Google’s capital spending forecast moved to as much as $205 billion. Last quarter, the top end was $190 billion. The bottom of the new range, $195 billion, sits above what the company previously called its ceiling.
That’s the part investors should read twice. Not the size of the number. The fact that it moved.
You can shrug at $15 billion. Plenty of people did. But strip away the scale and what Google communicated is that it can’t accurately forecast its costs, and companies that can’t forecast costs tend to surprise you again later.
The math doesn't close
Google is spending more money than it’s making. That’s the whole sentence. You don’t need a finance background to see the problem.
Now stack the pricing situation on top. Google faces competitive pressure from Chinese AI tools and pricing pressure to keep the cost of its models low. So spending climbs while prices stay flat or fall.
Spend more, collect the same. Or spend more and collect less. Neither version ends well.
Three other things went wrong at once
People seem to have finally noticed that SpaceX sucks. As of this writing its shares are worth almost half what they were at peak.
Investors are also nervous about Oracle’s datacenter buildout debt, which matters more than it sounds like it should. Oracle is the public market’s stand-in for OpenAI. When traders can’t short the private company, they take it out on the proxy.
And Nvidia has been in rounds of deal talks worth a combined three-quarters of a trillion dollars. Nvidia sits at the center of the circular financing here, more than OpenAI does. When the chip supplier keeps funding its own customers, that’s not always a sign of strength.
What a $250 billion guarantee actually signals
Nvidia guaranteeing OpenAI’s debt is a deal worth $250 billion. Read it as a demand signal if you like. Some people who watch this for a living read it differently.
The arrangement is “as much a reminder of funding strain in the AI build-out as it is a demand signal,” Billy Leung, Global X Management’s tech sector investment strategist, told Bloomberg.
Funding strain. Not demand. That’s a strategist at a firm with money in the space saying the quiet version out loud.
The China problem nobody has solved
A Chinese start-up released a new model, and people get nervous every time that happens. The nervousness has a specific source.
China’s biggest constraint is that, at least theoretically, its companies don’t have the same access to GPUs that US companies do. Their AI systems are competitive anyway.
Follow that thread and it goes somewhere uncomfortable for Nvidia. If competitive models don’t require the GPU volumes everyone assumed, there’s an end in sight to the chipmakers’ cash bonanza. It may also mean companies are building too many data centers.
Even the optimists expect an overbuild
I’ve talked to a lot of smart people who are more bullish on this than I am. I’ve been asking how AI companies plan to actually make money for three years now, and I still haven’t gotten a satisfactory answer.
Here’s what those bulls say when you press them. We will likely overbuild data centers during this period of exuberance. A lot of AI companies will die when the correction comes.
They’re invested anyway, on the theory that the survivors will pay for the failures. That’s the actual bull case. It has a pile of corpses built into it.
Watch the earnings calls this week
Meta, Amazon and Microsoft all report this week. Plenty of people expect them to announce they’re spending more than expected on the data center buildout too.
These pressures aren’t Google’s alone. They run through the entire AI sector, which is why one guidance revision moved the conversation this much.
The boosters are watching for the market top like everyone else. They know it’s coming. Nobody can time it in advance, and some investors are clearly getting cold feet and moving money elsewhere.
We may get numbers from the other giants that calm everyone down, and this stretch of AI anxiety passes. If you’re hunting for top signals instead, Elon Musk is a reliable one. SpaceX did just go public.
