Meta's Chief AI Officer Alexandr Wang: Llama 4 is not on the track Meta hoped for, clearly a disappointing result internally
Alexandr Wang, co-founder of Scale AI and current Chief AI Officer at Meta, recalled during an interview with David Senra that Zuckerberg called him for the first time after the Llama 4 release, asking, "What do you think we should do?"
Wang stated that Llama 4 is not on the track Meta hoped for, and it was clearly a disappointing result internally, while AI is obviously crucial for Meta's future. He pointed out that Zuckerberg, despite many achievements, still actively seeks external opinions, reflecting a quality of humility. The entire timeline from the first contact to the announcement of the deal was about five to six weeks, which is very fast.
The deal structure involves Meta acquiring a 49% non-voting stake in Scale AI for approximately $14.3 billion, with Scale continuing to operate as an independent company. Wang and some colleagues will join Meta and lead the Meta Superintelligence Labs. Wang mentioned that the acquisition was not initially anticipated, as Scale primarily serves enterprise and government clients, which is different from the logic of products like Instagram and WhatsApp, and it was only after discussing specific numbers that it became clear that the other party was seeking talent.
Wang was introduced to Zuckerberg around 2021-2022, and their first meeting took about a year to arrange. They have communicated roughly every six months since then, and Wang has received detailed replies from Zuckerberg via WhatsApp. The setbacks with Llama 4 prompted an increase in the frequency and depth of their contact. Wang described leaving Scale, which he had managed for nine years, as an emotionally difficult process, but ultimately believes the deal is beneficial for all parties, and that Meta's AI situation at the time provided space to build principles and culture from scratch.
The interview also covered Wang's plans to rebuild the AI lab with a small team and flat structure after joining, as well as the seven-month development process of the personal AI product Muse. After the deal, some of Scale's major clients, such as Google and OpenAI, adjusted their cooperative relationships.
Market Mechanism: Meta is responding to the setbacks of Llama 4 and competitive pressures from OpenAI and Anthropic through equity investment and talent acquisition. The event-driven nature is due to the model's trajectory not meeting expectations, leading to a rapid acquisition of external talent and data labeling capabilities; funds flow from Meta to Scale's shareholders and equity, benefiting Wang and the team joining Meta (gaining leadership and resources) as well as remaining shareholders of Scale (gaining valuation and liquidity), while the existing AI team at Meta faces pressure from restructuring and increased talent density.
Source: Public Information
ABAB AI Insight
Alexandr Wang co-founded Scale AI in 2016, focusing on data labeling and model evaluation services, and became one of the youngest self-made billionaires around the age of 24, with clients including major model companies like OpenAI and Google. After Llama 4 underperformed in 2025, Zuckerberg proactively reached out and completed a $14.3 billion, 49% equity deal in about five to six weeks, bringing Wang in to lead the newly established Superintelligence Labs.
In terms of capital strategy, Meta injected funds in the form of non-voting equity while incorporating Wang and key personnel internally, aiming to quickly gain insights into industry-wide data and evaluation processes, as well as the talent density of young operators, rather than simply purchasing data businesses; this continues Zuckerberg's approach of directly recruiting external founders at critical junctures, transforming external startup resources into a starting point for internal lab reconstruction.
In analogy, this is similar to early tech giants bringing in external teams through equity and talent transactions after product setbacks, such as some companies acquiring and integrating startup teams when their search or social products lagged; currently, Meta is transitioning from an open-source model approach to personal AI and superintelligence labs, attempting to catch up with closed frontier labs using small, high-density teams.
Essentially, this represents a technological substitution under capital concentration: when internal model trajectories deviate, external data labeling and operational capabilities are rapidly concentrated into new labs through quick equity and talent transactions, shifting AI research and development from a decentralized open-source path to an internally concentrated model focused on talent density and product direction control, with the mechanism being to exchange capital for direct control over key talent and industry knowledge, thereby shortening the reconstruction cycle.
ABAB News · Cognitive Laws
After model setbacks, talent changes trajectory faster than parameters.
Humility is reflected in actively seeking external input and then quickly completing transactions.
Equity investment is an accelerator for acquiring talent density.