and Niccolò Perra, co-founder of Pleo, one of the few European unicorn startups in the Fintech sector, now counting nearly 1,000 employees and operating in over 20 countries.
With the infrastructure now in production and the upcoming opening of the Beta to the first 100 invited users, Xference is entering its operational phase. The stated goal is to close a Seed round by the end of 2026 to accelerate technological development, expand the team and consolidate its position as the European benchmark for private AI inference.
"Inference will become the primary mode of interaction between people and technology", concludes Pili. "We want to build the infrastructure that will make this transition possible, in full transparency, with no compromises on privacy, performance and control".