The Boom Is Just Beginning — and That's the Scariest Sentence in Finance

The Boom Is Just Beginning — and That's the Scariest Sentence in Finance

Let me tell you something that's been sitting with me since the weekend. Alan Kohler — the man who's been reading Australian balance sheets longer than most of us have been alive — looked at the AI buildout and said the record investment boom we're living through is just beginning. Not peaking. Not cracking. Beginning.

I've been running hosting infrastructure for over a decade. I've watched this AI thing go from a curiosity to a capex monster to the single biggest financial event of my working life. And when a guy like Kohler — who called the dot-com bubble from the cheap seats and has been right more times than he's been wrong — says the greatest investment boom in history is still in its early innings, I don't get excited. I get careful. Because "just beginning" is the most dangerous phrase in finance, ent?

The News — What Alan Kohler Actually Said

Here's the setup. ABC News Australia published a Kohler explainer over the weekend that lays out the bull case in one sentence: artificial intelligence is driving what he describes as the greatest investment boom in history, with global AI spending expected to reach US$1 trillion next year. That's not a typo. One trillion dollars. A year. On AI infrastructure.

But Kohler didn't stop at the headline number. He said something that should be framed and hung on the wall of every data center operator in America: unless companies start generating substantial profits from AI products, the bubble may burst. He's been saying versions of this for a year — back in October 2025 he pointed out that the total cash tied up in AI and crypto bets was more than a quarter of global GDP, and called it probably the greatest technology investment boom or bubble in history. Notice he didn't pick a side then. He's still not picking a side now. He's telling you both things are true at the same time: the boom is just beginning, and the profit test hasn't been passed yet.

That's not a contradiction. That's a warning wrapped in a forecast.

The Two Readings — The Runway vs. The Reckoning

Here's where I have to be fair to both sides, because the numbers genuinely support two different futures.

The first reading: the boom is just beginning, and that's good news. The hyperscalers have committed roughly $700 billion to $770 billion in combined AI capex for 2026 — Amazon around $200 billion, Microsoft around $190 billion, Google in the $175 to $185 billion range, Meta $115 to $135 billion, plus Oracle and a dozen smaller players on top. That's up about 77 percent from roughly $410 billion in 2025. Analysts are already projecting $1 trillion-plus in 2027 — which is exactly the number Kohler is quoting for next year. Deloitte pegs global AI data center capex at $400 to $450 billion this year alone, with $250 to $300 billion of that going straight into chips. Dell'Oro says global data center spending hits $1 trillion a year by 2029. Data center capex is projected to jump from 1.4 percent of GDP in 2025 to 3.1 percent in 2027. If you run a hosting business, this reading says: demand is real, the money is real, and the runway is longer than anyone thought. Lock in capacity. Order hardware. Grow.

The second reading: the boom is just beginning, and that's terrifying. Because a boom that's "just beginning" at $1 trillion a year hasn't been tested yet. It's still being fueled by balance sheets, not by profits. The New York Times ran a piece on August 14 about how Amazon's and Alphabet's own profits are increasingly circular — tech companies investing in each other, buying each other's compute, booking revenue that ultimately traces back to the same handful of balance sheets. When the profit test finally comes — and it always comes — the question isn't whether the boom was real. It's whether the revenue underneath it was real. The 2001 telecom crash wasn't a lie about demand for bandwidth. It was a lie about how much anyone would pay for it.

The Numbers Behind "Just Beginning"

Let me be concrete about what "just beginning" actually means in dollars, because the scale is what most people miss.

We're talking about roughly $700 billion in hyperscaler capex in 2026, near-doubling to $1 trillion-plus in 2027, and $1 trillion a year as a steady-state run rate by 2029. That's not a cycle. That's an industry being built from scratch in real time — power plants, transmission lines, cooling systems, fiber, GPU factories, memory fabs. The five largest hyperscalers are projected to spend $660 to $725 billion this year, with roughly 75 percent of that — $450 to $500 billion — directly tied to AI infrastructure: GPUs, data centers, networking. The top four US cloud providers alone are expected to grow 2027 capex at about 40 percent, a record $210 billion-plus increase in a single year.

Here's what those numbers tell me as someone who actually runs servers: this is not 2021, when everyone threw money at crypto and the minute the music stopped, the colos emptied out. This is different in one crucial way — the money is coming from a handful of companies with real cash flows, real customers, and balance sheets big enough to keep writing checks even while Wall Street questions them. But that's also the vulnerability. When the spending is that concentrated, the whole boom is one bad earnings season away from a coordinated pullback. You don't need 100 companies to panic. You need four.

The Secondary Bottleneck Nobody's Talking About — The Profit Test

Everyone's obsessed with the visible bottlenecks — power, chips, cooling, transformers, grid interconnection. I've written about all of them. But Kohler's point points at the bottleneck that actually ends booms: the profit test.

Here's the uncomfortable math. The hyperscalers are spending $700 billion a year on infrastructure. AI revenue is growing fast — OpenAI reportedly hit a $40 billion annualized run rate, which is genuinely impressive — but it's still a fraction of the capex being poured into the ground. Microsoft recently disclosed that the overwhelming majority of its customers aren't paying meaningful amounts for AI yet. That's not a knock on anyone's product. It's the structural reality of a boom in its infrastructure phase: you build everything first, then figure out how to charge for it. The telecoms did the same thing in the late 1990s — they laid the fiber, and the fiber did eventually get used. But the companies that laid it went bankrupt before the usage showed up, and the people who bought the assets at fire-sale prices made the real fortunes.

That's the pattern I keep coming back to. The buildout phase rewards capital. The profit phase rewards whoever's left standing. If you're an independent hosting provider, you want to be the one left standing — not the one who bought into the boom at the top of the capex curve.

What This Means for Independent Hosting Providers

So what do you actually do with a "boom is just beginning" forecast? Let me give you the founder's version of the playbook.

First — lock multi-year colo and power contracts while the boom is still in its buildout phase. If Kohler is right that spending keeps growing into 2027 and beyond, power and space only get scarcer and pricier before they get cheaper. The best time to sign a three-year deal is before the next $200 billion capex wave hits the grid. Every month you wait is a month of rate increases baked into your cost base.

Second — don't build speculative capacity on "just beginning." This is the trap. The temptation when a respected analyst says the boom is just beginning is to overbuild — to assume unlimited demand forever. That's exactly how hosting companies died in 2001 and how crypto-mining operations died in 2022. Build for the demand you can see, not the demand you're promised. Hyperscalers can absorb years of negative returns on speculative data centers because they have other businesses. You can't.

Third — watch AI revenue disclosures like your business depends on them, because it does. The signal that the boom is ending won't be a headline. It'll be a footnote in an earnings call — "we're seeing softening demand for compute" — or a hyperscaler quietly trimming guidance. Set alerts for AI revenue and capex commentary from the big four. When the profit test starts getting passed, that's your green light to expand aggressively. When it starts getting failed, that's your cue to conserve cash.

Fourth — position as the capital-light alternative, because that's exactly what the profit test rewards. When hyperscalers hit the profit wall, they raise cloud prices to protect margins — and every price raise pushes customers toward independent hosting. I've seen this play out three times in my career. The boom's bust is the independent host's boom. Keep your pricing honest, keep your service sharp, and let the giants fight over who eats the $1 trillion.

The Bottom Line

Alan Kohler is right about both things at once. The greatest investment boom in history is just beginning — and it hasn't passed the only test that matters. That's not a contradiction. It's the shape of every great boom in financial history: the infrastructure gets built, the hype gets real, and then the profit test separates the businesses that were actually creating value from the ones that were just spending it.

For the hyperscalers, the boom being "just beginning" is a $1 trillion-a-year opportunity. For independent hosting providers, it's a warning dressed as good news. Build for what you can see. Keep your powder dry. And when the profit test finally lands — because it always lands — make sure you're the one holding the assets that actually make money, not the one holding the bill for someone else's boom.

Plan for the beginning. Survive the middle. Own the end. That's the whole game.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: ABC News Australia (Alan Kohler explainer, Aug 16 2026), The New York Times (Aug 14 2026), Deloitte AI data center capex forecast, Dell'Oro Group data center forecast, hyperscaler Q1 2026 earnings disclosures.