Andrea Pili has been through the startup cycle before. When his travel site Sardegna.com sold to Alpitour back in 2016, it offered the kind of modest exit that doesn't make headlines but does prove a founder knows how to build something people will pay for. Now he's back with a different kind of infrastructure play—one that might arrive at precisely the right regulatory moment, or possibly just ahead of actual enterprise demand.
Xference, his Cagliari-based venture, has closed a €700,000 pre-seed round led by Niccolò Perra, co-founder of expense management darling Pleo, alongside an investor from FSI. The startup's pitch is straightforward, if not exactly novel: help European enterprises run powerful AI models without sending their data to American hyperscalers or—perhaps more pressingly—running afoul of Brussels' thickening web of compliance requirements.
The company announced the raise in late February and quickly opened a seed round, targeting a close by year-end. Whether that timeline holds depends partly on how quickly customers materialize.
The Sovereignty Angle
Xference doesn't train its own models. Instead, it offers infrastructure for organizations to deploy open-source large language models—Qwen, Llama, Mistral, and others—either on their own premises or within certified European data centers. Think of it as the AWS of on-premises AI, minus the actual AWS part.
The value proposition centers on control. Servers can be air-gapped. Data never leaves a company's domain. Compliance boxes for GDPR and the incoming EU AI Act get checked almost by default. "Not your data processor," the company's materials emphasize, with no-code connectors designed to tap enterprise data sources without the usual hand-wringing over third-party access.
It's a clean pitch for a messy regulatory landscape, though whether mid-sized European firms are ready to pay for that peace of mind—and manage the operational complexity—remains an open question.
Infrastructure First, Customers Second
The funding follows the launch of Xference's distributed infrastructure at Aruba's IT3 data center in Bergamo. Pili told Data Center Dynamics the company is eyeing additional Italian and European sites, though he didn't specify locations or timelines. A beta program for the first 100 invited users is underway, offering an early look at how the platform handles everything from single-node deployments to thousand-server configurations.
The architecture promises low latency and energy efficiency across both traditional CPUs and AI accelerators—table stakes in a market where electricity costs can make or break unit economics.
Pili isn't working alone. His CTO, Michele Fadda, logged over a decade as IT Director at Tiscali, the Sardinian ISP that was once a dot-com darling before settling into regional relevance. Chief Strategy Officer Francesca Audino brings M&A and corporate strategy experience from stints in Milan, London, and New York. The seven-person team officially launched in January 2025 after a year of R&D, registering the company in February.
Crowded Space, Specific Niche
The sovereignty narrative has teeth, at least according to survey data. A Deloitte poll conducted last year found that 77% of European enterprises now weigh a vendor's country of origin when making purchasing decisions. Eighty-three percent view sovereign AI as strategically important. Nearly three in five primarily build with local vendors.
Those numbers suggest fertile ground. They also attract competition.
Germany's Aleph Alpha has raised over €500 million through November 2023 to build sovereign AI from the ground up, training proprietary models and offering full-stack services. IBM launched its Sovereign Core platform earlier this year, leveraging decades of enterprise relationships and regulatory expertise. Xference, by contrast, occupies a narrower wedge: inference infrastructure for open-source models, not model development or comprehensive cloud offerings.
That focus could be an advantage—fewer moving parts, faster time to value—or a limitation if customers want an all-in-one partner. Pili's bet is that enterprises already drowning in vendor relationships will appreciate a specialist.
The Timing Question
Xference arrives as European AI policy solidifies and geopolitical anxieties intensify. The question is whether corporate procurement cycles have caught up to the rhetoric. Regulated industries—finance, healthcare, government—seem like natural early adopters. But convincing a logistics company or a consumer goods manufacturer that they need on-premises AI infrastructure? That might take longer.
For now, the startup has capital, infrastructure, and a beta cohort presumably willing to kick the tires. The seed round will determine whether Pili's second act gets a real audience—or becomes another story about good timing that arrived just a bit too early.
