Latest / Minds, Bodies, and Terawatts / Meta Spent Almost Every Dollar It Earned Last Quarter on AI, Left With $784 Million and No Buybacks

Transcript

Meta Spent Almost Every Dollar It Earned Last Quarter on AI, Left With $784 Million and No Buybacks · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, the show where the biggest ad business on earth spent its entire allowance in a single quarter. Reuters reported Wednesday, July twenty-ninth, that Meta put almost every dollar it earned into artificial intelligence.
• 0:14
  [Guest] Thirty-one billion dollars on data centers, servers and chips in three months. What was left after every bill was seven hundred and eighty-four million.
• 0:23
  [Host] And a year ago, that same line?
• 0:25
  [Guest] Eight and a half billion. Same quarter, one year apart, and the spare cash is gone.
• 0:31
  [Host] Except Meta isn't broke. It closed June with about ninety billion in cash and securities. Free cash flow near zero is not out of money.
• 0:40
  [Guest] Fair. But that line is the discretionary money, the part that used to buy back stock. This quarter the dividend went out and the repurchases didn't.
• 0:50
  [Host] And the market's verdict?
• 0:51
  [Guest] Shares fell as much as ten percent after hours. If you hold an index fund, a slice of that was yours.
• 0:59
  [Host] Microsoft spent even more that same afternoon and went up. Same buildout, opposite reaction.
• 1:04
  [Guest] Because Microsoft sells that capacity by the hour and books it as revenue. Meta's spending makes its ads better, and ads are the business Meta already had.
• 1:14
  [Host] So the market isn't punishing the spending. It's punishing spending with no tenant.
• 1:19
  [Guest] That part turns out not to be true. On Wednesday's call, Zuckerberg said Meta has had offers to buy its compute at what he called a meaningful premium to what Meta paid for it.
• 1:31
  [Host] People are lining up to rent it and he's saying no?
• 1:35
  [Guest] For now. His argument is that selling the intelligence built on top of the compute earns a better margin than renting the compute out. He'd rather be the software company than the landlord.
• 1:47
  [Host] Why would anyone pay above cost for someone else's used servers?
• 1:51
  [Guest] The unscarcity dot ai piece "The Electron Gap" explains that. The binding constraint isn't silicon, it's megawatts. A chip order clears in months, while grid interconnection for a gigawatt campus runs years, and utility queues in the US are backed up well past the end of this decade. An already-energized data center therefore trades above what it cost, because money can't buy a queue position.
• 2:15
  [Host] So the asset Meta is being marked down for owning is the scarce one.
• 2:20
  [Guest] And the market already knew that. Bloomberg reported on July first that Zuckerberg was building exactly that rental business, and the stock rose by almost the same amount, in a single day.
• 2:32
  [Host] Same move, opposite direction, same month.
• 2:34
  [Guest] The market paid for the promise and charged for the invoice.
• 2:38
  [Host] And is the buildout even what wrecked the margin this quarter?
• 2:42
  [Guest] Mostly not. Susan Li, the chief financial officer, told analysts that two point four billion in legal charges plus the severance bill did that, and that without those two items operating income would have risen nine percent.
• 2:56
  [Host] The severance is the part I keep looking at.
• 2:59
  [Guest] Eight thousand jobs went in May. That severance sits in the same statement as a machine budget that keeps climbing. Payroll down, silicon up, one page apart.
• 3:09
  [Host] Patrick Deglon's line in Unscarcity is that when this much capital moves in one direction, it stops being a forecast and becomes a build order.
• 3:18
  [Guest] This week we learned who fronts the money for it. Shareholders paid Meta for the promise of collecting rent, then took every point of it back when the invoice arrived, on a building its own chief executive won't lease out while the megawatts behind it stay scarce. For more on compute landlords and who ends up owning the infrastructure, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.