Data Centers are Important, but Perhaps 30% to 40% of AI Processing Will Happen on End User and Edge Devices

AI PCs clearly are coming. So are AI phones. So one issue is what apps and features will make sense to run locally, on the edge devices, as opposed to at remote cloud data centers. 

And by some estimates, perhaps 30 percent to 40 percent of AI operations will happen on edge and end user devices. 

Title	Date	Publisher	Key Conclusions
"The Rise of Edge Computing in AI"	2023	IDC	Predicts 40% of AI workloads will shift to edge devices by 2025, driven by the need for real-time processing and privacy concerns.
"The Impact of AI on Data Center Markets"	2024	Gray Insights	Highlights the growing complexity of AI workloads in data centers; edge computing reduces latency for applications like autonomous vehicles and IoT devices.
"AI at the Edge: The New Computing Model"	2024	Gartner	Notes a 30-40% growth in edge AI applications in industries requiring real-time analytics, but large-scale training remains predominantly in data centers.
"Data Centers in the Age of AI"	2024	EFFECT Photonics	Suggests increasing decentralization with AI-driven analytics at the edge to meet sustainability goals and latency requirements for high-demand applications.
"The New Era of AI and its Impact on Data Centres"	2024	Technology Magazine	Emphasizes the importance of data centers for large AI models but recognizes that edge computing is growing for use cases requiring local processing.

And much of that AI processing will probably happen on smartphones. 

Study	Date	Publisher	Key Forecasts
Worldwide Generative AI Smartphone Forecast	July 2024	IDC	GenAI smartphone shipments to grow 363.6% in 2024, reaching 234.2M units (19% of market); 912M by 2028.
AI PCs and GenAI Smartphones Market Update	October 2024	Gartner	54.5M AI PCs (22% of PCs) shipped in 2024; combined AI PCs and GenAI smartphones to reach 295M units in 2024.
Smartphone and PC Industry Trends for 2024-2028	October 2024	Canalys	19% of PCs to be AI-capable in 2024, growing to 60% by 2027; smartphones to integrate GenAI more robustly by mid-decade.

Mostly, data centers will be needed for high-intensity enterprise and business operations such as training AI models, complex generative AI inferences, enterprise data processing, trend analysis and complex simulations. 

Task Type	Best Performed at Data Centers	Reasoning
Training of AI Models	Deep learning model training Large-scale data analytics	Requires vast computational resources (GPUs/TPUs), extensive memory, and high data throughput.
Complex Generative AI Tasks	High-resolution video rendering Advanced generative simulations	Demands high-performance hardware and is often time-insensitive, making centralized processing more efficient.
Big Data Processing	Batch processing Data aggregation and mining	Involves handling terabytes/petabytes of data, which is impractical for local devices.
Real-time Global Analytics	Cloud-based monitoring Predictive maintenance	Requires aggregation and processing of data from multiple sources across regions.
Highly Parallel Computation	Scientific simulations Cryptographic processing	Leveraging massive parallel processing clusters is more effective than limited on-device cores.
Complex Simulations	Climate modeling Large-scale physics or financial simulations	Demands high precision, vast data sets, and sustained processing, which exceeds the capabilities of on-device hardware.
Data Archiving and Backup	Cloud storage Long-term data management	Centralized data centers offer cost-effective, scalable, and reliable storage solutions.
Collaborative Workflows	Cloud-based co-editing Team project management	Requires simultaneous access and synchronization by multiple users, which is best managed through a central server.

But many consumer-facing operations can, and will, be provided directly on device. 

Camera image processing, language translation, predictive text operations  or speech interfaces already seem logical. Perhaps videoconferencing and onboard document search also will be seen as logical. 

Face recognition or biometric recognition obviously make sense for local processing for security reasons. And many local content recommendations will be able to operate using local processing as well. 

But what is “logical” still hinges on whether the particular operations on device and at the edge make more sense than operations at a remote processing site.  And issues such as battery life will play a part in that determination. 

Device	Key Features	Applications
AI PCs	On-device AI processing for real-time tasks	Productivity: AI-driven tools for editing, real-time transcription, and advanced virtual assistants.
	Generative AI for content creation	Content Creation: Automatic photo, video, and document editing.
	Predictive performance optimization	Gaming: AI-optimized game settings for better performance and user experience.
	Enhanced cybersecurity via AI threat detection	Security: AI-based malware detection and phishing prevention.
	Voice and natural language processing (NLP)	Collaboration: Smart meeting summaries, automatic translation, and intelligent chat responses.
AI Smartphones	Generative AI capabilities on-device	Camera Enhancements: AI for advanced photo editing, real-time effects, and improved low-light photography.
	Localized large language models (LLMs)	Personal Assistants: Context-aware responses, proactive suggestions, and personalized reminders.
	Neural processing units (NPUs) for faster AI computations	Health Monitoring: AI-based diagnostics, personalized fitness plans, and stress detection apps.
	Voice and gesture recognition	Accessibility: Enhanced voice-to-text, gesture-based navigation for differently-abled users.
	Integration of AI with IoT devices	Smart Home Control: Seamless management of smart appliances through conversational commands.