Meta’s shifting AI strategy causes internal confusion

Meta is overhauling its artificial intelligence plans as it shifts attention from its open-source Llama models to a new frontier system called Avocado, creating uncertainty inside the company.

Chief Executive Mark Zuckerberg previously promoted Llama as a foundation for broad industry progress. But mentions of the brand have faded as Meta pursues a proprietary successor. People familiar with the effort said employees expected Avocado to launch in late 2025. It is now planned for the first quarter of 2026, while teams continue performance testing. Meta said its training schedule remains on track.

The transition comes after Meta spent $14.3 billion to hire Scale AI founder Alexandr Wang and top researchers. The investment contributed to higher capital-spending guidance for 2025, raising Wall Street’s focus on returns. Analysts said Meta entered the year as a perceived AI leader but now faces questions about spending and results.

Llama 4’s weak reception accelerated internal debate over open source. Some employees were frustrated that DeepSeek’s R1 model adopted elements of Llama’s architecture, heightening concerns about risks. Avocado may not be released openly, a shift from Meta’s earlier posture.

The leadership shake-up following Llama 4 included removing long-time executive Chris Cox from oversight of the GenAI unit. Zuckerberg installed Wang as chief AI officer and head of the TBD Lab, where Avocado is being developed. Former GitHub CEO Nat Friedman and other senior hires have introduced new development methods that differ from Meta’s traditional processes.

Rival model releases have intensified pressure on Meta. Google’s Gemini 3, OpenAI’s GPT-5 updates and Anthropic’s Claude Opus 4.5 have drawn strong interest. Nvidia noted that major model developers rely heavily on its GPUs but did not reference Llama on its recent earnings call.

Inside Meta, long hours, restructuring and an October layoff of 600 employees have added strain. The cuts contributed to chief AI scientist Yann LeCun’s decision to depart.

Meta is also shifting resources toward AI products and away from some metaverse initiatives. The company continues to test outside AI models in products like Vibes and is expanding use of third-party cloud providers while building the Hyperion data center.

Zuckerberg told investors in October that Meta’s new AI labs are making progress and that the company is committed to competing at the frontier.