We all know people love to consume content on their phones. They stream videos, scroll through social feeds, and download apps. So it’s no surprise that, up until now,  mobile network infrastructure has been primarily designed to push data down to users. And lots of it — most networks were built to handle about 10 times more downloads than uploads.

But now that’s dramatically changing. 

AI-powered devices like smart glasses, AI companions, autonomous vehicles, robots, drones, and more don’t just receive data. They send it — and, again, lots of it. These devices capture the world through cameras, microphones, and sensors, and stream that information up to the cloud for AI to process in real time. As a result, we’re facing a growing wave of upload traffic that today's networks simply weren't designed to handle.

GSMA, the global trade association representing mobile network operators, has outlined three likely scenarios for future mobile traffic: In its low-growth scenario, uplink stays at around 15% of total traffic by 2040; in medium- and high-growth scenarios, driven largely by AI, that share rises to 25% and 35%, respectively. And Nokia has similarly noted that AI "increases uplink use in the home, injects automation and machine vision into industrial sites, and multiplies east-west movement between data centers."

Let’s take a deeper look at some of the devices that are driving these changes.

The fast-growing market for smart glasses is a prime example. For every byte of data they receive, they send nearly eight bytes back up to the network, continuously uploading video and audio feeds to cloud-based AI solutions to do things like answer questions or translate a conversation in real time.

AI companions are essentially pocket-sized AI assistants that are always on, always listening, always uploading. Unlike a smartphone that you pick up and put down, these devices run continuously, sending a steady stream of sensory data to the cloud. They represent what some are calling the "third core device" alongside phones and laptops, and they behave very differently from anything networks have had to support before.

Self-driving autonomous vehicles like taxis and delivery vehicles are already on the road in growing numbers. These vehicles need to constantly send telemetry data and, in some cases, live video feeds to human operators who can step in when the car encounters a situation it can't handle on its own. That's a strong, near-constant demand on upload speed and capacity.

Robots in factories and commercial drones in the air bring their own pressures to bear. Robots need to send video and sensor data to remote operators, while drones stream live footage during flight. Each one is, in effect, a small but persistent upload machine.

It’s clear these and similar devices will need more and more upload capacity. But they’ll also need to be fast and reliable. After all, many of the applications they’re supporting can’t afford any hiccups when safety and reliability are key. That’s why:

• Speed matters in real time. When a pair of smart glasses is translating a conversation, even a brief interruption can break the experience entirely. The AI models powering these features aren't built to handle spotty connections.

• Reliability is non-negotiable. When a self-driving car is sending data to a remote operator, or a robot is performing a task in a factory, a “pretty good” network isn’t good enough. It needs to perform consistently, every time.

• Upload coverage is the weak link. Today's networks are built with the powerful capability to push data down to devices. But devices like phones, glasses, and sensors transmit at a fraction of that power. That makes upload coverage, especially indoors and at the edges of a cell zone, the biggest bottleneck in the system.

About 9% of AI inference flows or pipelines — the core complex functions that include agentic AI — carry more upstream than downstream traffic, compared to 0.5% for typical, less-complex web traffic, according to a new report from Cisco.

It’s clear that modern mobile networks need to change their game. Today’s streaming video and downloads can tolerate delay with minimal impact because content is pre-buffered. But real-time and interactive AI applications place fundamentally different demands on the network, turning uplink into the new bottleneck, not downlink.

There are several ways that mobile networks can adapt to this changing world:

• Software upgrades can unlock more upload capacity from existing infrastructure — like smarter antenna configurations, better interference management, and AI-driven tools that optimize how the network allocates resources.

• Spectrum strategy matters, too. Certain frequency bands are better suited for upload coverage, and operators need to make the most of what they have.

• Antenna improvements can dramatically expand upload performance, in some cases doubling or even tripling capacity at the edges of a cell where coverage is weakest.

• AI-powered network management can help anticipate surges in upload demand before they cause congestion, rather than just reacting after the fact.

Shifting from a download-focused model to an upload-centric one isn't just a technical challenge. A significant revenue opportunity awaits those network operators who move early.

Networks that can guarantee reliable upload performance will be able to offer premium service tiers to businesses that depend on it, such as an autonomous vehicle fleet that needs assured data delivery, or a live event venue that wants to guarantee connectivity for thousands of simultaneous streams.

Operators can also open up their networks to developers through programmable interfaces, allowing AI applications to interact directly with the network and leverage the performance levels they need, when they need them. This creates an entirely new category of revenue tied to the growth of AI.

As AI continues to drive the need for bigger and faster upload capacity, businesses must begin to adjust their networks to meet the needs of applications in real time.

That’s why Vonage will soon be unveiling a new addition to its network API solution set. Quality on Demand will ensure low-latency, high-throughput connectivity using advanced quality-of-service differentiation and 5G network slicing. The network slice serves as an individual virtual network away from public use and traffic.

Need to stream video from drones, deliver lag-free mobile gaming, or power AR/VR experiences? Quality on Demand enables API-triggered adjustments to latency, throughput, and traffic priority — exactly when and where it matters most.