Strategic positioning: OpenAI and Meta represent the two dominant poles of AI strategy—closed-source frontier development vs. open-weight ecosystem capture. OpenAI positions itself as the safety-conscious, API-first provider of the most capable models (GPT-4o, o3 reasoning), monetizing through subscriptions and enterprise API consumption. Meta, via FAIR, positions Llama as the "Linux of AI"—free, modifiable, and designed to commoditize the model layer while Meta monetizes through ad targeting, recommendation systems, and hardware (Meta Ray-Ban, Quest). OpenAI's moat is exclusivity and performance; Meta's moat is distribution and zero marginal cost for model weights.

Product and ecosystem: OpenAI's product stack is narrow but deep—ChatGPT (300M+ weekly active users), API (dominant in developer tools), and Sora (video generation). Its moat is brand trust and API stickiness, but it faces existential risk from commoditization: as Llama 4 approaches GPT-4o quality, enterprises may prefer self-hosted, uncensored, and cheaper alternatives. Meta's ecosystem is broader but shallower: Llama 4 powers internal ad optimization (projected $10B+ incremental revenue in 2025), but developer adoption outside of academia remains fragmented. The critical metric is not model benchmarks but inference cost per token: Meta's open weights allow anyone to run Llama 4 on their own hardware, undercutting OpenAI's API margins by 10–20x in high-volume scenarios.

Recent momentum: OpenAI's 569 mentions vs Meta's 159 in the data reflects media dominance, but media mentions are noise. The signal is in talent flows and deployment patterns. Claude Code's viral adoption at MIT and Stanford (7+3 mentions) is a canary: Anthropic is poaching the academic pipeline that once fed OpenAI. Meanwhile, Meta's $60B+ AI spend in 2025 is overwhelmingly inference-side: they are building custom AI chips (MTIA) and optimizing Llama for on-device deployment. OpenAI's recent o3 release (strong on math/coding) signals a bet on agentic workflows; Meta's Llama 4 release (multimodal, MoE) signals a bet on ubiquitous, low-cost inference.

The critical question: Can OpenAI maintain a premium pricing model when Meta gives away comparable models for free? The answer depends on whether OpenAI's "safety tax" and API reliability justify 10x+ cost premiums for enterprise customers. If Llama 4 matches GPT-4o on enterprise benchmarks within 12 months, OpenAI faces a classic innovator's dilemma: defend high margins against a free alternative, or compete on price and destroy its own unit economics. Meta's risk is the opposite: if open-weight models enable catastrophic misuse, regulation could force liability onto model distributors—potentially breaking Meta's open-source strategy entirely.