Europe is converging on Cloud-Edge-IoT. The order still decides.

Field note v1.1 — written 17 August 2026 during and after the 13:00–15:30 CEST meeting; first published 13:48 CEST the same day. Version history at the end; the current version is always the one you are reading.

Monday, 13:00, Gothenburg. The August meeting of the Working Group Standards & Platforms — SEK Forum IoT together with the W3C Smart City Nordic Community Group — and Torbjörn Lahrin is walking us through the past, present and future of the group. Reports, autumn planning, and then a run of slides that say more about where this field is going than most conference keynotes. This note is what I took from them, and what WINNIIO thinks it means.

Six programmes, one theme

Europe is converging around Cloud-Edge-IoT. CEI-Sphere builds the ecosystem and the interoperability. O-CEI develops open, sustainable platforms for edge computing. COP-PILOT orchestrates interoperable Cloud-Edge-IoT services across domains. MetaOS orchestrates distributed computing across the edge-to-cloud continuum. Swarms enables decentralised intelligence across distributed edge devices. And around all of it, the Data Spaces work. Six programmes, one theme: Cloud + Edge + IoT + Data Spaces + AI, treated as one continuum instead of five departments that never speak to each other. The stated aim is to lay the foundation for how data is shared between IoT devices and cloud solutions inside the EU, and to get computational power closer to the data.

If you work with IoT and have not looked at these, the working group's advice was blunt: look them up. The umbrella is EUCloudEdgeIoT.eu, and CEI-Sphere runs open task forces on open source and standardisation, market uptake and outreach that anyone can ask to join.

The interesting part was always the outcome

In 2022 I wrote a piece called IoT is dead. Long live IoT! The point was never the sensors. Logistics, automotive, real estate, steel, healthcare — the outcome, the impact, has always been the interesting part, and it has always been available. But we built fragmented views of reality, and the fragments separated what was needed from the context it was needed in. So we reached for generalised approaches, one size for everyone, and called them best practice. Best practice is rarely cutting edge.

The fix is an order of operations, not a product. Start with the impact you want. Find the decision that changes it. Then, and only then, go and get the data. Data last. That order is the whole of SMILE, the methodology WINNIIO publishes openly — v6.4.4 is on Zenodo, free to read and free to reuse.

Compute moves to the data, and the AI changes shape

Where is this heading? Compute moves to the data. Data stays where it is created, and the experts and the models travel to it. AI sits on top of an honest, shared picture of reality — not on top of last month's Excel export.

And the AI itself is changing. Physical AI — AI systems that directly interact with the physical world — is now on the standards agenda. The slide named the committees that will have to cooperate: JTC 1/SC 41 (IoT and digital twin), SC 42 (AI), SC 27 (security), SC 38 (cloud), SC 6, IEC/TC 65 (industrial automation), IEC/TC 47 and ISO/TC 299 (robotics), with SC 41 as a possible integrator, and JTC 1/AG 2 already drafting a technical report. Two lines from that slide are worth repeating verbatim: safety and trustworthiness cannot be treated as optional.

Physics does not care what a language model believes. That is why we think the next step is from LLMs to LQMs — Large Quantitative Models. A language model is trained on statistics about text; a quantitative model is built on the equations of the system it describes — the heat balance of a furnace, the flow in a grid, the pharmacokinetics of a body. Language to talk to it. Physics to trust it. And the safest way to try anything in the real world is to not try it in the real world first: try it, break it, optimise it in a physics-based copy of that environment — the factory, the grid, the hospital ward, the body — and only then let it touch reality. Everything virtually first. The only dashboard that truly works is reality itself.

The human loop gets drawn into the architecture

Who is in the loop when it does touch reality? Humans. Human-Machine Teaming — humans and intelligent machine systems as interdependent partners, human context, judgment and ethical oversight combined with machine speed and autonomous execution — is on the same agenda, with ISO/IEC 25589 for the framework and ISO/IEC 42109 for the use cases. The reason it matters was on the slide: digital twin conversations often assume full automation. They shouldn't. Operations centres, utilities, manufacturing, infrastructure, healthcare — the applications listed are the ones where a wrong autonomous decision costs the most. The working group's expectation is that future digital twin architectures will explicitly include human decision-making loops.

That has been the WINNIIO position since day one. People in control. AI assists, it does not dictate. Every decision overridable. It is good to see the standards catching up, and it is worth saying why it is not a soft principle: a twin that cannot show a human why it wants to act is a twin nobody with a licence to lose will let act.

Testing and certification are how a twin becomes a procurement requirement

The quietest slide was the one an industrial buyer should care about most. AhG34 — conformity assessment and quality assurance for IoT and digital twin systems — has handed its recommendations to AG 6: conformance assessment, quality assurance, certification considerations, applicability guidance. That is the sequence a field goes through as it matures. Early stages are architecture and terminology; later stages are testing, certification and compliance. IoT, and above all digital twins, are moving into the later phase.

Why it matters was stated in one sentence on the slide, and it is the sentence a procurement officer has been waiting for: digital twins only become procurement requirements once they can be tested and certified. Until then a twin is a project. After that it is a line in a tender, with a conformance clause behind it. It is also why we score every twin against reality before we call it one — the Twin Readiness Score exists because a twin that cannot be tested cannot be trusted, bought or insured.

Why any nation should read this as a resilience question

Because it is one, long before it is a technology question. A country whose hospitals, grids and factories only work while a cloud in another jurisdiction answers has built a single point of failure and called it digitalisation. Edge-native means the local loop keeps running when the link is cut. Data spaces mean the data never has to leave to be useful. Sovereignty is architecture, not policy. We wrote the longer argument as The Resilience Musts of a Nation.

And Sweden? We have the standardisation seat, the pilots, the engineers, and a total-defence reason to get this right before 2030. The pieces exist. The order is what is missing. The Swedish version of the argument is Sverige 2030 — Sverige i tiden, with an English edition.

Where to plug in

• The Working Group Standards & Platforms meets monthly, hybrid, chaired by Torbjörn Lahrin. To be on the send list, email tobbe@of-us.se.

• SEK Forum IoT (Svensk Elstandard) — registration via felix.soneback@elstandard.se.

• W3C Nordic Chapter Smart City Community Group — w3.org/community/smartcity-nordic; create a W3C account and join the group.

• Europe's Cloud-Edge-IoT programmes — eucloudedgeiot.eu, and the CEI-Sphere task forces at ceisphere.eu.

SMILE — connecting outcomes, people and AI through reality.

Version history

• v1.0 — 17 August 2026, 13:48 CEST — first published, from the slides as presented, while the meeting was still running.

• v1.1 — 17 August 2026, 15:25 CEST — timestamp and version history added; no change to the argument.