Alphabet's Porat says next year's AI capex will exceed this year's roughly $200 billion

Porat's phrase "capacity constrained" ^[4] fits two readings. In the first, Alphabet has more orders for compute than it can serve. In the second, demand is real but the return on the next dollar of capex is untested, and the constraint is only the limit of what Alphabet has built so far. We lean to the first, because Porat tied the spending to named customer groups ^[6]. The account gives no utilisation rate, backlog or revenue figure for those groups.

Her only word for the size of next year's rise was "significantly" ^[3]. Against a base of roughly $200 billion ^[1], a 10% rise adds $20 billion ^[10] and a 25% rise adds $50 billion ^[11].

Porat said, "We do see this as an opportunity that extends for quite some time." ^[5] She offered it as a forecast. The $200 billion is a cost already incurred, and the revenue it earns will turn on how much of the capacity customers use.

She also framed the spending as long-horizon. "The thing that is true throughout history is anything that is not built for the long run won't endure for the long run," she said ^[7]. Her case is for judging the spending over years, and it leaves utilisation today open.

We would call the demand thesis wrong if next year's figure comes in at or near this year's $200 billion ^[1], or if the capacity built against it sits partly unused.