Welcome Back to XcessAI

When people think about AI performance, they usually focus on the model.

GPT-5 or Claude? How many parameters? How much context? Which benchmark?

Far fewer people stop to think about something much more fundamental.

The connection between you and the model.

As artificial intelligence becomes embedded into more workflows, the quality of your internet connection increasingly becomes part of your productivity. Today's AI doesn’t require enormous bandwidth, but tomorrow's AI almost certainly will.

This week, let's explore an overlooked question.

How much internet do you actually need for AI?

The surprising answer is: probably much less than you think.

Most interactions with ChatGPT, Claude or Gemini involve exchanging relatively small amounts of text. Even long conversations consume surprisingly little bandwidth compared with streaming 4K video or downloading games.

A typical AI session may only transfer a few megabytes of data.

From a purely bandwidth perspective, almost any modern broadband connection is sufficient.

If AI only involved text, internet speed would barely matter.

But AI is no longer just text.

What users perceive as "speed" is often something else entirely.

Latency.

Bandwidth determines how much information can travel.

Latency determines how quickly it starts moving.

That distinction becomes increasingly important as AI moves from answering questions to becoming an active collaborator.

Imagine asking an AI to review a document, search internal databases, retrieve market data, generate a presentation and send an email.

None of these tasks require extraordinary bandwidth.

But each requires dozens or even hundreds of small interactions between systems.

Every additional delay accumulates.

A one-second delay repeated fifty times quickly becomes nearly a minute.

As AI agents begin coordinating multiple tools autonomously, responsiveness becomes just as important as intelligence itself.

Most internet providers advertise download speeds.

500 Mbps. 1 Gbps. 10 Gbps.

These numbers look impressive. Yet for professional AI users, consistency may matter far more than maximum speed.

An unstable connection interrupts long-running workflows.

Video meetings freeze. File uploads fail. Cloud-based coding sessions disconnect. AI agents lose access to external services.

A slower but highly reliable connection often delivers a better experience than a faster connection that fluctuates throughout the day.

As businesses automate more critical processes, network reliability quietly becomes another form of operational resilience.

Unlike traditional software, modern AI rarely runs on your laptop.

The real work happens inside enormous data centres containing hundreds of thousands of GPUs connected through ultra-fast networking.

Every request you submit travels across fibre networks, enters hyperscale infrastructure, is processed and returns the result within seconds.

The faster and more reliable that journey becomes, the more natural AI feels.

This is one reason technology companies continue investing heavily in networking infrastructure alongside compute.

Moving intelligence efficiently is becoming almost as important as creating it.

Today's AI applications remain relatively lightweight. Tomorrow's probably won't.

Consider what is already emerging.

These systems exchange vastly more information than a simple chatbot. Some will communicate continuously rather than only when a user submits a prompt. Others will coordinate multiple AI models simultaneously.

In that environment, bandwidth, latency and reliability all become strategic resources. The infrastructure supporting AI must evolve alongside the models themselves.

Recently, researchers in Japan demonstrated optical fibre capable of transmitting approximately 125,000 gigabytes per second over more than 1,800 kilometres.

The headline understandably focused on the astonishing speed.

Yet the more interesting development may be what made it possible.

Instead of relying on thicker cables, the researchers designed a new fibre architecture capable of carrying dramatically more information through infrastructure with dimensions similar to today's optical fibre.

Whether this specific technology reaches commercial deployment remains uncertain.

But it highlights a broader trend.

The world is preparing for a future where moving enormous quantities of data efficiently becomes increasingly valuable.

AI is one of the reasons why.

For individual users, probably not.

Most people already have enough bandwidth to use today's AI tools comfortably.

For organisations, the answer becomes more interesting.

Companies operating thousands of AI agents, processing continuous streams of video, running real-time manufacturing systems or connecting distributed engineering teams may eventually discover that networking quality directly influences productivity.

Just as cloud computing created demand for better cybersecurity, AI may gradually increase demand for better connectivity.

The competitive advantage may not come from having the fastest internet.

It may come from having the most reliable path between people, data and intelligence.

Every technological revolution creates new bottlenecks.

The Industrial Revolution depended on railways. The internet depended on fibre optics. Cloud computing depended on hyperscale data centres. Artificial intelligence depends on all of them.

Today's conversations focus on models because that is where the visible innovation happens.

Behind the scenes, however, another race is underway.

Faster fibre. Lower latency. More resilient networks. Smarter routing. More efficient infrastructure.

These may sound like engineering details. In reality, they determine how quickly intelligence can move through the economy.

Perhaps the future of AI will not be constrained by how intelligent our models become. It may be constrained by how quickly we can connect them to everything else.

Until next time,
Stay adaptive. Stay strategic.
And keep exploring the frontier of AI.

Fabio Lopes
XcessAI

💡Next week: I’m breaking down one of the most misunderstood AI shifts happening right now. Stay tuned. Subscribe above.

Read our previous episodes online!