Turbulências na Meta: Desafios Internos e Novos Modelos de IA

Turmoil at Meta as New AI Models Develop

Meta Platforms, formerly known as Facebook, finds itself in a turbulent phase as it navigates the complex landscape of AI model development. The company is struggling to keep pace with emerging technologies while confronting significant internal challenges, such as leadership conflicts, talent attrition, and delays in launching new AI models like the Avocado.

Current Challenges at Meta

Leadership Conflicts and Restructuring

One of the most pressing issues at Meta is the restructuring of its AI division, marked by conflicts between new and existing leadership. In June 2025, Mark Zuckerberg appointed Alexandr Wang, only 28, as the new Director of AI, stirring division among employees. While new appointees like Wang and Nat Friedman push for innovation, veterans such as Chris Cox, removed from AI leadership, and Andrew Bosworth see their roles diminished.

The AI division has been reorganized into four teams:

TBD Lab
Product
Infrastructure
FAIR

This restructuring has resulted in confusion regarding project ownership and communication lines.

Talent Loss and Attrition

Meta has witnessed alarming talent departures, significantly impacting its AI capabilities. Key exits include:

Jitendra Malik: Former VP, now at Amazon.
Yann LeCun: Chief AI Scientist, who criticized Meta’s strategy on large language models (LLMs) and started his own venture.
A group of engineers from Superintelligence Labs (MSL), including Bert Maher and Tony Liu, left shortly after the team’s establishment in November 2025.

Recruitment continues to be problematic, with further losses, such as Avi Verma and Ethan Knight, returning to OpenAI.

Resource and Cultural Pressures

The internal culture at Meta has shifted dramatically, transitioning to a “show, don’t memo” ethos, which includes tighter deadlines and workloads exceeding 70 hours weekly. Additionally, salary disputes and computing resource limitations have hampered operational efficiency.

AI Model Development and Delays

Amid these challenges, Meta plans to release two new AI models.

Avocado Model

The Avocado model represents a new frontier for Meta but has faced significant setbacks, delaying its launch from late 2025 to the first quarter of 2026 due to performance issues during model training. Unlike the open-source model Llama, Avocado is intended to be proprietary. This strategic pivot comes after the disappointing outcomes associated with Llama 4, which was released in April 2025.

Successor to Llama

Additionally, the successor to Llama has also been postponed. Meta is reconsidering its strategy, moving toward a closed model due to risks associated with open-sourcing superintelligent models. Mark Zuckerberg confirmed this shift, emphasizing the need for caution in development.

Broader Implications

With an investment of $14.3 billion in new AI leadership and more than $70 billion allocated for capital expenditures in 2025, Meta is navigating a challenging landscape compared to competitors like Google and OpenAI, who have heavily invested in AI infrastructure and proprietary chips.

Investor concerns are escalating as they grapple with execution risks, talent loss, and uncertainty regarding return on investment (ROI). A recent study indicates that 53% of investors expect a return from AI within six months, contrasting with only 16% of CEOs who share that sentiment, creating a climate of mistrust.

“53% of investors expect a return in AI within six months, while only 16% of CEOs share the same expectation.”

Conclusion

As Meta grapples with its leadership challenges in AI, it faces a critical juncture in its quest for AI leadership. The company must address internal strife and mitigate talent loss swiftly to ensure that its AI development trajectory is not compromised. How effectively Meta navigates these obstacles may ultimately shape its future in this vital and competitive sector.

Sources

Share this post

Like this post? Share it with your friends!