• There are many potential use cases in the enterprise for generative AI, but many will be enabled by existing cloud solutions.
• Some use cases requiring real-time responses may emerge, generating modest demand for MEC and/or 5G services.
Expectations of demand for 5G and multi-access edge computing (MEC) services from the enterprise segment are established – in part – on enabling artificial intelligence (AI) to be used in real-time applications. AI requires considerable computing power, usually achieved in the cloud where its demanding requirements can be scaled, but where such resources are too distant (due to network latency) to be relied upon for use cases where seconds or milliseconds in application response time can determine success or failure. There are other reasons why MEC makes sense in this scenario, including both the security benefits and cost savings achieved by not sending massive amounts of data to and from the cloud. With the recent hype around generative AI and the potential impact on various professions, industries, and organizations, it is worth considering whether its uptake will mean even more demand for MEC and/or 5G.
Looking across the enterprise, one can see several potential generative AI use cases happening in an office environment, with office applications, where there is presumably plenty of connectivity and computing power available already, and where the applications aren’t ‘mission critical’ in terms of keeping core operations from falling down. In such cases, cloud-based AI will suffice.
For example, if you are a developer accessing generative AI to help debug code, or if you are a marketer trying to come up with new messaging, or an IT support person troubleshooting a problem for an end user, then you might be driving demand for generative AI solutions, but not for MEC or 5G.
Drivers in Edge Computing, and Vendor Support for AI at the Edge
What’s driving demand for edge computing, including MEC services, are the need for high performance and low latency, real-time data processing, and cost savings (by not sending traffic to the cloud and back) as well as data privacy and security. And, of course, the availability of 5G itself is a driver since it can provide the speed and latency required.
Meanwhile, all edge computing infrastructure vendors are already designing edge systems with processing capabilities needed to support AI platforms. Some have AI server products that can be scaled for edge deployments. There are also hyperscalers, like Microsoft, which already have initiatives to bring their AI services to the network edge to support new use cases.
These requirements for supporting AI in applications at the edge are already seen as a growth driver for 5G and MEC services and will have long since been baked into any demand forecasts, including those published before the advent of generative AI. Those applications include smart manufacturing and logistics use cases leveraging computer vision, which relies heavily on AI, and various retail use cases both in the warehouse and on the shop floor, using AI-driven augmented/mixed reality applications (like Verizon’s ‘Connected Retail’ solution). In fact, AI and ML are used in several different retail use cases where there is lots of data to be harnessed and exploited – including inventory management, digital shopping assistants, and retail transactions (e.g., for fraud detection).
Impact of Generative AI will be Incremental
Solutions that are powering real-time applications will benefit from MEC/5G and fiber connectivity. When it comes to generative AI, one can imagine new customer-facing and retail use cases where the harnessing of ChatGPT or an equivalent for getting real-time, bespoke data is going to help. On the customer side, it can bring a more powerful chatbot experience to physical shopping using voice assistants, and in the back office it can help managers obtain enhanced real-time insights on stock levels. In such use cases, cloud-based AI may or may not suffice. This should apply to other industries where there is a real time need by humans, whether they are customers or staff, for ad hoc data insights on demand.
Another example might be in manufacturing where Siemens and Microsoft are partnering on generative AI for factory automation use cases. They are suggesting it will be used for engineer-assisted use cases like helping with human visual inspections, and they foresee integrating the functionality into mobile devices so the engineers can use natural speech. They also see it assisting engineers in debugging factory automation software in real time, using natural language inputs to generate code. Some of this probably will result in demand for AI-driven MEC and 5G, which is incremental to current forecasts, since these solutions are only being designed now.
For public MEC services, which are more likely to support connected car, gaming, and smart city use cases, AI is already a component of many such solutions. If there is anything to say about generative AI, then, at this early juncture, it might be that generative AI will at least help justify the need for edge compute infrastructure, with potential for a modest level of upside based on the AI hype at the moment that seems to be affecting all industries. While there aren’t really actual solutions yet, the automotive sector is definitely one of those industries, looking at natural language interfaces for owners/passengers in a self-driving vehicle. One can see the benefit of using a MEC and 5G service to enable that functionality, but probably not immediately if it’s first being used for simpler queries like GPS and Siri-like recommendations, or for non-real time operations like configuring a car prior to buying.
It’s too early to say for sure that generative AI will drive more than modest, incremental demand for MEC and 5G services. But at minimum, it should at least shore up existing expectations around demand from AI and ML in general while highlighting some areas that could generate a more meaningful increase in resources required if the use cases pan out and are adopted widely.