Xference, a sovereign AI inference provider built around a private data-center infrastructure in Europe, and Axelera AI, the European AI semiconductor company developing next-generation AI Inference acceleration hardware, today announced a collaboration aimed at advancing a fully European AI infrastructure, spanning from AI chips to inference orchestration.
As part of the agreement, Xference will become one of the first companies to integrate Axelera AI’s Europa next generation chipset. The shared goal is to build a European ecosystem for developing technologies that are independent of major non-European providers, with a strong focus on data sovereignty, energy efficiency and environmental sustainability.
Under the partnership, Xference will join Axelera AI’s ecosystem program designed to bring full solutions to customers globally.
Europa is Axelera AI’s next-generation architecture designed to extend AI acceleration from edge devices to enterprise servers and data centers, enabling demanding generative AI workloads while maintaining significantly lower power consumption than conventional GPU-based solutions.
“Sovereignty and privacy have to be talked about together - Fabrizio Del Maffeo, CEO & Co-founder Axelera AI, commented - It isn’t enough to control where your data lives if you can’t also control who has access to it and what it costs you to run it. Xference is exactly the kind of partner we want: a company building sovereign AI infrastructure from the ground up, that will ensure customers get the absolute best performance and control out of their AI infrastructure.”
Axelera AI is a hardware and software company purpose-built for AI inference acceleration, providing more than 600 customers worldwide with products built on its proprietary D-IMC technology and RISC-V dataflow architecture. Its upcoming Europa platform extends the company’s chip lineup from ultra-low-power modules for embedded and edge use to server-grade, multi-chip cards designed for enterprise and data-centre deployment, unified under Axelera’s Voyager software stack. Europa is designed to deliver substantially lower power consumption than comparable GPU-based inference cards.
“We built Xference to give organisations real control over how and where their AI runs, not just their data, but their costs and their infrastructure choices too - Andrea Pili, Founder, Xference, added - Partnering with Axelera lets us put a fully European stack, from chip to inference orchestration, to a real test: can it compete on performance while keeping that control where it belongs, with our customers, rather than with a handful of non-European suppliers.”
Xference operates a sovereign AI inference infrastructure designed to let organisations run private, on-premise AI models without depending on non-European hyperscale cloud providers. Built on a hardware-agnostic architecture, its orchestration engine dynamically selects the right model for each request to optimise performance, latency and energy consumption, a design particularly suited to regulated sectors such as healthcare, defence and security, where data sovereignty is a requirement rather than a preference.
Xference currently runs production workloads on its Italian data centre, with plans to expand across Europe and, longer term, into markets such as Brazil and Canada, where similar sovereignty concerns are gaining traction.
AI Private Inference Technology made in Europe
