EINDHOVEN, Netherlands - November 12, 2026 - Axelera AI, the leading provider of purpose-built AI hardware acceleration technology, unveiled the details behind Titania, its third-generation silicon architecture in a chiplet design. Titania is highly programmable and energy efficient and will deliver high performance at interactive speeds, all within the power and cooling limits of existing data centers, rather than the 100-kilowatt-plus liquid-cooled racks that today’s highest-density GPU systems require.
Performance expectations for Titania are up to 7x the throughput per watt compared to the best data center grade servers available, and up to a 20x better TCO. Each Titania-based system-in-package (SiP) will contain two Titania chiplets which are then packaged in a range of form factors, including standard PCIe cards as well as OCP-compliant UBB baseboards with eight Titania OAM modules connected in a scale-up network. In support of over $300 million in confirmed Titania commitments, these form factors will be available via OEM partners in an inference server configuration in both air- and liquid-cooled configurations.
Titania builds on the programmability, performance and efficiency of its shipping predecessors, Metis and Europa, and further expands the product line from embedded, edge and enterprise AI into large-scale data center inference. Designed for interactive and highly variable AI workloads of enterprises, sovereign clouds, decentralized and modular data centers, telcos and AI factories, Titania combines Axelera AI's high-performance, energy efficient Digital In-Memory Computing (D-IMC) architecture with a new chiplet-based design, high-capacity LPDDR6 memory, high bandwidth scale-up connectivity and programmable RISC-V compute in one package. Available in multiple form factors including industry standard PCIe cards as well as OCP-compliant boards, Titania offers customers the flexibility to solve for their inference needs. Supported through the Voyager toolchain, Titania deployments will support tensor, expert and pipeline parallelism, as well as speculative decoding, KV cache offloading and disaggregation out of the box. In practice, that means large models use every card in the servers, each user gets tokens faster and operators get more useful work per kilowatt from the rack they already own.
“The biggest challenge most organizations face as inference becomes the dominant AI workload is the tokenomics of their infrastructure,” said Fabrizio Del Maffeo, co-founder and CEO of Axelera AI. “For enterprises, it’s becoming impossible to own their own AI infrastructure due to high CAPEX expenditures including rebuilding their data center footprint. So they choose between a server that can only compute small or medium models and paying over one million dollars for a data center server that requires specialized infrastructure. Titania solves this problem and gives enterprises, neoclouds, telcos and sovereign clouds alike the opportunity to take advantage of owning and controlling their own inference infrastructure.”
Designed for the infrastructure customers already have
Titania builds on the D-IMC architecture first introduced with Axelera AI's Metis platform and extended through Europa, enhanced with floating-point support, more and smaller cores, and better coordination of data movement to deliver exceptional performance for generative AI and data center scale.
The Titania architecture’s highly efficient data movement and compute engine enables it to use low-cost, readily available LPDDR6 memory rather than relying on scarce high-bandwidth memory (HBM). LPDDR6 gives each package 128 GB of capacity at lower memory power than HBM, while avoiding any dependence on HBM supply or packaging requirements. Sparse mixture-of-experts models are large in total parameters but small in active parameters, so memory capacity and effective utilization matter more than peak bandwidth if it goes unused. The result is a platform designed to deliver high-performance inference while reducing power, cooling, and infrastructure requirements.
Titania will scale from PCIe cards for standard servers and OAM modules to multi-accelerator servers and larger AI infrastructure deployments. An eight SiP accelerator Titania reference server (codenamed AX8) is designed to provide 20 PFLOPS of FP8 compute, 1 TB of accelerator memory, and 19.6 TB/s of memory bandwidth within a standard 4U, air-cooled server, at 5-6kW typical, enabling large AI models to run within existing data center power and cooling environments. One Axelera server will hold a 1.6 trillion-parameter model at FP4 with the headroom for KV cache, two fit within a legacy 15 kW rack and five fit within a 30 kW air-cooled rack.
“We don’t ask anyone to trust a simulator,” said Evangelos Eleftheriou, CTO and co-founder of Axelera AI. “The architecture behind Titania already turns 88 percent of its memory bandwidth into tokens on Europa silicon that is shipping today, and Titania builds on that capability as the third generation of our architecture. It's not simply about maximizing peak compute. It's about how efficiently we can turn power, memory, and infrastructure into useful AI performance at scale.”
One architecture and software foundation from edge to data center
Titania is Axelera AI's third-generation AI acceleration technology, following Metis (optimized for performance and energy efficiency) and Europa, which extended the company's architecture into enterprise AI infrastructure.
The same Voyager SDK and AxeleraScript (AxScript) software foundation extends across all three generations, allowing customers to develop and optimize AI applications on Axelera AI technology today and carry those workloads forward as the portfolio expands. The Titania PCIe card is designed as a drop-in replacement for the Axelera Server 250p (a FHFL four-chip Europa card), so models deployed on Europa today move to Titania in 2028 without re-porting.
One card serves LLMs, vision-language models, and video: on-chip 8K decode and 4K encode with H.265 and AV1 codecs remove the host-CPU cost of video pipelines, so a single Axelera AI system runs language, speech and vision models alongside live video.
Titania will be offered across multiple form factors, with OEM and ODM partners building the technology into servers and larger systems. Axelera AI’s Titania goes into production in Q3 2027 and already has customer deployment commitments through the beginning of 2028.
Designed in Europe, trusted globally
Designed in Europe with support from the EuroHPC Joint Undertaking's DARE initiative, Titania extends Axelera AI's mission to make high-performance, energy-efficient AI infrastructure more accessible to enterprises, cloud providers, governments and AI infrastructure operators globally.
Axelera AI will provide a deeper technical look at the Titania architecture during SC26, November 15–19, in Chicago, IL., at booth 3251 in McCormick Place. Book a meeting with Axelera here.
About Axelera AI
Axelera AI is the leading European AI semiconductor company developing next-generation AI acceleration hardware for edge computing and data center applications. With an efficiency-first architecture that addresses AI’s critical energy and cooling constraints, Axelera AI has deployed across more than 600 customers, including those in telecommunications, aerospace and enterprise sectors. The company’s Metis, Europa and Titania architecture-based platforms deliver high-performance AI inference solutions that enable deployment at scale within real-world power and thermal budgets. Through its Partner Accelerator Network and Voyager SDK, Axelera AI provides a comprehensive ecosystem approach that ensures accessibility and rapid deployment.
Contact
Matt Stubbs
Voxus PR