There's Clearly No Salary Cap on Big Tech CapEx... Yet.
![](https://spyglass.org/content/images/thumbnail/108074114-1733867859655-gettyimages-2188101075-776246284_07_qzq6g7gf.jpeg)
Meta is saying it will spend $65B in 2025 from a CapEx perspective. Google is coming in at $75B. Microsoft is "good for" $80B. Amazon?
Amazon said Thursday it plans to boost its capital expenditures to $100 billion in 2025, as it continues its investments in artificial intelligence.
The capex figure exceeds last year’s spending of roughly $83 billion. Amazon CEO Andy Jassy had predicted in October that the company’s 2025 capex would surpass last year’s figure, primarily driven by growth in generative AI.
$100B! We have a winner! Those who thought we might see some sort of cap or even pull-back on spend, especially in the wake of DeepSeek, have thus far been wrong. Big Tech is going to the moon.
And whereas the other three had spend in the order of their market caps:
- Meta, the runt of the Big Tech litter at a mere $1.8T
- Google, the middle manager at $2.35T
- Microsoft, old man antitrust at $3.1T
Amazon is a bit of an outlier here. At $2.5T (though lower once the stock opens today after a mixed earnings forecast), you'd think Amazon would put their CapEx somewhere around $75B - $80B – right in between Google and Microsoft. But per above, last year they already spent $83B and this year, well, we have the “once-in-a-lifetime type of business opportunity,” as Andy Jassy noted. So there was no going backwards. And to be fair to Amazon, they also have other CapEx needs. So do the other businesses, of course, but Amazon perhaps even more so given their logistics and and physical networks.
Still, “the vast majority of that CapEx spend is on AI for AWS," noted Jassy.
So let's bust out the calculator (ChatGPT): these four companies are now planning to spend $320B on CapEx in 2025. By my count (well, ChatGPT's count), there are only around 50 companies in the world that are valued more than $320B. Yes, entire corporations. It's a massive amount of money.
But it's also not as much as the $500B which OpenAI, SoftBank, and Oracle are promising to spend on the 'Stargate Project'. Of course, that's over many years – and they may or may not hit that actual amount. For now, a "mere" $100B seems within range. You know, the amount that Amazon alone is spending on CapEx this year...
Perhaps the most interesting element of 'Stargate' remains that Microsoft is only involved as a "technology partner" (read: they're getting an executive producer credit to use their name). Of the $80B they have earmarked for CapEx in 2025, exactly $0 is going towards 'Stargate', as Satya Nadella wanted to make very clear to everyone, but especially to Wall Street.
And to be fair to Microsoft, they're already basically "eating for two" as it were, with their own CapEx needs and those of OpenAI – the sort of ceremonial member of Big Tech, the little brother, growing up fast. Because they have no real cap to ex, they have to keep a hand in Microsoft's purse. But SoftBank is seemingly about to open up their first checking account.
Back to DeepSeek, again, a lot of people seem surprised at this $320B number because one of the lessons from China's "breakthrough" was supposed to be that similar performance can be had on the cheap. Of course, it's not actually clear just how cheap it was gotten. Certainly cheaper than tens or hundreds of billions – but they also had the AI breakthroughs of Big Tech to piggyback off of. That plus the fact that a lot of this 2025 CapEx spend was undoubtedly already committed to things such as data center build-outs means that there probably wasn't a ton of wiggle room to pull back even if Big Tech wanted to. And I don't think they want to pull back because they clearly wish to follow the lesson of Jevon.
But really, they all follow the lesson of Wall Street. Which is: spend until we say stop. Then stop spending. Meta has already learned this lesson once in the age of the metaverse (so, a few years ago). I'm still certain that Google, Microsoft, and Amazon will have to learn it at some point in the future as well. This quarter's earnings brought a taste of it, but it's more just a reminder that Wall Street is there, watching.
One more thing: There are are least two members of the current crop of Big Tech missing from the above. One is NVIDIA because, of course, they are the company which is the beneficiary of a ton of this CapEx spend. Said another way: Big Tech's CapEx spend is why NVIDIA joined the gang and is sitting pretty right now at a $3.15T market cap, even after a recent pullback due to, yes, fears of what DeepSeek might mean for AI spend.
The other is the king of the hill: Apple. At $3.5T, if they were more in line with their peers, you'd think they'd be spending well north of $100B in CapEx this year. Instead, they're likely to be closer to $12B. That's not a typo. They've gone from $9B two years ago, to something like $11B last year. And extrapolating out from the Q1 numbers this year, you get to around $12B.
We all know that Apple isn't playing the same AI game as the rest, choosing instead to offload a lot of work and capabilities to partners, but it's staggering just how much less money they're spending on it relative to their peers. Perhaps this is prudent! And Wall Street certainly rewarded them post-DeepSeek, while dinging basically everyone else. But it may also be foolish. I'm not saying they have to spend $120B on AI, but you know, maybe more than a tenth of that (much of which likely isn't even related to AI, but their other CapEx needs)?
![](https://spyglass.org/content/images/thumbnail/Canary-in-Data-Center--1--1.png)
![](https://spyglass.org/content/images/thumbnail/maxresdefault-1.jpg)
![](https://spyglass.org/content/images/thumbnail/mgs22_a_nuclear_blast_with_digital_bits_--ar_21_--profile_zeg_f7dcabeb-2e9f-46f9-a963-7e94d70fb2be_0-3.png)