Meta AI Capex: Zuckerberg's Mea Culpa and Strategic Vision

Ben Thompson of Stratechery published a script of what Mark Zuckerberg should say on Meta's upcoming earnings call. In the imagined remarks, Zuckerberg delivers a mea culpa for his long-running obsession with building a platform – an error he traces from the slow shift to mobile through the twelve-figure losses at Reality Labs. The real focus, he says, should have been the ad business all along.

AI now sits at the center of the argument. Zuckerberg warns that every digital business faces an existential threat from AI and that Meta must invest in GPUs and data centers even if it means ballooning capex. The payoff, he claims, will come from a massive expansion of ad inventory. AI makes every pixel monetizable, which he compares to the inventory jumps that followed Stories and Reels – two moments where investors panicked and missed the best buying opportunities.

A second structural change: Meta plans to sell short-term access to a portion of its compute capacity, with the right to claw it back. The rental revenue would fund an even larger build-out. The rental price itself becomes a hurdle rate – if Meta cannot make more money on that compute through ads, it will let someone else use it. Thompson frames this as a discipline mechanism that forces the company to stay focused on what it actually does well, rather than what Zuckerberg wanted it to become.

AlphaScala's Alpha Score for Meta sits at 60, a Moderate rating, with the stock up 2.98% on the day near $600. The score reflects the tension between strong earnings momentum and the elevated capex risk that Thompson's script explicitly addresses.

The script does not offer specific financial targets. It argues, instead, that the company's revealed preference and user data will validate the spend – the same logic Zuckerberg used to defend the feed launch and the mobile shift. Thompson's Zuckerberg admits that investors have been skeptical before, notes that the 2022 capex spike for GPUs looked prescient after ChatGPT, and insists the current build-out will look the same in hindsight.

For a trader building a watchlist on Meta, the key questions are whether AI-driven inventory expansion materializes within the timeline Thompson outlines – a few years – and whether the compute rental plan actually disciplines capital allocation or simply defers a reckoning. The script offers conviction, not numbers, which is typical for a vision statement. The data that will matter comes in the real earnings call, where capex guidance and ad revenue growth rates will either confirm or contradict the thesis.