From Micro to Meta Edge: Architecting Generative and Agentic Edge AI Technologies and Autonomous Systems.

Copyright © 2025 1 European Conference on EDGE AI Technologies and Applications - EEAI 20-22 October 2025, Naples, Italy The intersection of imagination and execution, where edge AI learns to create, reason, and act. Copyright © 2025 European Conference on EDGE AI Technologies and Applications - EEAI Milan, Ita20-22 October 2025 Naples, Italy 2 Copyright © 2025 Ovidiu Vermesan, SINTEF AS, Norway From Micro to Meta Edge Architecting Generative and Agentic Edge AI Technologies and Autonomous Systems Copyright © 2025 Presentation Outline •Edge AI Granularity •Evolution of AI Models •Edge AI and Automotive Industry •Edge AI Dynamics •Edge AI Transformation •GenAI Engineering Tools •Edge AI Evolution 3 Copyright © 2025 Edge AI Granularity 4 Applications DSPs, FGPAs, CPUs, GPUs, ASICs Network Processing Unit (NPU), Intelligent Processing Unit (IPU). Tensor Processing Unit (TPU), Reduced Instr. Set Computer RISC-V, Neuromorphic. Computing units (industrial processing, panel units, etc.) , network computing units (intelligent routers, switches, gateways and other communications hardware), intelligent controllers (PLCs, RTUs, DCS). Micro and clustered servers to handle compute intensive tasks / workloads (e.g., high-end CPUs, GPUs, FGPAs, etc.), on premises edge computing, local edge. Cloud Infrastructure. Local, regional and national data centres. Federation of clouds and data centres. Multi-access edge computing (MEC) infrastructure. Fog processing and platforms Deep Edge Meta Edge Data Centers Micro Edge Micro Deep Meta MEC Cloud Fusion Fusion Continuum-X Continuum-X Continuum-X Continuum-X HW/SW/AI Data HW/SW/AI Data HW/SW/AI Data HW/SW/AI Data Sensors Actuators GPU ASIC TPU IPU Real-time Processes Fog MEC Computing Continuum Copyright © 2025 Evolution of AI Models – From Giants to Efficiency •2025 continues the trend towards model miniaturisation. •Processors for AI - Focus on training 5 Next-Gen AI needs more chips, faster, and smarter systems released: From generative AI to agentic AI and physical AI. Source: YOLE Group 2025 Adapting large models to specific uses accelerates demand for flexible, general-purpose GPUs and AI accelerators that can handle smaller-scale training and tune closer to the edge, expanding semiconductor demand beyond hyperscalers. Copyright © 2025 NVIDIA – From Gaming to Playing the Game •NVIDIA is no longer just a processor designer but now provides a complete data center solution with its DGX Pod / SuperPod platform. Intel and AMD try to follow this trend. •AI ASIC designers are seeking to weaken the AI monopoly of NVIDIA and AMD with highly energy-efficient processors. Large companies for whom processor development is not their core business rely on external firms to co-develop their AI ASICs. • Startups and established companies will partly face direct competition in the market. •AI hyperscalers are looking to control a larger part of the supply chain, starting with AI accelerators, to massively reduce their CAPEX. •Hyperscaler expansion is also taking place in ARM-based CPUs, which are more energyefficient than x86. •The slowdown of Intel and the more power-hungry x86 processors are making way for more energy-efficient ARM processors. 6 NVIDIA at the center of chess game - Data center value chain From cloud to edge, the path forward demands a unified evolution of the compute stack to match AI’s continuous movement, balancing throughput, latency, energy efficiency, and scalability. The future of innovation depends on this. Source: YOLE Group 2025 Next-Gen systems demand far more from HW than earlier models. Agentic AI, which involves autonomous decision-making, requires continual inference and advanced memory architectures to execute tasks. Physical AI, embedded in robotic systems, adds real-time sensory integration, mobility and processing, further increasing compute requirements. Copyright © 2025 Edge AI and Automotive Industry •The evolution of semiconductor process nodes from 180 nanometre (nm) to below 5nm marks a dramatic leap in performance, power efficiency and integration density. •Leading semiconductor companies are taking distinct strategic approaches to capture value in the automotive market. •Newer entrants and compute focused firms like Qualcomm and Nvidia are positioning themselves as end-to-end platform providers, software-defined capabilities offering centralised compute, AI acceleration, and scalable architectures. •Efforts are underway within groups like the Automotive Edge Computing Consortium and the broader RISC-V ecosystem to define chiplet-based reference architectures for vehicle applications. 7 Source: YOLE Group 2025 Copyright © 2025 Edge AI Dynamics •Edge AI evolves with multimodal models, agentic AI and industrial-grade deployments. Edge AI is entering the next phase, moving beyond TinyML models to running multimodal large language models (LLMs), small language models (SLMs) and vision-language models (VLMs) directly on industrial-grade edge devices. •Reasoning models offer a significant advancement due to their characteristics to solve complex problems with high accuracy. •Agentic AI create new opportunities to improve productivity, and decision making, and require trust and proven applications. •Multimodal GenAI is transforming applications by enabling the addition of new interfaces, features and functionalities. •Synthetic data help multiple industries and use cases as it lower the barriers for reliance on real-world data and enables opportunities for simulations and data-generation techniques. 8 05 04 03 02 01 Improved Creativity Ethical Considerations Real-Time Applications MultiModal Generative AI Hybrid Models GENERATIVE AI Copyright © 2025 Edge AI Transformation •The transformation is accelerated by: 9 •AI accelerators efficiency and performance that support real-time inference for models with billions of parameters meeting industrial latency and thermal constraints. •Edge AI solutions are powered for real-time, local processing. Real-world edge deployments focus on compact 3B–8B models to optimise latency, cost, and energy efficiency. •LLMs, SLMs and VLMs are becoming more edge-optimised. Architectural innovations and lightweight multimodal models are enabling inference at lower computational requirements. Zero-shot VLMs (ability of a model to understand and perform tasks without having been specifically trained on those tasks) are changing edge applications. •The edge AI ecosystem is maturing on the HW, SW and AI side with new frameworks offerings that manage the full lifecycle of edge AI, from model training and optimisation to deployment, over-the-air (OTA) updates, and runtime management. Copyright © 2025 Event Organisers 16 The objectives of LoLiPoP IoT (Long Life Power Platforms for Internet of Things) are to develop energy harvesting-based innovative Long Life Power Platforms that enable retrofit of wireless sensor network edge devices for asset tracking, condition and performance monitoring. www.lolipop-iot.eu LoLiPoP IoT EdgeAI-Trust AIMS5.0 SC4EU SC4EU is a unique Chips JU "Innovation Action" project to take the supply chain management of semiconductor production in Europe to a new level. A true demand platform, along with its ontology as a formal description of all information within the chain, facilitates close interaction and smooth, transparent collaboration, making even highly complex supply chains resilient, flexible, and agile. https://sc4.eu/ EdgeAI-Trust addresses an advanced, trustworthy edge AI ecosystem through cuttingedge hardware, software, and tools. The project aims to enhance decentralized EdgeAI operations that are secure, reliable, and sustainable. By integrating AI-based algorithms, devices, and APIs, EdgeAI-Trust fosters interoperability and secure data exchange across diverse platforms. from sensor-actuated devices to cloud systems, all within a dynamic zero trust environment. https://www.edgeai-trust.eu/ AIMS5.0 aims to boost the economy by adopting, extending, and implementing AIenabled HW and SW components and systems across the entire industrial value chain. New technologies from IoT and based on Semantic Web ontologies, ML and AI help European manufacturers to shift from Industry 4.0 to Industry 5.0, creating humancentric workplace conditions and a climate-friendly production. https://aims50.eu/ Copyright © 2025 Supporting Organizations 17 The European Technology Platform on Smart Systems Integration is an industrydriven policy initiative, defining research, development and innovation needs as well as policy requirements related to Smart Systems Integration and integrated Microand Nanosystems. The main objective is to develop a vision and to set up a Strategic Research Agenda. www.smart-systems-integration.org Inside Industry Association is the European Technology Platform for research, design and innovation on Intelligent Digital Systems and their applications. The Association is a membership organisation for the European research and innovation actors with more than 200 members and associates from all over Europe. www.inside-association.eu Chips Joint Undertaking supports research, development, innovation, and future manufacturing capacities in the European semiconductor ecosystem. Launched as part of the Chips for Europe Initiative, it confronts semiconductor shortages and strengthens Europe's digital autonomy, engaging a significant EU, national/regional and private industry funding of nearly €11 billion. https://portal.chips-ju.europa.eu/ EU AENEAS EPoSS INSIDE Chips JU AENEAS standing for Association for European NanoElectronics ActivitieS, is an industrial Association, established in 2006, providing unparalleled networking opportunities, policy influence & supported access to funding to all types RD&I participants in the field of micro and nanoelectronics enabled components and systems. https://aeneas-office.org/