A new economy is coming. Big tech is already building it, and not just here on Earth.
Nobody can tell you what it will actually look like. But nine of the world's biggest tech companies are already carrying $3 trillion in AI infrastructure commitments, five times what they spent last year, and some of them have stopped looking at Earth entirely.
They're building in space.
I found this out over a pizza in Cape Town, the night before the Springboks played the All Blacks during a chat with Bartek Ogonowski, founder and CEO of LEVRA, a UK start-up, who was in Cape Town for the game and, it turns out, keeps a close eye on where AI money is heading.
We were talking about how a new economy, it seems, is being built before anyone can clearly describe what it'll look like. However, if you follow the money, the companies building it appear to have made up their minds.
Ogonowski asked whether I knew about the infrastructure being built in space, which I'd only vaguely heard about. So convinced are the tech giants of that future, he said, that they're taking the infrastructure for it into orbit.
I suppose, in a way, it makes sense: like every gold rush, this one is running out of ground to dig - and not only ground, but water and electricity too.
AI data centres are power-hungry. They need enormous quantities of constant electricity, water for cooling and land near power grids that can actually support them.
This isn't just a Silicon Valley problem.
In Cape Town, where I'd shared that pizza with the founder, municipal authorities recently approved Equinix's proposal to develop two data centres in the city - expected to require about 170 megawatts of power capacity - despite objections from activist groups over the project's water use, electricity demand and environmental impact.
The debate in Cape Town is a glimpse of the larger problem: the AI economy may be digital, but the infrastructure it depends on is intensely physical.
According to the CEO, some companies have started looking at the one place with genuinely unlimited solar energy and no neighbours to object: space.
Who's actually doing this?
I went away and started digging: this is no longer entirely theoretical.
A startup called Starcloud has already launched its Starcloud-1 satellite carrying an Nvidia H100 GPU into orbit, and has demonstrated AI processing there, including running Google's open-source Gemma model, according to Nvidia's Technical Blog.
Still, this is only one small satellite doing a proof-of-concept run, not a data centre.
SpaceX and Google are exploring orbital computing and communications infrastructure too.
It won't happen soon, they say, though things do have a way of snowballing - estimates for meaningful, gigawatt-scale orbital capacity sit closer to the late 2030s or 2040s.
The idea: build computing capacity above the Earth, power it with abundant solar energy and connect the systems through high-speed optical links.
Nobody spending on orbital compute knows for certain whether it'll work, technically or financially. Yet there seems to be a pattern repeating at a bigger scale than before: build first, on the belief that the future will need what you've built, and worry about proving it later.
Meanwhile back on earth
This space push doesn't seem like a separate story from the money already flooding into AI infrastructure on earth - the very thing Ogonowski and I had discussed over that pizza.
Consider the scale of what's already committed down here.
Nine top tech companies (including Alphabet, Meta, Oracle, and Microsoft) carry roughly $3 trillion (about R48.15trn) in off-balance-sheet AI commitments for chips, data centres, power and networking, according to a comprehensive Wall Street Journal investigative analysis. That is five times what they have actually spent on capital expenditures in the past year.
And for the three months ending July 2026, Nvidia's official Investor Relations report posted a net profit of $59.7 billion (R961.6bn), a 126% increase from the previous year. How did it do that? Nvidia sits at the centre of the shift to accelerated AI computing.
"Compute is revenue… and demand is accelerating," said Jensen Huang, founder and CEO of Nvidia, when releasing the figures. He added that the AI infrastructure buildout "is at full steam".
Nvidia isn't just selling into this boom, it's helping finance it too, partnering with some of Wall Street's biggest financial institutions to mobilise hundreds of billions of dollars for AI infrastructure and chip purchases.
Anthropic is a good example: the company has committed $45bn (R724.89bn) over six years for compute capacity at Nscale's Monarch Compute Campus in West Virginia, a 460-megawatt facility that won't exist until late 2027, built to run on Nvidia chip architecture that hasn't shipped yet.
A reality check
According to Bain & Company's Global Technology Report, the scale of global AI infrastructure build-out will require roughly $2trn (R32.21trn) in annual revenue by 2030 just to break even. Even with efficiency gains, Bain's models project an $800bn annual revenue shortfall by decade's end.
This is a huge amount of infrastructure being built against a future that still has to arrive - and space computing is the same story, stretched out a decade further. But nobody, it seems, is claiming this pays for itself in the 2020s.
Sam Altman, CEO and co-founder of OpenAI, has admitted he was too ambitious about the pace of AI disruption, telling podcaster David Senra in August that he'd expected much faster disruption of software businesses and the wider economy after GPT-4 launched in 2023.
He said he was "wrong" about the speed, citing economic "inertia" as one reason businesses and society had adapted more slowly than expected.
But that hasn't slowed the building.
Regulators are already struggling to keep pace with AI infrastructure on Earth - in orbit, the rules barely exist at all yet.
One thing is for sure though: if money is an indicator of the future, the world's biggest tech companies are showing us what they believe it looks like.
We may not know exactly what the new economy will become, but they are building for it anyway, even in space. Rules or no rules.
Are we even ready for it? How close are we to all of this?
I asked Ogonowski that exact question. He didn't answer straight away. Instead, he asked me one back: "Did you see how close that AirLink plane flew over the DHL Stadium during the game?"
I had. Close enough that half the crowd looked up before they looked back at the scrum.
"That's where we are," he said. "That close. Still with room to move though."
*Warby is the Digital Editor at The National. She chases weak signals — the small conversations, details and behaviours that hint at where AI, technology and human behaviour are heading next.