Welcome to the world of Edge AI

The world has been using "narrow" AI applications like chatbots and predictive text for years. Then came generative AI (GenAI), and its poster child ChatGPT, ushering in a new generation of computing and spreading AI far and wide. The disruptive potential of this general-purpose technology has made AI a top priority for companies across all industries.

To date, the growth in GenAI has largely been driven by the development of cloud computing. This is because data and computing power are the two key elements that enable the technology. AI models, like Large Language Models, are trained on enormous amounts of data, typically in cloud-based data centres, since the process demands huge inputs of computation and power.

AI spreads from cloud to edge computing

Edge AI, by contrast, runs AI algorithms locally. The term "edge" in this context means on a hardware device, rather than in a centralised data centre. The shift from the cloud to the edge could be seismic, making AI available everywhere, from your smartphone, computer, laptop, and wearables, to drones, AR/VR (augmented/virtual reality), IoT (internet of things) devices, and cars.

Cloud AI and Edge AI are not an either/or proposition. Today, for instance, a query to ChatGPT is processed at a cloud data centre, not locally on your PC or smartphone. This makes sense for many AI uses today, and will continue to do so in future. But we expect AI to become hybrid, meaning a mix of Cloud AI and Edge AI. Some large AI models will require the computing power of expensive supercomputers.

However, for AI to add value to all aspects of consumer life and business productivity, some AI workloads will need to be pushed to devices at the edge of networks. These AI algorithms will run locally on devices like smartphones and PCs.

As virtual assistants and other apps that require a hardware upgrade are integrated into the devices we interact with every day, Edge AI is expected to grow in importance. GenAI running on everyday devices can offer massively improved capabilities and a better user experience, encouraging consumers to upgrade their smartphones and PCs, which will also be good for hardware manufacturers. 

However, AI is a longer-term trend, and it will take time for the technology to be broadly adopted. We expect semiconductor companies, as the providers of computing power - and the main beneficiaries of Cloud AI - to gain from the development of Edge AI. The major tech companies, with giant troves of data, along with manufacturers of consumer electronic goods and cars, will also benefit. 

Edge AI and Cloud AI are complementary

Cloud computing and Cloud AI will remain important, given the ever-increasing computing power needed to train and use AI models. However, Edge AI will be both complementary and necessary to bring AI into the mainstream.

Edge AI offers advantages that in some ways are more visible to consumers and businesses. Key among these are privacy, security, and personalisation. Because data is processed on the device and not sent to the cloud, users retain greater control over their data. As a result, Edge AI can reduce adoption hurdles when AI uses sensitive and personal data.

Increased privacy and data security permit better use of personal data. By adapting to the local environment and user preferences, Edge AI may improve the user experience of AI, for example by enabling new features on smartphones and IoT devices.

Edge AI also offers some practical solutions to problems that make Cloud AI difficult to implement in all circumstances. One of these is cost. Data centres are very expensive to build, equip, and operate. While they offer unrivalled computing power, the cost per query is typically higher than for queries processed at the edge.

AI servers are also more power hungry than traditional servers, making data centres significant energy guzzlers that can stretch energy grids to the limit. Edge AI should help to relieve pressure on local grids by distributing power consumption to where devices are being used.

Lower latency

Data processing done on a device is necessarily quicker than when data is transferred back and forth to the cloud. Cutting data processing time and delay, called latency, is imperative for certain applications, like self-driving cars. Here data processing needs to happen in real time, with near-zero latency, which simply isn't possible with Cloud AI. 

And finally, the use of Edge AI applications does not require a stable internet connection and can therefore facilitate the adoption of AI in more places. This will be critical for autonomous vehicles, drones and planes, as well as when using underground transport.

Personalised answers, greater productivity

GenAI and specifically Edge AI applications have the potential to fundamentally change how we interact with smartphones, PCs, and other devices to access information. Today, we mostly retrieve pre-produced information. In future, we will be able to ask questions in human language and receive appropriate, personalised answers in real time. 

With AI embedded locally in a device, it will become a tool that grows smarter, learns from its users, and becomes more personalised, leading to more productivity, creativity, and efficiency. At the same time, tech companies may benefit from faster device replacement cycles, premium pricing, and new revenue opportunities.