OpenAI Shutters Sora: A Win for Meta Platforms Amid a Rocky Stretch

Why It Matters

Sora’s failure highlights the high operating costs of standalone AI video services, reinforcing Meta’s strategy of embedding AI within existing social feeds to protect revenue and investor confidence.

Key Takeaways

• OpenAI spent $5.4 billion annually on Sora compute
• Sora generated only $2.1 million revenue before shutdown
• Meta’s Reels contributes ~25% of its $51.2 billion revenue
• Sora’s daily compute cost was $15 million, unsustainable at scale
• AI integration into existing feeds beats standalone AI video costs

Pulse Analysis

The race to dominate short‑form video has accelerated with AI at its core. OpenAI’s Sora launched in September 2025, quickly amassing 3.2 million downloads and topping the App Store, but its all‑AI content model demanded massive inference power. At an estimated $15 million per day, the platform’s compute bill ballooned to $5.4 billion a year, dwarfing the modest $2.1 million revenue it ever produced and prompting a swift shutdown.

For investors, Sora’s demise serves as a cautionary tale about the economics of pure AI feeds. Unlike traditional social apps that monetize user‑generated content, an AI‑only service must shoulder the full cost of rendering each view, a model that quickly becomes untenable as engagement scales. Meta’s Reels, by contrast, blends AI‑enhanced clips with human‑created videos, leveraging its existing ad infrastructure and user base. This hybrid approach allows Meta to capture roughly a quarter of its $51.2 billion annual revenue while keeping AI spend proportionate to incremental value.

Looking ahead, Meta is likely to double down on AI integration rather than launching another standalone product. Experiments such as the “Vibes” app suggest the company will continue testing AI‑driven features within its proven platforms, preserving monetization pathways and mitigating cost spikes. As the broader market watches, the lesson is clear: sustainable AI video growth hinges on embedding intelligence into established ecosystems, not on building isolated, compute‑heavy feeds.