One year ago Mark Zuckerberg made one of the largest talent bets in tech history. He poured $14.3 billion into Scale AI and lured its 28-year-old founder Alexandr Wang to lead a new unit called Meta Superintelligence Labs. The goal was simple. Catch up to OpenAI, Anthropic and Google in frontier models. Deliver something that could power Meta’s apps, glasses and future products.
Wang delivered Muse Spark in April. The model marked Meta’s first serious proprietary foundation model. It moved the company away from its long-standing open-weight Llama strategy. Yet the stock has lagged. Meta shares are down 18% over the past 12 months. That’s the worst performance among big tech names. Developers remain skeptical. And the pressure now sits squarely on Zuckerberg’s shoulders.
From Llama Stumbles to a Massive Talent Heist
Meta’s earlier open-source push with Llama won praise from some developers but failed to generate the excitement or revenue that closed models delivered for rivals. Llama 4’s release last April landed with a thud. Two months later Zuckerberg struck the Scale deal. It brought Wang, several of his top engineers and a deep understanding of data labeling that powers nearly every major AI lab.
Wang didn’t arrive alone. Former GitHub CEO Nat Friedman joined around the same time. The new lab pulled talent from Google DeepMind, Anthropic and OpenAI. Some recruits reportedly received packages worth up to $100 million. Wang himself became one of the company’s highest-paid employees. He now serves as chief AI officer.
In recent appearances Wang has described the vision as “personal super intelligence.” He told an audience in India earlier this year that the aim is AI “that knows you, your goals, your interests and helps you with whatever you’re focused on doing.” (Observer, Feb. 19, 2026)
Meta spent $72.2 billion on AI-related capital expenditures last year. It expects that number to climb to between $115 billion and $135 billion in 2026. The company has expanded data centers and computing capacity at a furious pace. Yet those numbers have not yet translated into market leadership or stock gains.
Analysts want evidence that Meta can turn its models into products people will pay for. “Meta needs to provide more proof points of both adoption and commercialization,” said Ralph Schackart, an analyst at William Blair. He still recommends buying the stock. The first-quarter revenue growth of 33% was the fastest since 2021. But investors want more than better ad targeting.
So far Muse Spark focuses on integration inside Facebook, Instagram, the standalone Meta AI app and devices such as Ray-Ban Meta glasses. That consumer-first approach differs from the developer-centric strategies of some competitors. Andrew Moore, CEO of enterprise startup Lovelace and former Google Cloud AI chief, sees potential if Meta can prove advantages in cost or latency. “If they do proprietary, computationally efficient models, that will be so different from what’s happening in this death match between the big guys. They might really benefit.”
But others aren’t convinced. Krish Subramanian, CEO of KOI AI and a former IBM Consulting product head, says developers show more excitement for Google’s offerings. The open-weight appeal of earlier Llama models has faded. “The lack of developer trust will come back to hit them if they don’t focus on third-party developers.” He notes it took Microsoft years to regain credibility with open-source communities.
Rob May, CEO of startup Neurometric, puts it bluntly. “I think the AI community largely ignores Meta at this point.” He called Muse Spark a “yawn” for many researchers because it remains less accessible than rival models. May says his previous regular contact with Meta on Llama issues has dried up. Messages go unanswered.
And. The internal picture looks messy. Meta conducted multiple rounds of layoffs this year, including about 8,000 workers in May. Cuts hit trust and safety teams. Some employees worry that reduced oversight could create problems as models grow more powerful. Wang addressed the topic on a podcast last month. “One of the things that is very important to me is safety for these models.”
Morale has suffered. Tension exists at the top of the AI organization. Sources say both Wang and Friedman face pressure to show meaningful revenue from their work. Meta tech chief Andrew Bosworth, a longtime Zuckerberg ally, could take a larger role if the newcomers stumble. A company spokesperson pointed to Wang’s public comments supporting the open-source community and said Muse Spark’s underlying technology would become available via API to outside developers this month. “We’re already testing with some early partners, and look forward to releasing it this month.”
Zuckerberg has taken a hands-on approach. He reportedly spends hours each week coding and reviewing AI work. He moved his desk closer to the lab. The CEO has signaled that selling the technology falls to him. Wang builds it. Zuckerberg positions it.
Recent reporting shows the bet carries risks. A CNBC video analysis published yesterday highlights low morale after layoffs, developer indifference and the need to prove revenue beyond advertising. Wang has pushed back on claims that Meta simply bought talent with cash. In a May interview he argued that many researchers joined for the culture, compute resources and clarity of mission, not just paychecks. (Observer, May 14, 2026)
Yet questions linger. Can a company whose primary business remains advertising and social media truly compete in a race defined by massive compute clusters and closed models? Will Muse Spark prove more than an “appetizer,” as Wang has called it, for bigger breakthroughs? And will developers return once the API lands?
Meta’s enormous user base gives it an advantage few rivals can match. Over 3.5 billion daily active users across its apps create vast amounts of interaction data. Wang’s background at Scale taught him exactly how to turn that data into training fuel. The company has shifted resources aggressively. Some engineers now generate synthetic data or solve coding problems to train models. That reallocation has sparked internal complaints and even a petition from over 1,600 workers.
Still, progress is visible. Muse Spark integrates directly into consumer experiences. Always-on AI agents are rolling out across apps and wearables. The bet on proprietary technology after years of open-weight advocacy represents a sharp strategic turn. Wang’s data-labeling expertise and network give Meta unusual visibility into competitors’ bottlenecks.
Zuckerberg now carries the heavier load. He must convince developers, advertisers and Wall Street that Meta’s AI can generate new revenue streams instead of simply sharpening its existing ad machine. The $14.3 billion wager bought talent and time. Results must follow. Short, sharp execution matters now. Longer-term questions about safety, trust and competitive differentiation will decide whether the investment pays off.
Wang has said he views AI as one of the most consequential technologies of our time. He left a company he built from age 19 to take this shot. Zuckerberg, for his part, has never been shy about bold swings. This one may define the next decade for both men and the social media giant they lead.
WebProNews is an iEntry Publication