The Mobile World Congress in Barcelona is approaching, and I’ll be there all week with numerous meetings. Key topics this year include #Sustainability & ESG, 5G Advanced, RedCap, AI & Telecoms, Edge Computing & Cloud AI, OpenRAN, and Network APIs.
One particularly impactful topic, closely tied to AI in our phones and telecom infrastructure, will shape the future in developed countries. 75% of the population in the top 10 nations owns a smartphone. In 2024, the number of smartphones in use is estimated to exceed 7 billion.
This smartphone ubiquity is transforming multiple sectors, including #Web3, which relies on blockchain, cryptocurrencies, the metaverse, and phygital experiences. As mobile devices grow more powerful, these innovations are becoming more accessible and seamlessly integrated into users’ daily lives. This facilitates the adoption of digital wallets, immersive platforms, and new decentralized business models.
On-Device Inference: A Key Driver of AI Innovation
On-device inference is fundamentally reshaping the AI landscape. This advancement not only enables more efficient models but also makes AI more accessible and deeply embedded in our everyday devices. It paves the way for unprecedented technological democratization.
➡ On-device inference plays a central role in AI innovation.
➡ Four major trends are enhancing the performance of embedded AI models.
➡ AI models are now more accessible and deployed at scale.
The Impact of On-Device Inference on AI Innovation
On-device inference marks a turning point in AI model quality and performance. Thanks to techniques like model distillation and new neural network architectures, AI models are becoming smaller while maintaining high accuracy. The results are impressive—DeepSeek R1, for instance, outperforms industry leaders like GPT-4 and Claude 3.5 Sonnet in areas such as reasoning, coding, and mathematics.
➡ Another major breakthrough is model size reduction. Techniques like quantization, pruning, and compression shrink models without sacrificing performance. This enables AI deployment directly on everyday devices—smartphones, PCs, and even vehicles.
AI model creation at your fingertips
One significant outcome of this evolution is the democratization of AI model development. In 2024, over 75% of large-scale AI models published had fewer than 100 billion parameters. With lower training costs and growing open-source collaboration, AI model development is now within reach for a broader audience.
A new era of AI applications
On-device inference is also fueling the rise of new AI applications. AI-powered document summarization, image editing, and real-time language translation are becoming everyday tools. Meanwhile, AI is emerging as the new user interface, enhancing interactions with personalized multimodal agents across various applications.
On-device inference is undoubtedly a critical driver of AI innovation. It enables faster, more private, and cost-efficient AI models while significantly expanding the range of applications, interfaces, and benefits. AI is now deeply embedded in our daily lives and numerous industries.
With these advancements, one question arises: Aren’t we all becoming AI model creators in some way?
📲 A WhatsApp Business channel will provide real-time MWC updates with photos, videos, and messages. If you’re attending, let’s grab a coffee ☕ and discuss your business!
Companies integrating on-device AI into their mobile apps
📌 Snapchat (SnapML for AI-powered filters)
📌 TikTok (local AI optimization for video processing)
📌 Adobe Photoshop Mobile (on-device AI enhancements for iOS/Android)
📌 Samsung & Google Photos (AI-driven photo editing and optimization)
📌 Apple Siri & Google Assistant (partial on-device execution on iOS and Android)