Key Takeaways
AI computing is expanding from data centres towards the end user, creating a more robust AI experience.
What might an edge AI experience look like? Imagine an AI-powered phone that translates conversations in real time, or a self-driving car with a personalised AI assistant.
The expansion to edge AI can create investment opportunities in edge AI devices, services and AI infrastructure.
 | The following content has been prepared by Allianz Global Investors GmbH (AllianzGI), and is reproduced with permission by Voya Investment Management (Voya IM). Certain information may be received from sources Voya IM considers reliable; Voya IM does not represent that such information is accurate or complete. Any opinions expressed herein are subject to change. Nothing contained herein should be construed as (i) an offer to buy any security or (ii) a recommendation as to the advisability of investing in, purchasing or selling any security. |
Generative Artificial Intelligence (AI) computing is moving from centralised data centres to “edge AI”, promising faster and more personalised experiences for users—and potentially attractive opportunities for investors.
Generative AI emerged as a transformative force in 2023, capturing the world's attention and attracting significant investments. But generative AI is still in its early stages, and we believe its benefits are currently concentrated in a limited part of the value chain. We believe that this is likely to change as generative AI is adopted more broadly in the coming years, and the world of “edge AI” seems set to be a key growth area with analysis estimating that by 2025, Edge AI will be responsible for half of all enterprise data created1.
Edge AI, explained
Currently, most AI compute activity (including generative AI) is centralised in massive data centres, which enables AI models to leverage a tremendous amount of processing power. But as generative AI becomes more widely adopted, we expect that computing will expand towards the “edge,” closer to the end user. In this scenario, smartphones, cars, PCs and edge servers will shoulder more of the compute load, enabling faster and lower-latency experiences.
What might this look like to end users like us? Imagine AI-powered smartphones that can provide real-time, accurate conversational translations while traveling abroad. Or a self-driving car equipped with a personalised AI assistant. Companies that take advantage of edge AI will have an opportunity to differentiate themselves, offer more personalised and efficient services, and gain a competitive edge.
Why moving to edge AI makes sense
To understand why edge AI can make such a difference compared with a centralised model, it’s important to understand the two main types of compute activities for AI: training and inference.
- Training involves developing intelligence within an AI model, similar to how humans spend years in education and training before entering the workforce. The training process for a generative AI model like ChatGPT involves analysing vast amounts of text from websites, books and Wikipedia articles. This training is facilitated by thousands of interconnected high-end graphics processing units (GPUs), which are best positioned to do the kind of processing that this phase demands.
- Once an AI model is adequately trained, it is deployed for broader use, a process known as inference. During inference, the AI model performs tasks in response to user requests, such as generating photos, providing restaurant recommendations or summarising recent events. Inference compute can surpass the demands of training, especially when millions of users are actively engaging with generative AI applications.
Conducting AI inferencing exclusively at data centres can be costly and can result in sluggish user experiences. The best way to solve this problem is to distribute the inference compute across the network – among data centres, edge servers and the devices that are closest to the user. This has already happened with mobile games and social media photo filters, where the inference compute is distributed at the edge to enhance the user’s experience. Generative AI is likely to follow a similar trajectory.
How investors can plug into edge AI
Recently, several companies have made announcements regarding upcoming edge AI product launches:
- A few major semiconductor companies have unveiled new neural processing units (NPUs), which are chips specifically designed to run AI algorithms. (One of these companies has even projected that AI-powered PCs could reach a staggering 100 million units within the next two years2.)
- Another prominent semiconductor company has introduced a mobile computing platform that boasts on-device generative AI capabilities.
- One of the world's largest smartphone manufacturers has expressed its intention to integrate AI into every smart device they produce.
The introduction of new edge AI devices and services carries significant investment implications. First, edge AI has the potential to reinvigorate PC and smartphone sales, which have experienced sluggish growth3 as users increasingly wait to upgrade their devices. In addition, the adoption of edge AI devices paves the way for the emergence of the next "killer app," similar to the explosion of mobile apps that followed the initial wave of smartphones. And finally, we believe that the success of edge AI is likely to drive further research efforts aimed at achieving breakthroughs in new AI models, thereby requiring companies to invest more in AI infrastructure.