Why 2062 is closer than it looks

This is an edited version of the keynote by Vocus Chief Technology Officer Nikos Katinakis at CommsDay Wholesale Congress 2026 on 13 October 2026.

Who among us can predict the future? Nobody – if we could, none of us would have to work anymore.

But remember The Jetsons? The cartoon was set in 2062, and in the very first episode there’s Rosey, the “robot maid”. Today, more than 50 million Roombas have been sold, so you could say we’re 36 years ahead of schedule. The flying cars aren’t here yet but they’re coming soon to China, Dubai and Tokyo.  

By 2062, all of us will have some kind of robot at home. Before I moved back from the US last year, a robotics company invited me to try out a $20,000 robot that folds clothes, does the ironing and washes dishes. I didn’t get the chance, but my New York-based colleagues complained it was too big for their tiny apartments.

So why am I talking about 2062? It’s 36 years away, which is just about the life of fibre. Whatever we put in the ground today will still be there in 2062. And we are going to use it.

Humanity gets the timing wrong, not the demand

As a society we are quite good at predicting the future. But we usually get the timing wrong.

When the Sydney Harbour Bridge opened in 1932, journalists called it an over-engineered monstrosity. Anyone who tries to cross the Harbour at peak hour today would agree it could use five more lanes in each direction.

We can’t predict how things will get used.

In 2006, a household used around two gigabytes of broadband data a month. Nobody could have accurately predicted that in 2026 we’d be using 250x that – 500GB. Many families are now pushing towards a terabyte a month.

Where AI’s demand comes from

We’ve been talking a lot recently at Vocus about how much work we’re doing to build new digital infrastructure for Australia. Thousands of fibre strands in each new duct rather than a dozen or two in a direct-buried cable as they are today.

So I want to talk about why Australia needs that much fibre. 

One example is right in front of all of us. Most of us have stopped reading search results on Google and now just read the AI summary. And then when it suggests a follow-up question, you say, yes, give me the next answer, and then you end up having a chat with it.

The text of that chat isn’t that much data. One text question and simple reply is about 8KB. Peanuts. It’s the modern version of sending a text message.

Yet Google went from processing 9.7 trillion tokens a month in 2024 to 3.2 quadrillion in 2026 – that’s 329x the volume in just two years.

Meta has said that a single fibre link between two of its AI data centre needs twice as much capacity as its entire global fibre backbone it spent 10 years building.

AI queries aren’t staying as text. Increasingly we are asking it: can you fix my picture? Can you edit this document? Live voice chat is becoming popular to talk to agents.

North-south vs. east-west

The question and answer you see on your screen is the ‘north-south’ traffic. That’s the data that goes between you and the data centre. It isn’t much.

The real volume is in the other direction, east-west – traffic running between data centres as they process your query.

As the AI “thinks” to answer your query, it calls on data stored in many different places – its own model, the web, your own company’s data and third party systems hooked up to the AI.  

Every time you send a query, the traffic between data centres can be up to 1,000 times what you sent. That is why we need so much fibre.

With AI, your answer on your screen is just the tip of the iceberg; everything else is out of sight – compute, data, energy and connectivity.

The long pole in the tent: connectivity

AI runs on the extremely powerful chips made by Nvidia or Google and others (more and more). It is bringing out new generations of those every six months. Getting power connected to a new data centre takes 12 to 36 months, and building the data centre itself can take 18 to 36 months.

However, building a subsea data cable from Australia to the US takes three to five years.

Our new Sydney to Melbourne route is about 900 kilometres to drive but around 1,100 kilometres to build. That’s three years of round the clock work.

We’re building it with 6,912 fibre strands. That might sound like crazy compared to what Australia has now, but the ducted fibre networks in Europe, India and the US show that it’s not.

Everything I’ve talked about is five to ten years out at best, and none of us can predict the future. Thirty years from now, we’ll be back building more.

Two paths aren’t enough anymore

AI changes how we operate. One of the certainties of life is that cables will get cut. When a 12-strand cable is cut, a crew has to drive out and then dig and find it, but it’s spliced back together within hours. It takes just 20 minutes to fuse together a 12 strand cable.

A pipe with 6,912 strands is different. In my work in the US, we had one ducted fibre system in Los Angeles that someone yanked out of the ground. It took eight days, working around the clock, to put it back together.

A hyperscaler with multi-billion-dollar data centres can’t wait that long. In fact they can’t take the risk that two diverse cables might be cut at the same time. So increasingly we’re talking to them about triversity, or quadversity – three or four independent cables between sites. You can see how demand for fibre multiplies quickly.

And why is the connectivity so vital? Because as devices – including robots – proliferate in our lives to help make life easier, they all depend on connectivity to work. These estimates are all based on our current usage patterns… but those will change too. We will get more used to the models we use, we will send more pictures, more complicated agents, we will create more video…

So, yes, gadgets get the headlines and the glory, but it’s actually a fast, dependable network that delivers the future.

Late, but never wrong

“The future” may have come late, but humanity has rarely been wrong in predicting it. At some point the demand turns up.

Google Glass was ahead of its time, but Meta sold seven million pairs of smart glasses last year. Humanoids at home? Absolutely. Air taxis are coming. Overseas, driverless taxis are the norm in some cities. Here in Australia, you can buy a car and have it drive you from A to B by itself.

All of these things need connectivity, and that isn’t going away. So we’re not just building capacity. We’re running the network as if the world depends on it, because it does.

Remember the name: the Australian Digital Infrastructure Platform. We are going to keep talking about massive investments and big cables, because the demand is coming.