Zuckerberg pivots Meta toward paid AI as open source strategy fades

Zuckerberg shifts Meta AI strategy

Reporting indicates Mark Zuckerberg has become more involved in day-to-day AI decisions as Meta reorganizes to prioritize revenue-generating AI work. Instead of focusing primarily on open-source models, Meta is now pushing toward AI systems that customers pay for.

The shift marks a major change for a company that spent years promoting open access AI. It also places Meta more directly in competition with rivals that already sell controlled AI models.

A pricey new AI direction

Multiple reports say Meta recruited a high-paid AI team and personally courted researchers with multiyear pay packages that in some cases reached into the tens or hundreds of millions of dollars.

Much of that effort is now focused on building AI models that can eventually generate direct revenue, rather than being freely shared with developers and researchers.

Avocado signals closed model plans

Meta is developing a next-generation model codenamed Avocado that Bloomberg reports may debut in spring 2026 and could be made available only through controlled access rather than full public release.

A closed approach would allow Meta to tightly control access and sell usage to customers. That would represent its biggest move away from open source AI to date.

Llama disappointment sparked changes

Llama 4 received mixed reactions from developers, and some reviewers raised concerns about its real-world performance and benchmark reporting, which contributed to internal changes at Meta.

Some employees tied to the project were sidelined, and Zuckerberg began rebuilding the AI organization with a much more aggressive and centralized approach.

TBD Lab takes center stage

Zuckerberg now spends much of his time working closely with a new internal group called TBD Lab. The team sits physically near his desk at Meta headquarters.

The setup allows the CEO to closely monitor progress and stay deeply involved in decisions tied to Meta’s most important AI products.

Training pulls from rival models

People familiar with the work say the TBD Lab team has used outputs from several third-party open source models during Avocado’s development, including models from Google, OpenAI, and Alibaba.

The approach reflects how competitive the AI race has become, even as Meta previously warned about relying on technology tied to China.

China comments contrast current approach

Earlier this year, Zuckerberg raised concerns about Chinese AI models and potential state influence. He argued that open source US models were critical.

Now, Meta is training its next model using Chinese technology, highlighting how competitive pressure has reshaped earlier positions.

Alexandr Wang gains major influence

Alexandr Wang is playing a central role after Meta invested about $14.3 billion in his firm, Scale AI, and recruited him to help lead the company’s advanced AI efforts.

While not a traditional AI researcher, Wang is seen internally as a fast-moving operator with deep industry connections, and his influence on the team is significant.

Open source quietly deprioritized

After Llama 4, Meta leadership directed some employees to stop publicly discussing open source and Llama products. The company reassessed its stance and refocused its messaging.

The move signaled that open source was no longer central to Meta’s AI identity, even if some open models remain available internally or for limited purposes.

LeCun exit underscores shift

Yann LeCun has confirmed he plans to leave Meta to start his own company. Reports and public comments indicate disagreements over research direction, and resourcing were among the factors that preceded his decision.

Before his exit, employees were encouraged to keep him out of the spotlight as Meta’s strategy shifted toward closed and revenue-generating AI models.

AI spending

Meta has said it will materially increase capital spending on AI, including plans that could reach about $65 billion in 2025, and has discussed multi-year infrastructure investments totaling tens or hundreds of billions of dollars.

Some investors worry the costs may take years to pay off, even as Meta argues that AI already boosts its advertising business and overall platform value.

Superintelligence message raises concerns

Meta’s stated goal is to achieve superintelligence, meaning AI systems that outperform humans at many tasks. The term has generated attention and concern both internally and externally.

Internal research found the term sparked fear among regulators, especially in Europe, adding complexity to Meta’s messaging and public relations strategies.

Explore how Zuckerberg’s high-stakes hiring spree is stirring tension inside Meta’s AI teams.

High stakes pivot still unfolding

Meta’s pivot toward paid AI is still in its early stages, with few public results to date. Pressure is rising as competitors release their own AI models quickly.

Success will depend on execution, leadership, and trust from users, investors, and regulators.

See why Microsoft’s latest move could spell fresh trouble for Zuckerberg and Meta.

What do you think about Meta’s shift away from open source AI? Share your thoughts.

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• $8 Billion privacy lawsuit pits Meta investors against Mark Zuckerberg
• Mark Zuckerberg fires shots at Altman as Meta eyes AI superintelligence lead
• Elon Musk and Mark Zuckerberg say Smartphones are fading, but Tim Cook Says The Best is Yet to Come