The Three Biggest Cloud Companies on Earth Are Spending 102% of Their Cloud Revenue — and That Number Changes Everything

The Three Biggest Cloud Companies on Earth Are Spending 102% of Their Cloud Revenue — and That Number Changes Everything

I've been running hosting infrastructure for over a decade, and I've watched a lot of spending cycles come and go. But let me tell you something — I've never seen anything quite like what UBS put on paper this week. Amazon, Alphabet, and Microsoft are on track to spend 102% of their combined cloud revenue on capital expenditures this year. Let me say that again, slowly: they're spending more than the entire cloud business generates. Every single dollar of cloud income gets recycled into AI infrastructure — and then some.

That's not a typo, and it's not a headline writer being dramatic. It's the single most important number in infrastructure right now, and almost nobody outside the finance world is talking about what it actually means for the rest of us.

The Setup — What the 102% Actually Means

Let's get the accounting straight first, because the nuance matters. When UBS says Amazon, Alphabet, and Microsoft will collectively plow 102% of their cloud revenue into capex in 2026, it does not mean those companies are losing money. These are enormous, diversified businesses — Amazon sells groceries, Alphabet sells ads, Microsoft sells software licenses. Their cloud divisions are profitable on a standalone basis, and the broader companies generate plenty of cash.

What the ratio actually tells you is where the money goes. Every dollar of cloud profit — plus a little extra pulled from elsewhere in the balance sheet — is being funneled straight back into data centers, GPUs, and networking. The cloud isn't being run as a business anymore. It's being run as a construction fund for the AI buildout.

Ent? Think about what that means if you're the one paying the cloud bill every month.

The Numbers — a $4.1 Trillion Bet in Three Years

The scale is the story. UBS projects hyperscalers will spend roughly $4.1 trillion on AI infrastructure from 2026 through 2028 — more than three times the $1.3 trillion deployed across the previous six years. The year-by-year path: $492 billion in 2025, $1.009 trillion in 2026, $1.447 trillion in 2027, and $1.619 trillion in 2028. The growth rate slows after this year, sure. But the dollar amount keeps climbing past a trillion and a half a year — and that pattern, UBS argues, is what separates this cycle from a normal technology upgrade.

Break it down by company and it gets even more vivid. UBS's cumulative three-year estimates: Alphabet at about $938 billion, Meta Platforms at roughly $683 billion, Microsoft at about $672 billion, and Amazon at around $628 billion. SpaceX and Oracle add an estimated $335 billion and $276 billion. CoreWeave and Nebius are in there too. Moody's has the broader group at $785 billion for 2026 and close to a trillion in 2027. Morgan Stanley says the five largest cloud companies will spend $805 billion this year and $1.1 trillion in 2027 — more than the GDP of Saudi Arabia.

And the ratio easing to 99% next year and 94% in 2028 isn't discipline. It's math — the denominator finally starts catching up. The spending itself does not slow down.

The Two Readings — Platform Shift or Bubble Math

Here's where I get honest with you, because there are two ways to read this, and both of them are true at the same time.

Reading one: this is what a platform shift looks like. When Amazon built AWS, it spent ahead of revenue for years. When Microsoft bet the company on Azure, same story. The hyperscalers are doing what they've always done — building the railroad before the trains exist. If AI is genuinely a once-in-a-generation platform shift, then spending 102% of cloud revenue for a couple of years is the price of entry. The companies that convert that infrastructure into recurring AI revenue will own the next decade.

Reading two: this is what bubble math looks like. Citi projects combined 2027 capex of $801 billion for Alphabet, Meta, and Amazon — with all three expected to post negative free cash flow in 2027 and 2028. Forbes counted $3.1 trillion in off-balance-sheet commitments backing the buildout — roughly five times annual capex. Epoch AI's tracker says aggregate cash capex is on track to exceed aggregate operating cash flow in the third quarter of this year. And a data center does not generate attractive returns merely because it holds expensive GPUs. Capacity has to stay utilized. Customers have to pay. AI services have to produce enough revenue to cover depreciation, electricity, financing, and operating costs.

I've seen this movie before, and the difference between a shift and a bubble is exactly one thing: whether the revenue shows up before the bill does.

The Part Nobody's Talking About — the Depreciation Wall

Here's the number I keep coming back to, and it's not the capex. It's the depreciation.

All that spending lands on the balance sheet as property, plant, and equipment — then it runs through the income statement over multiple years as depreciation. The pace of that is a management estimate, and it's the one dial nobody is watching. Amazon's trailing four-quarter capex already exceeds its trailing four-quarter operating cash flow — 102%, by one analyst's count. The depreciation drag is the 2027-2028 risk nobody's pricing into anything.

When that wall hits, the hyperscalers have exactly two levers. They can stretch depreciation schedules — which is accounting, not economics — or they can raise prices. And that is where the bill lands on you.

The Secondary Bottleneck — the Bill Always Comes Due

Let me connect this to the thing I actually run. When the three biggest cloud companies spend more on infrastructure than their cloud businesses earn, that money does not come from nowhere. It comes from debt markets, from investors, and eventually — it comes from the customers who can't leave.

Look at the egress data. Per-gigabyte transfer prices have barely moved since 2020 — AWS still charges about $0.09 a gigabyte, Azure $0.087, Google up to $0.12 — while the volume of data flowing through those pipes has grown three to seven times. A mid-market SaaS company pushing 50 terabytes a month out of AWS is paying $54,000 a year in bandwidth alone. That's the hidden transfer tax, and it's exactly where the AI buildout's bill is going to show up first. Egress does not get cheaper when a provider is trying to cover $1.6 trillion a year in capex. It gets more creative.

Watch the AI-inference line items on your next cloud bill. Watch the compute instance prices. Watch what happens to the startup credits and free tiers when margins compress. The 102% ratio means the cloud's profit center is now a loss leader for the buildout — and loss leaders eventually get their money back from somebody.

What This Means for Independent Hosting Providers

If you run an independent hosting business, this is the most important economic signal you'll get all year. Here's what I'd do.

First — watch the depreciation line, not the capex headline. Capex is the sales pitch. Depreciation is the bill. When the D&A number starts moving through the income statement in 2027-2028, that's when cloud pricing strategy changes. Position your business ahead of that, not behind it.

Second — expect cloud prices to rise, and be ready to catch the refugees. The 102% ratio has to be paid back. When AWS and Azure raise prices, the first people to move are exactly the customers you want — the ones who were only on the cloud because the free credits made it look cheap.

Third — lock your hardware and colo now. Supplier pricing is still distorted by the buildout. When the hyperscalers' cost pressure mounts, they don't absorb it — they pass it upstream to component makers and downstream to customers. Get your capacity priced while the window is open.

Fourth — be the capital-light alternative, and say it out loud. You don't have to amortize $1.6 trillion a year. Your incremental cost of serving a customer is your real cost, and it's a fraction of theirs. That's not marketing — that's math.

Fifth — don't bet your business on cloud credits or startup programs. When margins compress, the freebies are the first thing to go. Build your stack on something you control.

The Bottom Line

Spending 102% of revenue on infrastructure is either genius or insanity, and the verdict won't come from a press release. It'll come from a depreciation schedule and a pricing email.

Me? I'm not picking a side. I'm just building my business around the one thing that's guaranteed: when the biggest players on earth are spending more than they earn, the bill eventually gets passed to somebody. Make sure it isn't you.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: UBS via 24/7 Wall St. and PrimeXBT, Moody's, Morgan Stanley, Citi, Forbes, Epoch AI, Silicon Analysts, AWS/Azure/Google Cloud egress pricing data.