Meta Needs AI but maybe not AGI
Why is Meta trying to compete at the AI frontier and pursue “personal superintelligence”? It is a question I return to every earnings call, because Zuckerberg has yet to offer an economically compelling answer.
Meta unquestionably needs AI. Machine learning drives ad targeting, recommendations and engagement, and the results have been fabulous: users spend more time on Reels, advertisers receive better conversions, and Meta’s core grab-eyeballs-and-sell-ads flywheel is kicking a lot of ass. Indeed, European regulators now object that their platforms are becoming too addictive—a ringing implicit endorsement if ever I saw one.
Building much of that technology internally absolutely makes strategic sense. Meta does not want its core products dependent on frontier labs—or, worst case, Google, its largest advertising rival—charging it high rents for essential models.
But none of that explains why Meta needs to own the frontier.
Facebook, Instagram and WhatsApp are innately social products: their principal value comes from interactions amongst us humans. AI can improve discovery, creation, advertising and moderation, but it is not obvious that these tasks require Meta to build the world’s best general-purpose model and deliver superintelligent agents. I am, for example, in a fair few WhatsApp groups, and in none of them are we currently trying to solve cold fusion, although breakthrough prostate inflammation treatments would make some of us happy….
The strongest justification is platform risk. Meta was partially disintermediated on mobile by Apple and Google. If personal AI assistants become the next operating layer—and AI glasses become an important interface (not my bet, but still)—Meta does not want to depend on somebody else’s intelligence. Meta’s official strategy increasingly points in this direction: distributing a personal assistant across its apps and glasses, rather than merely improving the existing feed.
That is plausible, but it still does not establish that winning the frontier race is necessary, rather than remaining sufficiently close to the frontier and exploiting Meta’s unparalleled distribution.
Meta expects approximately $130–145 billion of capex in 2026. It does not disclose how much of that represents incremental superintelligence investment rather than infrastructure supporting ads, recommendations and its established products. That opacity is part of the problem. $50B maybe? Guessing is fun.
I genuinely admire Zuckerberg’s willingness to repeatedly bet Meta’s balance sheet on the future. Tim Cook could have used a little more of this. But willingness to make enormous bets, and judgment about which enormous bets to make, are very different CEO capabilities.
AI is not the metaverse; its importance and commercial usefulness are already far clearer. But the uncomfortable similarity is that Meta may again be spending extraordinary sums to solve a supposed strategic vulnerability without clearly articulating how the resulting product fits its business model. It’s becoming hard to entirely rule out CEO ego, especially when the board effectively has no say.
Perhaps personal superintelligence becomes the next platform, Meta’s apps and glasses become its dominant distribution, and Zuckerberg looks prescient again. But so far, Meta has made a much better case that it needs AI than that it needs to win AGI.
Disclosure: I have no position in Meta, but I would add it in a heartbeat if they would just fully commit to sticking to the knitting of content and advertising.