Superintelligence, Superclusters, and star engineers: Meta doubles down on AI with $72 billion infrastructure push

Meta AI

Mark Zuckerberg-led Meta is making clear that it's not just chasing AI — it’s betting the house on it. In its Q2 earnings call, the company revealed a significant capital expenditure forecast: $66–72 billion in 2025, up nearly $30 billion from this year. And that’s just the start. CFO Susan Li confirmed the company will “significantly ramp” AI infrastructure spending again in 2026, viewing it as Meta’s strategic moat in the race for top-tier models and product experiences.

What does that money buy? Titan-scale data centers like Prometheus in Ohio (targeting 1 gigawatt of compute) and Hyperion in Louisiana — which Zuckerberg claims could eventually scale to 5 gigawatts. For perspective, that’s enough power to run several million homes. These AI superclusters, along with others still under wraps, are designed to turn Meta into a serious model-building powerhouse. But they’re also raising red flags. One Georgia facility has already strained local water systems, drying up taps in nearby homes.

To match its hardware bets, Meta is going hard on talent too. Superintelligence Labs — its new elite AI unit — is on a hiring spree, offering massive pay packages to lure engineers and researchers. Compensation is now Meta’s second-largest expense after infrastructure.

Zuckerberg also shared his longer-term vision: “personal superintelligence.” Think AI that understands you — accessed through Meta’s smart glasses and VR headsets. It’s a philosophical shift: AI not just as a backend tool, but as a life companion.

Despite Reality Labs posting another $4.5 billion in losses, Wall Street liked what it saw. Meta’s Q2 revenue hit $47.5 billion, and its Q3 forecast of $47.5–50.5 billion beat expectations. The ad business — still Meta’s golden goose — is now powered by AI tools that help brands generate better creatives, translations, and video content.