Viet Nguyen, President, 5G Americas (June 2025) –
Recently, I was privileged to be able to chair a wonderful track at the Edge Computing Expo at TechEx in Santa Clara that was focused on edge computing and the Internet of Things. Over the course of the day and in a panel on scaling edge networks, the running question emerged of how AI, and specifically generative AI, was going to impact network requirements at the edge.
At the same time, there has also been a raging debate on LinkedIn and other social platforms among some very top-tier, senior wireless analysts over the direction of how much mobile data traffic is going to increase or decrease in the Age of AI. On the one hand, we have some powerful voices lending guidance that AI is going to increase mobile data use in wireless networks – specifically on the uplink (upload) side. While others suggest that on-device AI processing will reduce the demand for mobile wireless data.
Who’s right? Who’s wrong? Well, I certainly have my opinions. And so do many others.
According to the data from GSMA, AI is projected to add between 20% and 80% additional traffic to mobile networks beyond current forecasts, depending on the adoption trajectory. That is not marginal. Ericsson finds that GenAI will be pervasive across device hardware, operating systems and applications, leading to net-new traffic traffic growth particularly on the uplink, especially from applications with high adoption and high data rate requirements.
A lot of that traffic will be carried over 5G networks. In 2024, the 5G share of mobile data traffic reached 35 percent and is expected to grow to 80 percent in 2030, according to Ericsson. With global 5G connections reaching 2.4 billion as of Q1 2025 and projected to reach 8 billion by 2029, 5G monthly mobile data is expected to reach 225 exabytes per month by 2030.
In the age of AI, it has become increasingly popular to assume that moving computation closer to the edge or even directly onto devices will alleviate network stress. However, such assumptions underestimate the complex dynamics between compute distribution, user behavior, and service architecture. Things get messier at the edge and create network stress in different ways.
The underlying reason is both simple and structural. AI changes how data is consumed and created, and these shifts ripple through the mobile ecosystem. AI-driven applications—from real-time assistants to generative content tools—are not passive clients of data services. They are active engines of demand, initiating traffic patterns that are distinct from conventional mobile use cases. Even when AI processing occurs at the edge, the associated ecosystem—from data synchronization and federated learning to content delivery and contextual inference—demands new bandwidth and creates persistent uplink and downlink flows that challenge previous optimization models.
As we stand at the intersection of widespread AI adoption and edge network expansion, it is clear in my view that mobile data growth will not plateau. It will escalate—and the edge will be a primary engine of this growth.
But let’s take a look at the very real and very convincing arguments from both sides:
| Reasons for INCREASE in Mobile Data Demand | Reasons for DECREASE in Mobile Data Demand |
| 1. Proliferation of AI Assistants and Chatbots Cloud-based AI assistants increase both uplink and downlink traffic through frequent, complex interactions. | 1. On-Device AI Processing Reduces Data Transmission Local AI processing (voice, image, text) reduces the need to send data to the cloud, minimizing network traffic. |
| 2. Growth of Generative AI Applications AI tools for creating images, video, and text require significant data transfer, especially when processed in the cloud. | 2. Decline in Mobile App Usage Due to AI Assistants AI assistants replace many app functions, potentially reducing data-intensive app activity. |
| 3. AI-Driven Video Streaming Personalized recommendations and higher video consumption driven by AI increase mobile data usage. | 3. Increased Use of Wi-Fi Over Mobile Networks AI features that work offline or locally further shift usage to Wi-Fi instead of mobile networks. |
| 4. Expansion of AR/VR AI-powered AR and VR require real-time data exchange and high bandwidth, driving up traffic. | 4. Plateau in Mobile Data Growth Efficiency gains and changing usage patterns may flatten growth rates by 2027. |
| 5. Explosion of IoT and Smart Devices AI in IoT leads to billions of new devices generating and transmitting data. | 5. Privacy and Security Regulations Favor Local Processing Privacy laws and user concerns push sensitive tasks to on-device AI, reducing network data. |
| 6. Hybrid Edge and Cloud AI Processing Workloads sent between devices, edge, and cloud increase total data transmitted. | 6. Offline Functionality Enabled by AI On-device AI allows apps to work offline, reducing reliance on mobile data. |
| 7. Personalized Content and Advertising AI-driven personalization increases engagement and time spent on data-intensive platforms. | 7. Efficiency Gains from AI-Driven Compression and Optimization AI optimizes/compresses data before transmission, minimizing data sent over networks. |
| 8. Enterprise AI Adoption Businesses use AI for analytics, automation, and remote work, increasing real-time data needs. | 8. Reduced Latency and Faster Responses Locally Local processing provides faster responses, making cloud/edge transfers less necessary. |
| 9. Real-Time Analytics and Monitoring AI enables continuous data uploads from devices and sensors for real-time monitoring. | 9. Consolidation of Apps and Services AI-driven consolidation leads to fewer, more efficient apps and lower overall data use. |
| 10. AI-Enabled Network Automation and Optimization AI enables new high-bandwidth services and higher throughput, raising total data handled. | 10. Diminishing Returns and Plateau in AI Model Performance Slower AI model improvements mean fewer new data-intensive apps, shifting toward efficiency. |
The answer is, of course, not clear. That’s the challenge with forecasting. There are a lot of factors to consider.
One thing to keep in mind is that edge workloads are very different than cloud workloads. In cloud environments in a typical data center, things are very standardized and controlled, as data centers are meant to deliver on *scalability* of compute and storage. Scalable AI workloads can work well in a containerized Kubernetes kind of environment, so AI workloads are very optimized. The servers and racks are all standardized, the air is at the right temperature and well-conditioned, and security is very tight. Have you ever tried to get inside a data center? It’s like Fort Knox.
On the other hand, at the network edge, where all the sensors and devices exist, things are quite a lot messier. You might have an exposed camera or toaster connecting wirelessly to the network. The edge server might be sitting in a closet. Security is a bit of a mess. There’s no real standardization – indeed, every edge site is unique with very different levels of compute, storage, and wired or wireless connectivity. In fact, in these messy edge environments, many devices don’t even have an operating system. They’re just running on firmware.
So if you introduce AI workloads into this environment, the LLM needs to sit somewhere. Not every device manufacturer will want to add an entire operating system onto their device to run AI workloads sitting on top of it. This means unless there’s an integrated AI system on a chip, all of the computation will need to be run on the edge server – and the data transferred either via wire or wirelessly.
Let’s admit it, device manufacturers are generally loathe to add more cost and complexity onto their devices if the market for them is unproven. Does the camera itself need to have a built-in chatbot? Or should that chatbot features sit on the edge server? The push-pull with device manufacturers entering into the “smart device” space will broadly determine how much data will continue to need to be delivered over the network versus processed locally on device.
Indeed, there are certainly some devices like smartphones which have a lot of support for on-device AI. The silicon used in smartphones can drive a lot of onboard processing very quickly. But as you start to move into the higher nm ranges for IoT devices, the use case gets dramatically murkier.
Consider, for instance, all the optical cameras that might exist on a mobile robot walking or rolling around a manufacturing facility. Those are likely using wired optics that connect to some kind of on-board compute. But let’s assume for a moment that the robot is passing by an area that have some additional remote cameras. Is the data from those remote cameras sitting on some kind of storage device locally? Will the robot be downloading the data as it walks by?
So will on-device AI suppress network traffic overall? While it is true that some localized AI inference tasks reduce latency and enhance privacy, the technical constraints of devices—especially limitations in processing power, memory, and energy efficiency—mean that high-performance AI will continue to depend on edge servers. And even when on-device models operate autonomously, they still require periodic updates, training synchronizations, and integration with broader services that rely on cloud and edge processing. These background operations alone can create significant uplink demand, with generative applications spiking usage up to 25 Mbps during peak interactions.
More importantly, AI changes user behavior. Content recommendations powered by AI drive longer session times, more video playback, and increased user interaction across applications. Social platforms and streaming services see measurable gains in engagement from algorithmic personalization, directly translating into more data consumption per user. This indirect traffic—while not always visible as part of the AI stack—is nonetheless generated by AI’s influence on the application layer.
So in this context, edge computing becomes not just a tactical optimization but a strategic imperative. You have to start strategizing about how the AI workloads will interact in this messy edge environment before you deploy. Network planning must be adjusted to fulfill the latency, bandwidth, and scalability requirements of modern AI services while positioning carriers to better manage the spatial distribution of load.
Of course, Edge processing is not free of network implications. Proximity reduces latency but does not eliminate the need for transmission. In fact, edge architectures often introduce additional traffic layers, especially in hybrid scenarios where edge nodes feed insights back to centralized systems for orchestration, analytics, or compliance.
Overall, infrastructure investments are reflecting this new reality. Telecommunications edge computing spending is projected to grow from $25 billion in 2023 to $46.5 billion by 2028. Similarly, edge data centers are expected to grow at over 17% annually, reaching $109 billion globally by 2034. These figures are not driven by aspirations of reducing data – they’re driven by the necessity of supporting more of it, closer to where it’s created and consumed.
And not just data, even sustainability considerations reinforce this trajectory. While on-device AI offers energy savings for some workloads, cloud and centralized data centers face growing scrutiny for their power demands. The US Department of Energy notes data centers are projected to consume 9% of total U.S. electricity demand by 2030. Edge computing offers a middle ground: distributed enough to minimize latency and reduce transmission energy, yet powerful enough to handle high-complexity tasks efficiently. But make no mistake – this does not reduce the volume of traffic. It redistributes it and enables more of it.
So where do we stand?
Across regional markets, the story is consistent. North America and North East Asia have already exceeded 50 percent penetration rate for 5G networks, enabling sophisticated AI applications to flourish on robust infrastructure. Meanwhile, markets like India are becoming global data leaders thanks to affordable mobile access and high user engagement. These parallel developments will only intensify AI usage and its resulting impact on mobile traffic.
As a wireless industry, we must re-calibrate our assumptions. AI does not represent a neat substitution of local compute for global bandwidth. Instead, it introduces new paradigms of engagement, coordination, and application complexity that expand the data landscape. Edge networks are central to this evolution — not as a cap on growth, but as the very mechanism that enables and amplifies it.
Preparing for this future requires us to embrace hybrid architectures, modernize radio access networks, and invest in agile capacity planning that anticipates unpredictable demand patterns. The edge is where intelligence meets mobility. And as AI continues to shape how we interact with information and each other, the data it generates will increasingly move across, and because of, the edge.
-Viet
