A year after Mark Zuckerberg recruited Scale AI co-founder Alexandr Wang to inject urgency into Meta's lagging artificial intelligence division, the $1.5 trillion company has released Muse Spark—its most credible AI model yet and a potential sign that the high-stakes bet is beginning to pay off.

The Wunderkind Gambit

Wang, who was 28 when recruited, represents an unconventional choice for one of tech's most consequential AI pivots. Rather than promote from within or hire a veteran researcher, Zuckerberg invested $15 billion into Wang's data-labeling startup and brought the founder aboard with unusual autonomy to rebuild Meta's research organization from scratch. According to interviews with current and former employees, the move reflected growing internal concern that rivals like OpenAI, Google, and Anthropic were pulling further ahead after Llama 4's disappointing reception.

Inside TBD Lab

The secretive unit Wang assembled—known internally as TBD Lab—operates from a secure area of Meta's Menlo Park headquarters requiring special badges to enter. Both Wang and Zuckerberg maintain offices inside the space, with non-TBD staff reportedly caught attempting unauthorized entry on multiple occasions. The group has grown to approximately 100 researchers drawing multimillion-dollar salaries, making it one of the most expensive talent concentrations in Silicon Valley. Wang has fostered a deliberately non-hierarchical startup culture, including regular boba tea happy hours and podcast appearances where he argued that 'the very small team where everyone is cracked is always going to move faster than the large org.'

Early Stumbles and Internal Tensions

TBD Lab's first year hasn't been without friction. Ruoming Pang, a former Apple executive poached to lead research efforts, departed after just seven months for OpenAI—a significant loss given the tight timeline. More pointedly, internal tensions erupted when Wang suggested Muse Spark had been developed 'from scratch' despite using code and datasets associated with Llama 4, according to people familiar with the project. The episode highlighted deepening divides between TBD and Meta's established AI teams, who felt their contributions went unacknowledged.

What's Working—and What Isn't

Muse Spark has earned praise for visual understanding capabilities but acknowledged weaknesses in coding tasks—several employees tasked with testing the model reportedly continued preferring Anthropic's Claude for software development work. The model was trained using some third-party open-source models, including Chinese ones, with insiders drawing comparisons to DeepSeek's latest releases. Wang has also advocated shifting Meta away from its longstanding open source approach toward proprietary models, a significant philosophical pivot that hasn't been universally embraced.

The Road Ahead

Future TBD Lab models are expected to prioritize coding improvements, agentic task completion, and advanced multimodal capabilities including video generation—successors reportedly launching within months. 'It was a rough start for him to find his power at the company,' said one associate. 'But he's found his groove.' Carnegie Mellon professor Russ Salakhutdinov, Meta's former VP of AI research, offered measured endorsement: 'Alex knows what he doesn't know and he's willing to listen. The amount of work TBD Lab was able to do in a short amount of time is very impressive.'

Key Takeaways

  • Zuckerberg invested $15 billion in Wang's Scale AI and hired the 28-year-old founder to lead Meta's AI revival
  • Muse Spark represents TBD Lab's first major release, praised for vision but trailing rivals in coding
  • Internal tensions persist over credit for the model and questions about incremental versus transformative progress
  • The secretive unit operates from a secure Menlo Park facility with roughly 100 researchers on premium salaries

The Bottom Line

Meta's AI turnaround is real but incomplete—Muse Spark shows promise while exposing how far the company still needs to climb. Wang has delivered results faster than skeptics expected, yet the gap between 'impressive for less than a year' and 'leading frontier model' remains substantial. If TBD Lab's next releases don't close that distance decisively, this chapter reads less like a comeback story and more like expensive groundwork.