AI at the Edge: The Expanding AI Ecosystem in 2025

Hi, I’m Olivier Blanchard. I am the Research Director and Practice Lead for AI Devices, Automotive, and AI Device Semiconductors at the Futurum Research Group.

What we’re thinking here in our holistic practice, looking at all of our different technology categories, is that AI devices are actually taking off this year. It’s probably going to be one of the bigger trends in the overall tech sector. Up until now, we’ve had AI living in the cloud. In terms of training, inferencing services, most of it has been handled in the cloud through data centers and cloud services. But what we’re doing this year, or what we’re seeing this year rather, is a migration not away from the cloud, but into devices towards the edge. So essentially, it’s an expansion of the AI ecosystem from the cloud outward to the edge to devices. And these devices are AI PCs – that’s obviously one of the big developments for this year – I’ll come back to it in a second. 

Another one is AI-enabled mobile devices, which actually isn’t new, but we’re going to start seeing a lot more on-device agentic AI entering the market this year. And also, all of the other devices that are sort of peripherals to PCs and mobile. So that’s the wearables, like your watches, and XRs, so smart glasses. It’s also drones, it’s also smart cameras, smart speakers. All of the little ecosystem of intelligent devices that you can interact with, either by voice or other types of interfaces. And we’re also seeing an expansion of that into the automotive space, where cars aren’t just about self-driving and ADAS. They’re also about agentic experiences inside the vehicle for the drivers and passengers. 

So all of these things together are essentially driven by advancements in two areas. One is semiconductors, so the small semiconductors. The semiconductors that go into your devices, your PCs, your mobile phones, your watches, your smart glasses, your speakers, all of your wearables, and in the vehicles are getting much better. A lot of them are equipped with something called an NPU, which is a neural processing engine, or unit, that allows AI workloads to happen on-device. And increasingly, what we’re seeing is not just inference, which is sort of the AI interacting with you on devices, it’s also the training itself. And we’re now able, with AI PCs, mobile devices, and vehicles, to train AI directly on the device without necessarily needing a cloud connection, or connection to the cloud, or to a data center. So that’s a huge thing, because it allows training of AI to remain local, to be secure, and to be a little bit more immediate. And that’s also with the inferencing, which is kind of the interactions that we have with AI. 

AI models are becoming a lot more efficient to train. And so what used to have to be trained in the cloud a year ago, that required, you know, 70, 100 billion parameters, is much smaller now, and can run directly on devices, or can be trained directly on the device. So we’re gonna see an expansion of training AI models from the cloud into smaller solutions, like small servers, more local, and also some of that training being moved to AI devices. 

All of that and more can be found in our eBook about our predictions for 2025. We’re talking about not just AI devices, but a lot of other technology categories as well that all play into this big AI revolution that we have. Also, I recommend that you follow us on the socials. So we’re on LinkedIn, obviously, we’re on X, anywhere you can find us, anywhere you can find me, where I talk about AI devices, and also follow us on our websites, where you’ll find a lot of other resources, and that’s futurumgroup.com. Thanks a lot, happy reading, and I hope to see you around.