Divide the capex midpoint by the share gain and each of those two percentage points carries $68.75 billion of spending [1], which is a crude figure and worth saying so before anyone repeats it: the $137.5 billion is this year's guidance [6] and the two points were banked through a quarter reported on July 30 [1], so the sum sets a forward bill against a backward receipt. It is still the first output figure in this argument that can be set against a spending figure at all, and that is the thing that has changed.
The persuasive work in the Bernstein note is done by the spread rather than the level. Twenty-seven percent against minus one is 28 points of separation [2], and a gap that wide survives most quibbles about modelling method. What it does not survive is scrutiny of the comparison group: Google's network is the third-party inventory business, while the same note says Meta is on track to surpass Google Search this year [2][4], a different and far larger denominator. The two-point share figure is measured across the platforms Bernstein tracks [3], which makes it the sturdier of the two claims, or rather the one I would actually underwrite.
The other side of the ledger is where it gets awkward. Meta is the only one of the big four without a public cloud, and Zuckerberg's remark that one was in the works has been followed by nothing [8], so where Amazon, Alphabet and Microsoft can rent surplus compute to outsiders [8], every Meta accelerator either improves ranking inside a business that is more than 95% advertising [5] or sits there depreciating. That is the allocation decision, stated plainly.
Hatch is the proposed second line, and its price tells you how it is meant to work. At as much as $199.99 a month, or $2,399.88 a year [13], covering one year of capex at the midpoint would take roughly 57 million subscribers on the top tier [3]. Bank of America's stated case is different: slow adoption feeds subscription revenue and sharpens content and ad targeting [16], with the global user base as the adoption advantage and differentiation from ChatGPT and Claude still to be demonstrated [11][15]. Jeff Marks' line that the jury is out on new revenue streams [14] is the honest position.
This is probably wrong, but the 12% drawdown [7] looks less like doubt about the 27% and more like a bet on its deceleration. The growth could be mostly pricing and mix, which the reported figures do not decompose [1][3]. The minus 1% network line is the easiest comparison in the set to reverse [2]. Or the settlement safeguards trim the inventory the gains ran through [9], in which case the up-to-$18 billion is the cheap part, since Cramer's over-ten-times framing implies more than $180 billion of avoided exposure [5] against a settlement worth about 13% of a single year's capex [4]. Disproof of my read would be next year's guidance rising again while the spread narrows toward, say, 15 points, which would mean the next two points of share cost more than these two did.
