Only One or Two Hyperscalers Will Win — and Wall Street Is Already Picking Them

Only One or Two Hyperscalers Will Win — and Wall Street Is Already Picking Them

Let me tell you something that's been sitting with me since I watched the clip. A chief market strategist on CNBC, Matt Maley from Miller Tabak, looked straight into the camera and said the quiet part out loud: only one or two hyperscalers will be the winners in this AI race. Not three. Not four. One or two. And the more I dig into the credit markets this week, the more I'm convinced he's not being dramatic. He's being late.

Because the bond market and the credit-default-swap market have been picking winners and losers for months now. They just do it in numbers nobody outside the credit desks bothers to read. I've been telling you for weeks to watch the debt numbers — the Oracle downgrade, the $334 billion in data center debt issued this year, the off-balance-sheet obligations nobody wants to count. This is that story finally showing up on the nightly business news, wearing a suit and a confident smile.

The Maley Thesis — What "One or Two Winners" Actually Means

Maley didn't invent this idea. He just said it in plain English on "The Exchange" on July 31, which is more than most people on Wall Street will do. His argument is the classic consolidation playbook: the entry barriers to hyperscale are stratospheric — billions in infrastructure, research, talent, global data centers — and the network effects are self-reinforcing. The more businesses build on your platform, the more attractive your platform becomes, and the harder it is for anyone else to break in.

That's not a controversial take in cloud computing. AWS, Azure, and Google Cloud together hold roughly 63 percent of the market — 28, 21, and 14 percent respectively in the latest numbers. And here's the kicker: the gap is widening, not narrowing. Google Cloud grew 63 percent year over year, Azure 40 percent, AWS 28 percent. The big are getting bigger, and the also-rans are spending billions just to stand still.

The Credit Market Is the Real Scoreboard

But talk is cheap. The credit-default-swap market is where investors put actual money behind their opinions about who survives. And that scoreboard is brutal right now.

Oracle's CDS — the cost of insuring its debt against default — hit a multi-year high this week at over 215 basis points. At the end of last year it was around 145. That's a 48 percent jump in the cost of protecting against Oracle going bust, in about seven months. Remember, this is the company S&P downgraded to BBB- — one notch above junk — with negative $23.7 billion in free cash flow against $55.7 billion in capex. The market is not stupid. It's reading the same balance sheet I am.

Then there's CoreWeave. Its CDS topped 855 basis points this week — the highest among big tech companies investing in AI. That implies roughly a 50 percent chance of default over the next five years by the standard pricing model. A former crypto miner, junk-rated, free cash flow negative since at least 2022, operating about 50 data centers across North America and Europe. The market is pricing in a coin flip on whether it survives. SpaceX's swaps are around 185 basis points. Nvidia, Meta, Amazon, and Alphabet all hit new highs too — just from a much stronger starting position.

Maley put it perfectly: "The market is picking winners and losers in the AI race, and the number of companies on the negative side is growing." Read that again. The negative side is growing.

The Bond Market Is Saying the Same Thing

If you think CDS is too exotic, look at plain old bonds. BlackRock sold a $12.5 billion bond Monday at a yield of 7.53 percent — one of the highest levels for a blue-chip data center debt offering since the AI borrowing binge started. Seven and a half percent. For BlackRock. That's not a growth story; that's a risk premium.

Amazon — Amazon! — sold $25 billion in bonds this month and got only slightly more orders than there were bonds for sale. The norm used to be four times oversubscribed. Four times. Now it's a coin flip on whether the book fills. Oracle's notes due in 2054 are yielding 7.8 percent, up almost a full percentage point this year. And Apollo's chief economist Torsten Slok asked the question that should terrify every boardroom in this industry: whether the all-in yield climbs to a level where the marginal data center dollar no longer clears its return hurdle, forcing the capex cycle to self-throttle.

Translation: the money is starting to cost more than the machines can earn back.

The Off-Balance-Sheet Shuffle Nobody Wants to Talk About

And here's where it gets really interesting for the "one or two winners" thesis. Look at what Meta is doing. Instead of borrowing the money itself, Meta is finding partners to develop projects and keeping the debt off its own books. BlackRock sold that $12.5 billion bond for a data center in Texas that Meta is developing — BlackRock entities hold 80 percent of the project's equity, Meta keeps about 20 percent and guarantees the lease payments.

That's clever. It keeps Meta's reported debt down while the project gets built. But it doesn't make the debt disappear — it just moves it somewhere harder to see. I flagged this exact pattern weeks ago: the $1.65 trillion in hidden off-balance-sheet obligations nobody wants to count. The more of this financial engineering you see, the more you understand why the credit markets are treating these companies differently. They're not fooled. They know the difference between a real balance sheet and a shell game.

The Counter-Argument — "But Cloud Is Still Growing"

Now, the pushback. Every time I write about this, somebody says: "But Allan, cloud revenue is still growing. Google Cloud grew 63 percent. These are profitable businesses. You're being too negative."

Fair. Cloud growth is real. But here's what that argument misses: the growth is increasingly concentrated in the companies with the strongest balance sheets, and the losers are funding their growth with debt that's getting more expensive by the week. When your cost of capital goes up and your growth requires ever more capital, the math eventually stops working. That's not a prediction. That's arithmetic.

And remember what happened the last time I wrote about this: Meta admitted its own $145 billion capex plan had overshot, Microsoft pulled back on 200 megawatts of leases, and 75 data center projects got blocked in a single quarter. The buildout is cooling even as the borrowing heats up. That combination — less demand, more debt — is exactly the setup that produces consolidation.

What This Means for Independent Hosting Providers

If you're running an independent hosting business, this is not a spectator sport. Here's what I'd be doing right now:

First, watch the CDS spreads, not the headlines. Oracle at 215 basis points, CoreWeave at 855 — these are your early-warning system. When a big borrower's credit protection costs spike, expect supply chain disruption, order cancellations, and fire-sale hardware within two to four quarters. Plan accordingly.

Second, lock your equipment orders now. When stressed hyperscalers start cutting costs, they don't stop buying GPUs — they stop paying for the little stuff, and vendors reprice everything to compensate. Get your hardware commitments locked before the repricing wave hits.

Third, position yourself as the capital-light alternative. Every point of yield on hyperscaler debt is a point of advantage for you. They're borrowing at 7.5 percent to build capacity they may not fill. You're buying what you need, when you need it, with cash. That's not a weakness. That's a competitive weapon.

Fourth, model your pricing against a 10 percent-plus cost of construction capital. If a new-build competitor has to clear a 10 percent hurdle just to break even, you don't have to undercut them — you just have to not raise prices as fast. The pricing floor is rising. Make sure you're on the right side of it.

The Structural Reality — Consolidation Is the Feature, Not the Bug

Here's the uncomfortable truth nobody on the business shows wants to admit: consolidation is not a bug in this system. It's the design. Hyperscale economics only work at scale, and scale only works when there are fewer players. The whole architecture of this industry — the capex, the power deals, the network effects — is built to end with one or two giants standing.

That's not a conspiracy. It's what happens when an industry requires $700 billion a year in capital and the capital starts demanding a return. The market isn't punishing AI. It's just starting to behave like a market instead of a casino. And in markets, the weak get acquired or eliminated, and the strong get stronger.

The Bottom Line

Matt Maley said one or two hyperscalers will win. He's right, and the credit markets have been saying it louder than any talking head for months. The question isn't whether consolidation happens. It's whether you're positioned on the right side of it.

If you're a hyperscaler's customer, start thinking about your exit now — because when the losers get folded into the winners, contracts get rewritten and prices get reset. If you're an independent provider, this is your moment. The giants are fighting over the last seats at the table, and the customers they drop are looking for somewhere to land. Make sure that somewhere is you.

The one or two that survive won't need your business. Everybody else will. And everybody else is a lot of business.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.