Meta bet $14.3 billion（約2.3兆円） on Alexandr Wang to build a proprietary AI model; now Zuckerberg must prove it can make money beyond ads.

What happened

Meta spent over $14 billion（約2.2兆円） last year to hire Alexandr Wang and other top engineers from Scale AI to overhaul its AI strategy. Wang's team delivered Muse Spark, Meta's first proprietary foundation model (launched in April), designed to integrate into Facebook, Instagram, Ray-Ban Meta glasses, and Meta's standalone AI apps—a sharp shift from Meta's previous open-source Llama approach.

Why it matters

Meta's stock has fallen 18% over the past 12 months despite 33% revenue growth in the first quarter, because Wall Street wants proof Meta can build an AI-first paid product, not just use AI to improve its ads. Analysts say Meta must show both adoption and revenue from new AI tools; historically, Meta derives 98% of revenue from ads, and non-ad subscription efforts have not succeeded. The stakes are high: Meta has a $200 billion（約32兆円） a year advertising business to protect.

What to watch

Meta says it is testing Muse Spark access with early partners via API and plans to release it this month. The critical test is whether developers, burnt by the failed Llama 4 rollout last year, will trust Meta's proprietary model—and whether Meta can demonstrate a cost or efficiency advantage (e.g., cheaper or faster operation) that competitors like Google do not offer.