Multiverse & Axelera partner to push AI to the edge

Multiverse Computing has partnered with Dutch chipmaker Axelera AI to run advanced artificial intelligence models directly on edge devices, potentially shifting some workloads away from cloud-based GPUs and datacentres.

The collaboration brings Multiverse's compressed AI models to Axelera's Metis platform and its forthcoming Europa platform. The companies are positioning the combination for use cases that require local processing because of power limits, latency requirements, or privacy constraints.

Model compression has become a focal point for organisations looking to deploy larger models outside the datacentre. Smaller models reduce memory requirements and chip demands, and lower energy use where a constant connection to cloud infrastructure is impractical or costly.

Multiverse says its CompactifAI technology can shrink large language models by up to 95%, with a 2% to 3% loss in precision. That reduction can make it possible to run models on smaller systems and changes the trade-offs between centralised compute and distributed deployments.

Axelera sells hardware designed for AI inference at the edge, a category that has grown with the rise of computer vision systems in industrial settings and retail. It has also gained attention as generative AI spreads beyond knowledge work into operational environments that depend on near real-time response.

The integration is expected to let workloads that normally require "datacenter-class infrastructure" run on compact devices. The companies also plan a commercialisation programme around the joint offering.

Enrique Lizaso, co-founder and CEO of Multiverse Computing, said the partnership is designed to run more capable models closer to where data is generated. "Our mission is to make state-of-the-art AI radically more efficient and accessible. By combining Multiverse's advanced compressed AI models with Axelera's high-performance edge platforms, we can bring powerful reasoning capabilities to devices where latency, privacy and energy consumption are critical."

Edge deployment

Edge AI deployments typically focus on inference rather than training. Many industrial environments also have limited bandwidth, and some sites operate with intermittent connectivity. Local inference can keep systems running when networks fail and reduce the volume of data that needs to be transmitted.

The companies framed the partnership around both inference and on-device fine-tuning. Fine-tuning at the edge remains technically demanding because it requires memory and compute headroom. Compression can improve feasibility by reducing the size of the base model and the resources needed to adjust it.

Axelera said the collaboration targets customers across industrial, retail, mobility, defence and smart city deployments. Fabrizio Del Maffeo, co-founder and CEO of Axelera AI, said the deal broadens the applications its platform can address. "Axelera AI is committed to delivering the most powerful and efficient AI inference solutions to the world. Enabling Multiverse Computing's compressed AI models to run on our Metis and future Europa platforms will unlock new classes of applications for our customers, from industrial and retail to mobility, defense, smart cities and more."

Metis is Axelera's first-generation edge AI platform. Europa is its next platform, which the company has signalled as part of an expanded product line that also includes Titania chiplets. The partnership focuses on integrating compressed models into Metis and Europa, rather than developing a separate product line.

European sovereignty

Both companies are backed by the European Innovation Council and its fund, and they linked the partnership to European policy goals on technological sovereignty. The idea of a "sovereign" AI stack has gained prominence as governments and enterprises reassess dependencies on US hyperscalers and non-European chip supply chains.

Supporters argue that local development and manufacturing can reduce exposure to geopolitical shocks and supply constraints. They also point to regulatory requirements and public-sector procurement rules that can favour European providers in some contexts.

Ekaterina Zaharieva, Commissioner for Startups, Research and Innovation, said the partnership shows how European deep-tech firms can combine hardware and models within the region.

"Europe's competitiveness in the next decade will depend on our ability to combine world-class chips with trustworthy, efficient AI. Collaborations like the one between Multiverse Computing and Axelera AI, both supported significantly by the European Innovation Council and its Fund, show how European deep-tech companies, when connected to each other, work together to deliver sovereign strategic digital technologies that are developed and scaled in Europe while serving global markets."

The work will cover "ultra-efficient inference and fine-tuning engines," linked to lower energy consumption on edge devices, local fine-tuning to keep sensitive data on-site, and deploying models across large fleets at a lower cost per device.

Multiverse is headquartered in Donostia-San Sebastián in Spain, with offices in Europe, the US and Canada. Axelera is based in Eindhoven and has staff across several European countries as well as Taiwan and the US. After integrating the models on Axelera's platforms, the companies plan to move into a dedicated commercialisation phase.