By Stuart Kerr, Technology Correspondent
Published: 11/09/2025 | Updated: 11/09/2025
Contact: [email protected] | @LiveAIWire
The Shift from Cloud to Edge
For years, artificial intelligence has relied on massive cloud-based models, with billions—or even trillions—of parameters. But a quiet revolution is underway. As IEEE Spectrum reports, tiny models are powering edge devices, enabling AI to run directly on hardware such as smartphones, wearables, and IoT gadgets. These compact systems, sometimes with as few as 94 million parameters, can deliver surprisingly strong performance without the need for constant cloud access.
The appeal is clear: lower latency, reduced reliance on data centres, and enhanced privacy. Instead of sending sensitive data to the cloud for processing, your phone itself can handle voice recognition, image analysis, or translation in real time.
Why Smaller Can Be Smarter
As IEEE Spectrum noted in another analysis on the limits of cloud-bound AI, the latest giant models are not always the best fit for edge computing. By stripping down architectures and applying new quantisation and pruning techniques, engineers can deliver models that are efficient enough for everyday hardware, but still accurate enough for practical use.
Google, for example, recently announced a new wave of on-device small language models capable of handling multimodal input—text, images, even audio—without relying on cloud servers. These systems also integrate retrieval-augmented generation (RAG) and function calling, tools once reserved for heavyweight models.
Edge AI in Academia
The academic community has responded with a flurry of research. The SHAKTI project introduced a 2.5 billion parameter model for edge AI, optimised to run on smartphones and wearables. Another influential paper presented EdgeViTs, lightweight vision transformers designed specifically for mobile inference, balancing high accuracy with reduced energy consumption.
These breakthroughs underscore a fundamental shift: instead of scaling models endlessly upward, researchers are finding new value in scaling down.
The Broader Context
This trend reflects a wider theme in the evolution of technology. Just as our reporting in Beyond Algorithms — Hidden Carbon & Water showed how AI’s environmental costs extend beyond carbon, the move toward edge AI highlights overlooked dimensions of efficiency—like energy, bandwidth, and user privacy.
It also intersects with media and industry trends. In the same way our article on Can Publishers Survive Zero‑Click Era? explored shifts in information flows, edge AI decentralises computing power, challenging cloud monopolies. And like our coverage of AI and Emotional Manipulation, it raises new questions about user trust, autonomy, and control.
A New Frontier for Everyday AI
The promise of tiny models is not just technical—it is cultural. By embedding intelligence directly into devices, AI becomes more personal, more private, and more accessible. No longer tethered to a cloud, users gain faster performance and greater control over their data.
If the last decade of AI was about scale, the next may be about restraint. Tiny models are proving that smaller doesn’t just mean cheaper or weaker—it can also mean smarter.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire. He writes about artificial intelligence, ethics, and how technology is reshaping everyday life. Read more