Highlights
- Edge AI can help enterprises unlock actionable insights for real-time responsiveness.
- We see three major driving factors for edge AI adoption—purpose-built hardware, AI frameworks, and novel security techniques.
- An edge-native and strategic approach will be the way forward for enterprises to bring out the best of edge and AI.
On this page
Edge AI use cases
From improving customer experience to meeting sustainability goals, edge AI can help enterprises drive business outcomes.
With the Internet of Things (IoT) and the massive amount of data generated, the potential for enterprises to draw actionable insights for real-time responsiveness is huge. However, this requires more computing power, greater system performance, robust networks, and higher bandwidth. Enterprises must first address challenges such as network latency, bandwidth issues, and data security.
Edge AI, which combines the power of edge computing and artificial intelligence (AI), holds the answer. It can help enterprises run machine learning tasks directly on resource-constrained edge devices and make decisions quickly at the edge of a network. In other words, enterprises can truly get an edge with edge AI. Edge analytics is projected to reach $47.4 billion by 2030, growing at a compound annual growth rate of 24.9%.
There are numerous applications for edge AI. Some of them are:
- Smart factories, smart grids, and smart cities – Quality assurance, surveillance, and monitoring are areas where the technology can make a difference. Through smart grids and smart cities, it can also reduce energy drains with a smaller energy footprint, lower the networking traffic of centralized servers, and lead to low bandwidth utilization.
- 5G mobile edge computing – Edge AI can boost self-driving vehicles, driver monitoring systems, and vehicle-to-everything communication.
- Connected healthcare – The technology can transform clinical decision support systems, wearables, and remote surgery.
- Smart retail – Edge AI can improve in-store experiences and reduce shrinkage or loss of inventory from theft, errors, fraud, waste, and damage.
- Improving customer experience – As chatbots, voice-based interfaces, biometrics, and virtual worlds redefine customer engagement, edge AI can help enterprises deliver on the promise of anytime, anywhere services.
Edge AI adoption
We see three things emerging as driving forces behind edge AI adoption.
These are:
- Purpose-built hardware: Current edge hardware and the accelerators and multi-core central processing units they use for AI processing may be suboptimal for powerful AI workloads in the future. The existing computing architecture may be unable to keep up with the real-time response and energy required for continuous data processing. New AI chips are paving the way for innovative edge AI use cases. These AI chips with different architectures, such as memristor-based computation-in-memory, and computing technologies, such as neuromorphic computing, can address some bottlenecks around energy consumption and real-time response.
- AI frameworks and architecture: Software technologies and AI frameworks accelerate AI orchestration, integration, and deployment in edge computing. Besides, the continuous evolution of model compression techniques, neural architectures, Kubernetes variants, and edge-native frameworks enables larger AI models to run on edge devices without compromising AI performance. TinyML, a subfield of machine learning (ML) with a target energy cost of 1mW or less per inference action, is advancing swiftly and promises new capabilities for edge AI cost effectively.
- Novel security techniques: Enterprises need to build new defenses for AI-specific vulnerabilities or attacks at the edge, such as model extraction, model inversion, model inference attacks, and adversarial input attack. Novel security techniques such as confidential computing, homomorphic encryption, multiparty computation, and federated learning can help mitigate edge AI-specific risks. These add security layers to existing cybersecurity strategies, such as zero trust, secure access service edge, and defense-in-depth strategy.
A strategic approach
Edge-native is the way forward.
Enterprises mostly implement edge AI as an extension to their cloud solutions with AI models trained on the cloud and time-critical inferencing executed at the edge—what some call a cloud-out strategy. However, some take an edge-in strategy—adopting edge AI with applications and services developed specifically for the edge, along with some support services in the cloud. Cloud out or edge-in, a strategic, consistent, and enterprise-wide approach is a must to accelerate edge AI deployment and extend the value derived from the technology. Also, an edge-native mindset—with infrastructure and applications that leverage cloud-native principles while considering the unique characteristics of the edge—can bring out the best of the edge and AI.
Overcoming challenges
Robust processes, ethical practices, and edge-native solutions are key to success with edge AI.
Enterprises embracing edge AI often face challenges such as managing and governing data in a distributed environment and achieving optimality between data volume, model performance, and hardware capacity. Security challenges related to data, hardware, software, and applications caused by dynamic changes and distributed scale, and compliance requirements for individual edge locations are other complexities. To accelerate their edge AI deployment and extend the value of their investment, businesses need to:
- Establish robust processes to proactively identify, qualify, and prioritize appropriate use cases.
- Bring edge AI solutions under the existing enterprise AI and data governance framework.
- Adopt edge AI tools, edge operations platform, and edge-native solutions to execute edge AI at scale.
- Formulate a cohesive edge-to-cloud zero trust security model to mitigate data, access, hardware, network, and AI vulnerabilities.
- Embed AI ethical practices (guardrails, fairness, explainability) and integrate privacy-preserving techniques like federated learning.
- Regulate edge AI solutions and applications with a risk-based approach and establish controls based on the level of risks.
Edge AI is expected to evolve quickly and play a key role in driving business innovation. Enterprises focused on growth and the future would do well to edge in now with edge computing, AI, and cloud.
