TLDR: Axelera AI has announced Europa, a groundbreaking AI processor unit (AIPU) designed to redefine performance, power efficiency, and affordability for multi-user generative AI and computer vision applications. The Europa AIPU delivers up to 629 TOPS at INT8 precision and boasts 3 to 5 times greater performance efficiency compared to leading industry solutions, with shipments expected to begin in the first half of 2026.
EINDHOVEN, Netherlands – Axelera AI, a prominent innovator in purpose-built AI hardware acceleration technology, today announced the launch of its revolutionary Europa™ AI Processor Unit (AIPU). This new offering is poised to establish a new industry benchmark for performance-to-price ratio in multi-user generative AI and computer vision applications, catering to a broad spectrum of uses from edge devices to enterprise servers.
The Europa AIPU is engineered for exceptional processing power and remarkable energy and thermal efficiency, all within a compact design that supports multiple form factor options. This makes it an ideal solution for demanding, multi-modal AI inference workloads.
At the heart of Europa’s capabilities are its eight second-generation AI cores, each integrating Axelera’s advanced Digital In-Memory Compute (D-IMC) technology and large vector engines. This sophisticated architecture enables the Europa AIPU to deliver an impressive 629 TOPS (Tera Operations Per Second) at INT8 precision. Complementing these AI cores are two clusters of eight dedicated RISC-V vector processing cores, which handle non-AI pre- and post-processing tasks with a peak performance of 4915 GOPS (Giga Operations Per Second). An integrated H.264/H.265 decoder further enhances efficiency by offloading media tasks, freeing the host CPU for critical business logic and application-level workloads. These programmable cores, accessible via the Voyager SDK, promise a superior total cost of ownership by reducing the need to offload tasks to other hardware.
Memory bottlenecks are effectively eliminated through Europa’s purpose-built architecture, which incorporates 128MB of on-chip L2 SRAM and a 256-bit LPDDR5 interface, providing a substantial 200GB/s bandwidth. This high-throughput design translates into a significant competitive advantage, delivering 3 to 5 times greater performance efficiency (measured by performance per watt and performance per dollar) compared to leading industry solutions in the same product category. Furthermore, Europa offers best-in-class performance for generative AI solutions.
Fabrizio del Maffeo, CEO of Axelera AI, emphasized the company’s vision, stating, ‘Organizations shouldn’t have to choose between the raw compute power they want and the real-world usability they need.’ This philosophy underpins Europa’s design, aiming to make cutting-edge AI applications accessible and cost-effective for any organization.
Also Read:
- XMOS Advances Development Tools with Generative AI for Enhanced Abstraction
- On-Device AI Reshapes Consumer Electronics Landscape, MediaTek at the Forefront
Axelera AI will also offer the Europa AIPU in a PCIe card form factor, providing flexible and scalable configurations ranging from single-chip 16GB to four-chip 256GB solutions. These cards are capable of processing complex workloads locally and can simultaneously preprocess and decode multiple 4K HEVC265 video streams with minimal latency and economical power consumption. Shipments for both the Europa AIPU and the PCIe card are slated to commence in the first half of 2026. Axelera AI’s product portfolio also includes the Metis AIPU platforms, which cater to very low-power, size-constrained edge device applications.
