Meta's 2025 talent poaching via MSL nabbed experts like Alexandr Wang with $100M+ deals from OpenAI, driving Llama 4, "Mango" video, and "Avocado" text models for H1 2026. Meta Compute launched January 2026 targets tens of gigawatts this decade via $600B U.S. spend, nuclear pacts, led by Santosh Janardhan/Daniel Gross.
Early 2026, Meta ranks fourth in innovation (behind OpenAI/Microsoft/Google) but leads open-source, powering billions via apps while trailing enterprise monetization. Poaching rebuilt Llama edge; Meta Compute integrates hardware/energy for $250B savings by 2030.
Meta Compute unifies data centers for superintelligence scale and open AI delivery, no commercialization yet amid $10B Google Cloud use. Future hints at energy-industrial moat via Prometheus/Hyperion superclusters, potentially reshaping U.S. power grids but risking capex scrutiny amid Reality Labs layoffs.
This infrastructure race bets compute over algorithms amid energy crunches, accelerating U.S.-centric AI but questioning sustainability. Meta Compute lifts to #2-3 contender, if gigawatts deliver; execution risks talent churn, nuclear delays versus hyperscalers.
COMPUTE MAY WIN AI WARS BUT WHO WILL FOOT THE POWER BILL?