Mark Zuckerberg, CEO of Meta Platform Inc., wears a pair of Meta Oakley Vanguard AI glasses during the Meta Connect event in Menlo Park, California, U.S., on Wednesday, Sept. 17, 2025.
David Paul Morris | Bloomberg | getty images
after about 10 months meta After spending billions of dollars to bring on Scale AI’s Alexander Wang as the centerpiece of Mark Zuckerberg’s AI overhaul, the company finally unveiled Muse Spark on Wednesday, its first new model since the transition. A big question is will users pay for it?
While rivals like OpenAI, Anthropic and Google Meta, which led the boom in artificial intelligence with powerful models and popular chatbots as well as other services, has spent heavily on AI, but has yet to see any new revenue sources from it.
In June, Meta spent more than $14 billion to hire Wang and some of his top engineers and researchers, soon forming Meta Superintelligence Labs as a new distinct entity. And in January, the company told Wall Street it planned to pour in between $115 billion and $135 billion in capital spending this year, nearly double its 2025 capital spending figure.
“This has been a year of basically no releases and a lot of hiring, and then the capex concerns for this year are obvious,” Morningstar analyst Malik Ahmed Khan said in an interview. “I think Meta needs to show investors and operators that they are doing something worthwhile. This is the first step.”
The second step for Meta is to make the model work and figure out how to monetize it, Khan said.
The Muse Spark, Meta’s newly released model, is proprietary, a sharp change from its predecessor family of models called Llama, which included open-source offerings, although the company said it plans to release some open-source versions eventually. Zuckerberg shook up his company’s strategy after the April release of Llama 4, which failed to attract developers.
Alexander Wang speaks on CNBC’s “Squawk Box” outside the World Economic Forum in Davos, Switzerland on January 23, 2025.
Gerry Miller cnbc
Arun Chandrasekaran, an analyst at Gartner, described the move as a “major change” and said it “signals an intention to move away” from the Llama brand.
Taking inspiration from other leading AI labs, Meta aims to eventually provide paid API access to Muse Spark to third parties after an initial “private API preview” with “select parties.”
But the meta is very late to the game. OpenAI and Anthropic are collectively valued at more than $1 trillion due to the popularity of their models and services, and Google has embedded Gemini in its portfolio of apps and products, while also selling access to Gemini models through its cloud unit.
To be successful, Meta’s AI technology needs to be good enough to compete with top models while also providing a new business opportunity.
‘Crown jewel’
Andrew Boon, an analyst at Citizens, said Meta’s clear advantage is the more than 3 billion people who use Facebook, Instagram and WhatsApp each month. And the business opportunity for Meta has nothing to do with trying to attract developers who are currently attracted to OpenAI, Anthropic, Gemini, and a number of Chinese models, but rather to focus on its core market: advertising.
“This is the crown jewel, it just needs to continue to improve,” said Boon, who recommends buying the stock.
Khan shares that sentiment.
“I believe this would be a killer use case from a meta perspective,” Khan said, adding that the goal is to “make ads more engaging and improve targeting.”
Advertising accounted for 98% of Meta’s $200 billion Income Last year. The company has made many efforts to diversify its business, spending tens of billions of dollars specifically to make the metaverse possible. But Meta’s advertising model is something that is constantly being worked on, and the company’s investments in AI have helped improve its targeting capabilities and provide better tools for marketers.
Khan said that as advertisers see a return on investment from their meta spend, they reinvest that money in more ads on the platform. So it makes sense that they would be willing to pay for AI services if they can get even better results.
Meta declined to comment about its API plans after its initial announcement.
Doris Shin, CEO of AI startup Disare, said that based on technical benchmarks released by Meta, comparing Muse Spark to rivals, the new AI model appears to excel in areas related to image and video processing. These are important features for advertisers who want to create dynamic campaigns for audiences who have grown accustomed to watching short-form videos on Reels or seeing cat photos on Facebook and Instagram.
“Compared to Cloud and Gemini, I think it definitely feels like it has more consumer traction,” Zin said of Muse Spark.
However, Zuckerberg has long-standing ambitions that go far beyond advertising. Their approach with Llama was targeted at developers and getting the best and brightest minds in AI using Meta’s tools, even if they weren’t paying for them.
With the switch to a proprietary model, the pitch to developers becomes more difficult. Joseph Ott, CEO of AI startup Samu Legal Technologies, said he is unsure where he will find value.
“The only reason I use Llama is so I can get it right,” Ott said, referring to the practice of optimizing AI models.
Many developers use so-called open-weight AI models, such as those provided by Chinese tech companies, as a basis for training AI models to meet their specific use cases. Ott said it’s not clear what will make Meta’s Muse Spark stand out against free or cheaper alternatives and leading proprietary AI models.
Ulrik Stig Hansen, co-founder of AI and data training startup Encord, said it is important for Meta to develop its own AI foundation models to avoid future dependency on third parties. As one of the few companies with the resources and computing infrastructure necessary to build and maintain large AI models, Meta wants to ensure it remains relevant in the hottest market on the planet.
“This is about AI sovereignty and being a player in the game,” Hansen said. “They want to be seen and known as an AI company.”
As far as Meta’s massive investment in Wang and his team is concerned, Boone said the latest benchmarks show that Zuckerberg got what he wanted, and now it’s “back on mark.”
“We’ve just given you a state-of-the-art frontier model,” Boone said, referring to the team behind the Muse Spark. “What are you going to do with it?”
Watch: Why Meta’s new AI model, Muse Spark, is such a big deal
Correction: Advertising accounted for 98% of Meta’s $200 billion revenue last year. This image was incorrectly depicted in an earlier version.
