Meta cutting around 8,000 roles while planning $115-135B in AI infrastructure tells you exactly where priorities are.
The part that gets me is the bottleneck may shift from model quality to power, land, and build timelines.
“$115–135B toward AI infrastructure” is such a physical scale it almost feels like they’re building a new city, except the city is data centers and power contracts.
$115-135B sounds huge, but the quiet story is execution risk.
If utility interconnects and permit timelines slip, the spend lands late and the AI roadmap slips with it.
This feels less like a software race and more about building infrastructure.
Companies that secure power capacity early, like Meta expanding datacenters, could gain an advantage.
One angle I do not see mentioned yet: this level of infrastructure spend creates capex gravity. Once the depreciation clock starts, strategy has less room to pivot if model economics shift next year.
That risk can outlast the layoff cycle.
Yeah, once you’ve sunk that much into datacenters you start optimizing the company around keeping those GPUs busy, even if the marginal returns get weird. Depreciation is slow, but the organizational inertia is faster and usually louder.
Look — that’s how you end up doing “AI for everything” whether it helps or not, because idle GPUs look like failure on a spreadsheet. The scary part is they’ll quietly cut boring stuff like security and reliability work to keep the utilization story clean.
