Zuckerberg wants to sell personal AI assistants to more than a billion people
Three things from Meta’s earnings call caught my attention:
1. Meta hasn’t yet reached the limits of what AI can do within the company
Meta is testing whether its new Muse Wonder model can autonomously improve its own advertising and recommendation systems. Zuckerberg says they’re still seeing incremental improvements from using more computing power. That’s a positive signal for expanding AI infrastructure.
2. Muse is too costly, but Meta wants to make it available to a billion people
Meta can’t make Muse available to more users because it doesn’t have enough computing power, even though it has already invested hundreds of billions of dollars in AI infrastructure. Meta expects more than a tenfold improvement in computing efficiency within a year. It believes this is the only way to fulfill its vision and make Muse available to a billion users.
3. New non-voting Class C shares to fund capital expenditure, with the forecast raised again by $5 billion
To fund AI development, Meta introduced a new class of shares that it will use instead of debt, following Google’s example. During the call, they mentioned several times that the market essentially doesn’t trust AI enough to let them raise as much debt financing as they need. That’s why they’re working hard to make their AI software profitable and looking for new ways to fund it. Given negative free cash flow in Q3 and capital expenditure of $135–145 billion in 2026, I see this as preparation for an even larger investment cycle.
