Mark Zuckerberg, CEO of Meta Platforms Inc., wears Meta Oakley Vanguard AI glasses during the MetaConnect event on Wednesday, September 17, 2025 in Menlo Park, California, USA.
David Paul Morris | Bloomberg | Getty Images
About 10 months later Meta After spending billions of dollars to bring Scale AI’s Alexandr Wang on board as the central figure in Mark Zuckerberg’s AI overhaul, the company finally announced its first new model on Wednesday. One of the big questions is: Will users pay?
OpenAI, Anthropic, google Meta, which has been at the forefront of the artificial intelligence boom with its powerful models, popular chatbots and other services, has been spending heavily on AI but has yet to show any new revenue streams from it.
In June, Meta spent more than $14 billion to hire Mr. Wang and several of its top engineers and researchers, and soon established a new elite force, the Meta Superintelligence Institute. And in January, the company told Wall Street it planned to spend $115 billion to $135 billion on capital spending this year, nearly double its 2025 capital spending.
“There were basically no releases this year, there wasn’t a ton of hiring, but the concerns about capital spending this year have become more pronounced,” Morningstar analyst Malik Ahmed Khan said in an interview. “I think Meta needed to show investors and operators that it was making a real commitment. That’s the first step.”
Khan said Meta’s second step is to find a way to make the model work and monetize it.
Meta’s newly released model, Muse Spark, is proprietary and is a significant change from its previous model family called Llama, which consisted of open source products, but the company said it plans to eventually release some open source versions. Zuckerberg overhauled the company’s strategy after Llama 4, released in April, failed to attract developers.
Alexander Wang speaks on CNBC’s “Squawk Box” outside the World Economic Forum in Davos, Switzerland, on January 23, 2025.
Jerry Miller | CNBC
Gartner analyst Arun Chandrasekaran described the move as a “major change” and said it “signals an intention to move away” from the Rama brand.
Taking cues from other frontier AI labs, Meta aims to eventually offer paid API access to Muse Spark to third parties after an initial “private API preview” with “selected parties.”
But the meta was way behind the game. OpenAI and Anthropic are collectively worth well over $1 trillion due to the popularity of their models and services, and Google is incorporating Gemini into its portfolio of apps and products, while also selling access to the Gemini model through its cloud unit.
For Meta’s AI technology to be successful, it must be powerful enough to compete with top models, while also providing new business opportunities.
“Crown Jewel”
Citizens analyst Andrew Boone said Meta’s clear advantage is that more than 3 billion people use Facebook, Instagram and WhatsApp each month. And the business opportunity for Meta has nothing to do with trying to attract developers who currently flock to OpenAI, Anthropic, Gemini, and a host of Chinese models, but rather to focus on its core market: advertising.
“That’s the crown jewel and we need to keep improving,” Boone said, recommending the stock be bought.
Khan shares that sentiment.
“We believe that will be a killer use case from Meta’s 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 in revenue last year. The company has made a number of efforts to diversify its business, most notably spending tens of billions of dollars on the Metaverse. But Meta’s advertising model is the only one that has consistently worked, and the company’s investments in AI have helped improve targeting capabilities and give marketers better tools.
Khan said once advertisers see a return on investment from their meta spend, they will reinvest that money into more advertising on the platform. So it stands to reason that people are willing to pay for AI services if they can get better results.
Meta declined to comment on its API plans beyond the initial announcement.
Doris Xin, CEO of AI startup Disarray, said that based on technology benchmarks released by Meta comparing Muse Spark and competing products, the new AI model appears to be superior in areas related to image and video processing. These are important characteristics for advertisers looking to run dynamic campaigns to audiences accustomed to watching short-form videos on Reels or looking at cat photos on Facebook or Instagram.
“I think it’s definitely more consumer-oriented than Claude or Gemini,” Shin said of Muse Spark.
But Zuckerberg has long had ambitions that go far beyond advertising. His approach to Llama was to target developers and use Meta’s tools to acquire the best talent in AI, even if they weren’t paying for it.
Switching to a proprietary model will make selling to developers even more difficult. Joseph Ott, CEO of AI startup Sam Legal Technologies, said he doesn’t know where to find his value.
“The only reason I use Llama is because I can fine-tune it,” Ott said, referring to the practice of customizing AI models.
Many developers use so-called openweight AI models, such as those provided by Chinese technology companies, as the basis for training AI models for specific use cases. Ott said it’s unclear what makes Meta’s Muse Spark better than free or cheaper alternatives or leading proprietary AI models.
Ulrik Stig Hansen, co-founder of AI and data training startup Encord, said it was important for Meta to develop its own AI-based model to avoid future dependence on third parties. As one of the few companies with the resources and computing infrastructure needed to create and maintain AI models at scale, Meta hopes to remain relevant in the hottest market on the planet.
“It’s about AI sovereignty and being a player in the game,” Hansen said. “They want to be recognized and known as an AI company.”
Regarding Mehta’s large investment in Wang and his team, Boone said the latest benchmarks suggest Zuckerberg got what he wanted and now it’s “back to Mark.”
“We’ve just given you a cutting-edge frontier model,” Boone said, referring to the team behind Muse Spark. “So what are you going to do?”
WATCH: Meta launches new AI model ‘Muse Spark’.
Correction: Advertising accounted for 98% of Meta’s $200 billion in revenue last year. In previous versions, the diagram features were incorrect.
