Mark Zuckerberg, chief govt officer of Meta Platforms Inc., wears a pair of Meta Oakley Vanguard AI glasses in the course of the Meta Join occasion in Menlo Park, California, US, on Wednesday, Sept. 17, 2025.
David Paul Morris | Bloomberg | Getty Photographs
Virtually 10 months after Meta spent billions of {dollars} to usher in Scale AI’s Alexandr Wang because the centerpiece of Mark Zuckerberg’s AI overhaul, the corporate lastly revealed its first new mannequin on Wednesday. One massive query is — will customers pay for it?
Whereas rivals like OpenAI, Anthropic and Google have spearheaded the unreal intelligence growth with highly effective fashions and standard chatbots in addition to different companies, Meta has been a hefty spender on AI however has but to indicate any new income streams from it.
In June, Meta shelled out greater than $14 billion to rent Wang and a few of his prime engineers and researchers, quickly creating Meta Superintelligence Labs as a brand new elite unit. And in January, the corporate informed Wall Road it plans to pour between $115 billion and $135 billion this yr into capital expenditures, practically double its 2025 capex determine.
“It has been a yr of principally no releases and loads of hiring, after which the capex worries for this yr are pronounced,” mentioned Morningstar analyst Malik Ahmed Khan, in an interview. “I believe Meta needed to present traders and operators they’ve been engaged on one thing of substance. That is step one.”
Meta’s second step, Khan mentioned, is making the mannequin work and determining how one can monetize it.
Muse Spark, Meta’s newly launched mannequin, is proprietary, a pointy change from its predecessor household of fashions known as Llama, which consisted of open-source choices, although the corporate mentioned it does plan to finally launch some open-source variations. Zuckerberg shook up his firm’s technique after the April launch of Llama 4, which didn’t captivate builders.
Alexandr Wang speaks on CNBC’s “Squawk Field” exterior the World Financial Discussion board in Davos, Switzerland, on Jan. 23, 2025.
Gerry Miller | CNBC
Arun Chandrasekaran, an analyst at Gartner, described the transfer as a “main shift” and mentioned it “alerts an intention to maneuver away” from the Llama model.
Taking a cue from different frontier AI labs, Meta goals to finally supply third events paid API entry to Muse Spark after an preliminary “non-public API preview” with “choose events.”
However Meta could be very late to the sport. OpenAI and Anthropic are collectively valued at effectively over $1 trillion, due to the recognition of their fashions and companies, and Google has embedded Gemini throughout its portfolio of apps and merchandise, whereas additionally promoting entry to the Gemini fashions by way of its cloud unit.
Meta’s AI know-how, to succeed, must be adequate to compete with prime fashions whereas additionally offering a novel enterprise alternative.
‘Crown jewel’
Andrew Boone, an analyst at Residents, mentioned Meta’s clear benefit is the greater than 3 billion individuals who use Fb, Instagram and WhatsApp each month. And the enterprise alternative for Meta has nothing to do with attempting to draw builders, who at the moment swarm to OpenAI, Anthropic, Gemini and a number of Chinese language fashions, however quite to give attention to its core market: promoting.
“That is the crown jewel, that is what must proceed to enhance,” mentioned Boone, who recommends shopping for the inventory.
Khan shares that sentiment.
“I imagine that might be the killer use case from Meta’s perspective,” Khan mentioned, with the objective being to “make adverts extra participating and enhance concentrating on.”
Promoting accounted for 98% of Meta’s $200 billion in income final yr. The corporate has made quite a few efforts to diversify its enterprise, most notably spending tens of billions of {dollars} to attempt to make the metaverse occur. However Meta’s advert mannequin is the one factor that is persistently labored, and the corporate’s investments in AI have served to enhance its concentrating on capabilities and supply higher instruments for entrepreneurs.
Khan mentioned that as advertisers see returns on funding from their Meta spending, they reinvest that cash again into extra adverts on the platform. So it is sensible that they’d be prepared to pay for AI companies if they’ll get even higher outcomes.
Meta declined to remark about its API plans past its preliminary announcement.
Primarily based on the technical benchmarks Meta launched evaluating Muse Spark to rivals, the brand new AI mannequin seems to excel in areas associated to picture and video processing, mentioned Doris Xin, CEO of AI startup Disarray. These are necessary traits for advertisers searching for to make dynamic campaigns for an viewers that is grown accustomed to viewing short-form movies on Reels or gawking at cat pictures on Fb and Instagram.
“In comparison with like Claude and Gemini, I believe it undoubtedly feels prefer it has extra of a client bent,” Xin mentioned about Muse Spark.
Zuckerberg, nonetheless, has lengthy had ambitions that go effectively past promoting. His strategy with Llama was focused at builders and getting the very best and brightest minds in AI utilizing Meta’s instruments even when they weren’t paying for them.
With the swap to proprietary fashions, the pitch to builders turns into tougher. Joseph Ott, CEO of AI startup Samu Authorized Applied sciences, mentioned he is not sure about the place he would discover worth.
“The one cause I’d use Llama is that I may fine-tune it,” Ott mentioned, referring to the apply of customizing AI fashions.
Many builders use so-called open-weight AI fashions, like these supplied by Chinese language tech firms, as a foundation to coach AI fashions to fulfill their particular use circumstances. Ott mentioned it is unclear what would make Meta’s Muse Spark stand out in opposition to free or cheaper options and the main proprietary AI fashions.
Ulrik Stig Hansen, co-founder of AI and knowledge coaching startup Encord, mentioned it is necessary for Meta to develop its personal AI basis fashions to keep away from any future dependencies on third events. As one of many few firms with the assets and computing infrastructure essential to create and keep massive AI fashions, Meta needs to make sure that it stays related within the hottest market on the planet.
“It’s about AI sovereignty and being a participant within the sport,” Hansen mentioned. “They need to be perceived and generally known as an AI firm.”
As for Meta’s huge funding in Wang and his workforce, Boone mentioned the most recent benchmarks recommend that Zuckerberg obtained what he wished, and now it is “again on Mark.”
“We simply gave you a state-of-the-art frontier mannequin,” Boone mentioned, referring to the workforce behind Muse Spark. “What are you going to do with it?”
WATCH: Meta unveils its new AI mannequin: “Muse Spark.”
Correction: Promoting accounted for 98% of Meta’s $200 billion in income final yr. An earlier model mischaracterized the determine.
