It’s surprising that AOL even offered dial-up internet in 2025, but they’ve now discontinued it.
On September 30, 2025, AOL will finally discontinue dial-up internet service, along with the AOL Dialer and AOL Shield Browser. For millions of early internet surfers, the distinctive screech of a modem handshake heralded the arrival of the future. Now, the last of the original on-ramps is closing, and the highway it led to is unrecognisable.
Dial-up usage has plummeted over recent years. In 2015, AOL had around 1.5 million dial-up users, but as of 2021, that number had reportedly dropped to “the low thousands.” So, while this change may not affect many people, it’s still surprising that dial-up usage has persisted for so long. AOL originally launched its dial-up service in 1991, meaning it will be 34 years old when it’s finally shut down next month.
Unfortunately, some of the dial-up users may not have many other internet access options. Broadband infrastructure hasn’t reached some remote rural areas, or if it does, it’s expensive. Many AOL customers who were still subscribed to dial-up likely stuck with it out of necessity. Fortunately, AOL isn’t the only dial-up provider; there are still a few others offering this antiquated service.
The Velocity of Data is Increasing and Will Always Increase
Dial-up’s maximum speed was 56 kilobits per second. Today’s 1-gigabit fibre service, a conservative baseline in many markets, represents approximately 18,000 times faster performance in just 34 years. However, raw speed misses the point. We didn’t just get a faster internet; we got a fundamentally different internet that enabled entirely new categories of human experiences. Every improvement in network performance led to the invention of applications previously unimaginable. The progression from dial-up to today follows a classic exponential technology curve, with significant business implications.
A Predictable Path
The technical path from screech to symmetry follows predictable patterns. Copper was pushed to its physical limits. Cable operators learned to squeeze gigabits from coaxial networks designed for television. Fibre completely redefined the baseline, with new standards delivering 25 gigabits over the same glass that once carried megabits. Wireless caught up not only on speed but also on reliability, which is arguably even more crucial for real applications.
Meanwhile, satellite internet transformed from a joke into a credible broadband option through low-earth orbit constellations. Starlink now delivers speeds exceeding 100 megabits with latency in the mid-20 milliseconds, fast enough for video calls from moving vehicles. You can access 500 Mbps service in most major cities, with many markets offering multi-gigabit tiers.
The increases in internet speed didn’t just lead to faster internet; they also enabled the creation of entirely new products and services, such as streaming video at scale, cloud-first software development, real-time collaboration across continents, live commerce and interactive entertainment, and remote production workflows. In essence, high-speed internet has enabled the always-online world we live in today.
The current half-trillion dollar (and counting) hyperscale data centre buildouts represent the next velocity threshold. These investments will only generate returns if AI inference can reach customers over reliable pipes with minimal latency. Your future AI assistant won’t feel intelligent if it takes three seconds to respond to voice commands.
To truly enjoy augmented reality (AR), we’ll need motion-to-photon response times under 20 milliseconds. That’s not a nice-to-have spec; it’s a table stake for the category. We can expect latency for comfortable AR experiences to be measured in single-digit milliseconds at the edge.
Then, there’s the emerging world of AI agents. These systems will coordinate across multiple services, process real-time data streams, and respond to environmental changes instantly. An agent that takes 10 seconds to react to your calendar change or traffic pattern isn’t useful; it’s annoying.
We need to design for tomorrow’s velocity, not today’s average. Our customers won’t thank us for building applications that “work fine on gigabit.” They’ll abandon us when those applications feel slow compared to competitors who assumed 10-gigabit symmetrical service.
Next, upstream bandwidth and consistent low latency will matter as much as headline download speeds. The shift to cloud-rendered experiences, real-time collaboration, and always-on AI means our users are producing as much data as they consume. Jitter and packet loss will decide whether professional workflows feel instant or unusable.
Lastly, network architecture will become an even more competitive race. Companies leading the way in AI won’t just rent more GPUs; they’ll position compute at the network edge, minimising data round-trips and designing for congestion scenarios. If your product strategy relies on AI responsiveness, network topology becomes just as important as server specifications.
What’s Next?
The roadmap is surprisingly predictable. 10-gigabit residential service will become commonplace as fibre buildouts accelerate and cable operators deploy new standards. In-home wireless will prioritise reliability over raw speed, which is crucial for AR devices and local AI systems. Mobile networks will push compute to the edge to support real-time applications. It’s going to be exciting out there.
