Meta $META ( ▲ 0.32% ) is deep into its all-in AI pivot, and two new reports suggest the transition is getting messy inside the company. The New York Times and Bloomberg both detailed tensions between Meta’s longtime execs and Alexandr Wang, the 28-year-old wunderkind now leading its AI division.
According to the Times, Meta asked CTO Andrew Bosworth to cut 2 billion dollars from Reality Labs and redirect the money to Wang’s group. Reality Labs oversees the metaverse and AR glasses, the same division Meta once rebranded its entire company around. That budget trim aligns with last week’s report of a potential 30 percent cut to Reality Labs, which helped send Meta’s stock higher.
The bigger fight is about what the next AI model should actually be. Bosworth and chief product officer Chris Cox reportedly wanted Meta’s AI team to train its upcoming frontier model using Facebook and Instagram data. Their argument was simple: use AI to strengthen the feeds and the ads business that actually make money. Wang disagreed. He pushed to first catch up to OpenAI and Google before worrying about product integration.
Bloomberg’s reporting adds another twist. Meta is training its next model, code-named Avocado, using techniques borrowed from competitors. The TBD Lab, Zuckerberg’s handpicked team of AI all-stars, is “distilling” learning from other models, including Google’s Gemma, OpenAI’s gpt-oss, and Alibaba’s Qwen. Using a Chinese model like Qwen could complicate Meta’s ambitions to sell AI systems for national security contracts.
Even more significant is the strategic pivot behind the scenes. After years of aggressively championing open-source AI, Zuckerberg is now preparing to shift Meta toward closed, proprietary models. The goal is to turn AI into a high-margin business instead of giving away Llama for free. A closed model would allow Meta to charge enterprise customers, sell developer access, and compete directly with OpenAI and Google Cloud for revenue rather than influence.
In short, Meta is racing to build a superintelligent model, but also a super profitable one. And it is doing so by reallocating billions, borrowing ideas from rivals, and reversing one of the company’s loudest public stances on AI openness.