The build is ahead of the payoff

The candid line came from inside the room. On July 2, in an internal Meta town hall whose recording was obtained by Reuters, Mark Zuckerberg told employees that AI agent development over the previous four months 'has not accelerated in the way we expected,' and that May's sweeping reorganization 'wasn't as clean as it could have been' and 'hasn't come to fruition yet.' He said the meaningful benefits are still three to six months out.

The May reorganization "wasn't as clean as it could have been" and "hasn't come to fruition yet." — Mark Zuckerberg

What makes this worth noting is who is saying it. External skeptics have called the agent hype overdone for two years; the discontinuity here is that the builder itself is conceding the payoff is behind schedule — while the spending is already committed. Meta has guided its 2026 AI infrastructure budget as high as $145 billion, up from a range topping out near $135 billion in April, and that reorg cost roughly 8,000 jobs, about a tenth of the workforce, with some 7,000 people reassigned to AI teams.

The timing tells the rest. Days earlier, Meta's stock jumped about 9 to 10 percent on a report that it plans to sell its surplus AI compute as a cloud service — a new unit Zuckerberg framed as a 'natural hedge against overinvestment.' So within the same 48 hours the market cheered the glut of AI capacity, and the CEO privately explained why that glut exists: capacity was built for a payoff that isn't arriving on time. The admission is a data point the whole hyperscaler capex debate has been waiting for — the first from someone signing the checks.