Meta's Overmind Lab has released its first product. Is it possible to make money on it?

In early April, Meta unveiled its Muse Spark AI, the first to be created since Mark Zuckerberg actually personally demolished and then rehired an AI development team last year. But in 2026, simply owning your own AI is no longer enough - it is important to be able to answer investors' questions about how to monetize it.

A year ago in April 2025, Meta co-founder Mark Zuckerberg was probably not happy. His then-new Llama 4 family of AI models, which was supposed to displace competitors from OpenAI, Google, and Anthropic, received a very cool reception from the community. Moreover, the company was accused of falsifying the results: specially retrained AI models were allegedly sent to the tests, not the ones that the company later published. Officially, of course, Meta denied everything, but the famous researcher Jan Lekun, who was Meta's chief AI researcher at the time, later admitted that "the results were a little bit falsified".

Amid the tumult of failure, Zuckerberg went into "founder mode," that is, without relying on managers, he took direct control of the process. He fired virtually all AI employees and personally recruited new ones, declaring his goal to create "personal superintelligence for everyone." Whatever that means.

His main and most expensive "acquisition" in Meta's history was Alexander Wang, the then 28-year-old co-founder and CEO of ScaleAI. To get the young manager, Zuckerberg agreed to buy 49% of ScaleAI for $14.3 billion, in exchange for Wang's commitment to move to Meta and lead the Superintelligence project.

A talented young man, Alexander Wang founded ScaleAI in 2016 at the age of 19, dropping out of his studies at the prestigious Massachusetts Institute of Technology (MIT) for the sake of business.

The company has been busy marking up video data for AI autonomous cars.

"Someone had to train the AI to tell the difference between a paper bag and a pedestrian," Forbes writes. This was done by over 200,000 people hired for pennies in countries such as Kenya, the Philippines and Venezuela.

The bet to serve the AI market has been very successful, and with new generative AI models coming in late 2022, it's simply stellar. Almost all major AI labs have partnered with ScaleAI for data partitioning, which is how Wang met Zuckerberg in 2019.

After the failure of Llama 4, Zuckerberg began to increasingly "groom" Wang, which culminated in an epic deal.

Lecun called Wang "young" and "inexperienced," which is true on one hand - ScaleAI had never built its own AI. On the other hand, the company had worked on data partitioning with all the leading AI labs and Wang had a thorough knowledge of what they were doing.

In general, he had insights about competitors, plus good connections throughout Silicon Valley and in the U.S. government - as you can imagine, $14.3 billion is not paid simply for intelligence and beautiful eyes. But, as Carl Marks said, practice is the criterion of truth.

Finally, on April 8, Meta unveiled the first creation of Wang and his team - its new AI Muse Spark. Contrary to recent practice, Meta did not claim that its AI is the most powerful that mankind has ever seen. It was simply described as "competitive" in areas such as text, image, video and audio perception, reasoning, healthcare and agent management. Moreover, it is even modestly acknowledged that there are "performance gaps" in long-term agent management systems and programming (OpenAI and Anthropic AI are particularly strong in the latter).

Further efforts to develop AI and scale computing power are promised. Meta said its AI-related capital expenditures in 2026 will be between $115 billion and $135 billion, nearly double last year.

What will they pay for?

Muse Spark is a proprietary model, meaning that it can only be used with Meta's permission, just like other major AI companies. This is a radical departure from the principles of "open AI" proclaimed earlier by Zuckerberg himself - Llama's models were opt-in, i.e. available for free use and modification. Some criticize him for this, but apparently business is more important than principles.

This approach opens up the possibility of monetization through selling access to the API programming interface, but developers are unlikely to rush out to buy it, especially given the model's admitted weakness in programming.

Andrew Boone, an analyst at Citizens, told CNBC that Meta's obvious advantage is the more than 3 billion people who use Facebook, Instagram and WhatsApp every month. And the business opportunity for Meta has nothing to do with trying to attract developers, who are currently moving en masse to the models of OpenAI, Anthropic, Google and a host of Chinese companies, but rather focusing on its core market: advertising.

According to Emarketer, Meta is doing very well indeed on the advertising side, and in 2026 it could surpass Google in ad sales for the first time, collecting $243.5 billion versus $239.5 for a competitor.

"It [advertising - Oninvest note] is the diamond in the crown, that's what you need to keep improving," Boone says and recommends buying Meta.

The main source of Meta's revenue is advertising and Muse Spark is being integrated into the company's ecosystem precisely to increase its efficiency, Freedom Broker capital markets analyst Saken Ismailov confirmed in an interview with Oninvest.

The model is planned to be integrated into the company's WhatsApp, Instagram, Facebook services, which will increase user engagement and time spent in these applications, and, accordingly, increase ad impressions. In addition, the new model will increase the effectiveness of ad targeting, and this, in turn, will allow the company to increase the average price more confidently, as advertisers will get a better conversion rate, he reasoned.

In addition to the advertising business, Meta will try to monetize the e-commerce business, where AI helps choose goods, Ismailov believes. This is an ambitious idea affecting a multibillion-dollar market, but its successful realization remains in doubt. OpenAI and Google have already advanced the furthest in this direction. Due to the integration of agents directly into mass products (search, Gemini, ChatGPT) and large partnerships with retail and payment networks, they turn the AI response into a "showcase of goods" with the possibility of purchase, the analyst notes.

After the release of Muse Spark, Meta's shares rose by almost 10%, major banks raised their target prices and recommendations on the company's shares and regarded the launch of Muse Spark as proof that Meta's huge investments in AI are paying off. The market was prepared for the model to be weaker than analogs from Anthropic, OpenAI and Google, but the gap in benchmark results was not as big as expected, Ismailov notes.

Three days after the release of Muse Spark, Wang rejoiced on social network X that the Meta AI app, which now incorporates the model, had climbed to the second spot in the U.S. AppStore and was first among AI apps. By April 22, it had slipped to seventh place, , passing three major AI competitors: ChatGPT, Claude, and Gemini.

It's not a bid for victory just yet, but the fact that Meta has managed to get back into the big AI game is perhaps beyond doubt.

This article was AI-translated and verified by a human editor