To The Moon

A year after Mark Zuckerberg appointed Alexandr Wang to put Meta Platforms, Inc.'s artificial intelligence efforts on a "wartime footing," the $1.5 trillion company has unveiled Muse Spark, its most credible AI model to date.

Zuckerberg entrusted the mission of revitalizing Meta's AI to a 28-year-old startup founder at the time, rather than a seasoned researcher. He bet that an outsider's sense of urgency and ambition could succeed where the company's established AI organization was struggling.

Based on interviews with current and former Meta employees and those close to Wang, the billionaire prodigy is now beginning to show results, while also navigating criticism of his inexperience, challenges in early research, and the intricate internal politics of working at a tech giant.

In nearly twelve months, Wang has assembled an elite research team with multi-million-dollar compensation packages, reshaped parts of Meta's AI operations, and has become one of the most influential executives internally. He was the sole Meta leader to accompany Zuckerberg to a White House dinner hosted by then-President Donald Trump and to meet with top Silicon Valley figures last year.

In April, Meta also released Muse Spark, the first major model to emerge from the secretive research team, TBD Labs, that Wang leads.

Supporters of Wang view the model's release as the clearest signal yet that Meta's AI rebuilding effort is making progress. They believe subsequent models, expected in the coming months, could further close the gap with Alphabet, OpenAI, and Anthropic.

Rus Salakhutdinov, a computer science professor at Carnegie Mellon University and former vice president of AI research at Meta, stated, "The amount of work TBD Labs has done in a short time is impressive. Alex knows what he doesn't know, and he's willing to listen."

Others within Meta are far less convinced. Critics describe Wang's leadership style as frenetic and impatient, arguing he overstates what is often incremental progress. Some current and former employees are skeptical about Meta's ability to achieve leadership in cutting-edge AI under Wang's direction.

A former Meta AI employee said, "TBD folks, including Alex and Zuck, set a pretty low bar for Muse Spark, both internally and externally. Other labs are moving incredibly fast."

Meta stated, "Alex's results speak for themselves: in less than a year, he helped build one of the industry's strongest research teams and led Meta's Superintelligence Lab to release Muse Spark, laying the scientific and technical groundwork for scaling more advanced models. We look forward to everyone seeing what they do next."

Meta is investing tens of billions of dollars in AI, with investors demanding evidence these expenditures translate into revenue. Muse Spark and future TBD models are expected to improve Meta's content and ad targeting, and support a range of planned initiatives from AI assistants and business agents to digital avatars and wearable devices.

Wang was recruited last year after Meta's AI work suffered a series of setbacks, including a disappointing reception for the Llama 4 model and growing internal concern that competitors were pulling further ahead.

Zuckerberg's response was to invest $15 billion in Wang's data-labeling startup, Scale AI, and hire its co-founder.

Scale AI had worked closely with leading AI labs, and Zuckerberg believed Wang's network and operational intensity could help rebuild Meta's research organization.

According to insiders, Wang was granted unusual autonomy and secrecy, quickly assembling TBD Labs, a team of about 100 handpicked researchers working in a secure area of Meta's Menlo Park headquarters requiring special badges.

Both Wang and Zuckerberg have offices within this workspace, and non-TBD employees have occasionally been caught trying to sneak in.

According to several sources, TBD encountered some early startup problems. Some staff were poached by rivals, including former Apple executive Ruoming Pang, who left for OpenAI after just seven months.

Several sources indicated that certain research efforts faced challenges, including an attempt to develop a completely new codebase for model training.

Ultimately, according to sources, Muse Spark was built using some elements of Meta's existing AI infrastructure, including code and datasets related to Llama 4.

As the TBD team settled, Wang sought to chart a roadmap that reconciled his and Zuckerberg's vision for "personal superintelligence," the beliefs of individual researchers, and the practical realities of scaling infrastructure to train future model generations, according to people familiar with his thinking.

He also reshaped Meta's AI safety work through a new team internally called TBA, or "To Be Aligned."

In leadership discussions with executives including Zuckerberg, Wang prioritized advancing the models themselves, while some other leaders were more focused on shipping AI products quickly, according to sources.

An internal source said that in presentations for AI teams, Wang advocated for an idealistic pursuit of developing super-powerful AI capable of solving the world's hardest problems, contrasting with others' focus on social media applications.

Wang also argued for greater emphasis on proprietary models over Meta's longstanding open-source approach, according to several sources.

Wang sought to build support for his vision by fostering a non-hierarchical, startup-like culture within TBD. In a recent podcast, he argued that "small teams where everyone is a 'god-tier' will always move faster than large organizations with diffused responsibility," borrowing gamer slang for exceptionally talented engineers.

He also regularly hosted happy hours with boba tea to foster camaraderie within the secretive team, according to insiders.

The broader Meta workforce experienced a less pleasant period. Wang's first year coincided with company-wide restructuring and multiple rounds of layoffs aimed at offsetting its massive AI spending.

Some employees also protested company plans to install tracking software to capture their computer usage for training AI models. Meta informed staff in a memo on Tuesday that it would partially roll back the plan following strong backlash.

Muse Spark is also currently deployed primarily within Meta's own products, making external evaluation difficult. Wang has stated some external companies would gain access via private APIs, but that rollout is limited.

Muse Spark has been praised for its visual understanding capabilities, but Wang acknowledged it lags rivals in programming ability. Several employees said staff asked to test the model for software development tasks still preferred using Anthropic's Claude.

Future Meta models are expected to focus on programming, completing agent tasks, and more advanced multimodal capabilities, including video generation.

A person close to Wang said, "The process of finding his influence when he first joined the company had a rocky start. But he's found his rhythm now."