Meta's $70B AI gamble hits turbulence as Zuckerberg faces internal revolt and strategy shift

Meta Pivots Strategy After Llama 4 Disappointment

Mark Zuckerberg is orchestrating one of the most aggressive transformations in Meta's 20-year history, pumping at least $70 billion in capital expenditures into AI infrastructure this year as the company races to compete with OpenAI and Google1

1

. The strategic shift follows the disappointing April 2025 release of Llama 4, Meta's open-source AI model that underperformed rivals on coding tasks and complex problem solving1

1

. The setback prompted Zuckerberg to pursue what he calls "personal superintelligence" through a dramatic overhaul that includes recruiting top AI talent with compensation packages reaching hundreds of millions of dollars3

3

.

Meta is now developing a new AI model code-named Avocado, expected to debut in spring 2026, marking the company's biggest departure from its open-source strategy2

2

. Unlike previous Llama iterations, Avocado will adopt a closed-source approach that Meta can tightly control and monetize, aligning with strategies used by rivals Google and OpenAI3

3

. The model aims to achieve performance comparable to Google's Gemini 2.5 upon release and Gemini 3 by summer, according to sources familiar with the matter1

1

.

Training Data Draws on Chinese Technology

The Avocado project is using several third-party models as part of its training process, including Google's Gemma, OpenAI's gpt-oss, and Qwen from Chinese tech giant Alibaba3

3

. This reliance on Chinese technology represents a notable shift for Zuckerberg, who raised concerns on Joe Rogan's podcast in January about Chinese models potentially being shaped by state censorship3

3

. The situation reflects how Chinese open-source models like Qwen have gained ground, with Nvidia CEO Jensen Huang stating that "China is well ahead -- way ahead on open-source"3

3

.

Historically, Chinese companies leveraged Llama to build their own AI models, but the dynamic now appears reversed, with Qwen serving as a reference point for Llama's successor2

2

. Alibaba's Qwen3-based app surpassed 10 million downloads within a week of public testing in November, outpacing initial launches of ChatGPT and DeepSeek2

2

.

Alexandr Wang Faces Zuckerberg Micromanagement

At the center of Meta's AI reboot is Alexandr Wang, the 28-year-old founder of Scale AI who joined Meta after the company acquired a 49% stake in his startup for over $14 billion4

4

. Wang now heads the secretive TBD Lab tasked with building Avocado, but sources say he has privately complained to associates that Zuckerberg micromanagement is "suffocating" progress5

5

. Zuckerberg has positioned himself as a mentor to Wang and spends much of his time working closely with the new hires in TBD Lab3

3

.

Some Meta employees have questioned whether Wang is equipped to manage massive research teams at the company's scale, noting his background is in AI data services rather than developing frontier AI models5

5

. Wang is an advocate of closed models, influencing Meta's strategic shift away from open-source AI3

3

.

Executive Departures Signal Internal Turmoil

Meta has experienced a wave of executive departures as the AI push creates corporate upheaval. Longtime chief legal officer Jennifer Newstead was recruited by Apple, while chief revenue officer John Hegeman announced plans to launch a startup5

5

. Yann LeCun, the company's storied chief AI scientist and Turing Award winner, is departing to start a new AI initiative after reportedly objecting to reporting to Wang and seeing his research priorities slashed5

5

.

The company has also laid off 600 employees from its AI divisions, framing the cuts as necessary to speed decision-making5

5

. Former GitHub CEO Nat Friedman, brought in to integrate AI models into Meta's products, has faced mounting pressure to deliver quickly, frustrating team members who felt products were rushed out to beat competitors5

5

.

Investor Concerns Mount Over Spending

Investor concerns have intensified as Zuckerberg signals AI spending could top $100 billion annually, up from $39 billion the previous year1

1

. At Meta's October earnings, the company failed to provide clarity on how AI technology would be integrated into its social media empire and monetized, triggering a stock drop of more than 10 percent that wiped over $208 billion from its valuation1

1

. The company has undertaken complex financial maneuvering to pay for new data centers and chips, tapping corporate bond markets and private creditors1

1

.

Zuckerberg has pledged to spend $600 billion on U.S. infrastructure projects over the next three years, most AI-related, while redistributing investments away from virtual reality and metaverse efforts toward AI glasses and related hardware3

3

. The year 2026 could determine whether Zuckerberg's vision becomes reality or shatters under pressure, with Avocado's performance critical to retaining the AI talent recruited through what insiders call the "Great Talent Heist"1

1

.