Meta’s New AI: A Game Changer or a Gamble?

Nearly a year after Meta Platforms Inc. made a substantial investment to bring Alexandr Wang and his team from Scale AI into the fold, the tech giant has finally unveiled its latest AI model. The central question now is whether users will be willing to pay for it.

While competitors such as OpenAI, Anthropic, and Google have dominated the artificial intelligence landscape with their sophisticated models, popular chatbots, and diverse services, Meta has been a significant spender in AI without yet demonstrating clear new revenue streams from these investments.

In June, Meta significantly bolstered its AI capabilities by acquiring Wang and a contingent of his top engineers and researchers, subsequently establishing Meta Superintelligence Labs as a dedicated elite unit. Furthermore, the company indicated to investors that it plans to allocate between $115 billion and $135 billion for capital expenditures this year, nearly doubling its 2025 figures.

“It’s been a year characterized by minimal product releases and extensive hiring, with pronounced capital expenditure concerns for the current year,” observed Malik Ahmed Khan, an analyst at Morningstar, in a recent interview. “Meta needed to demonstrate to investors and stakeholders that substantial progress has been made. This unveiling marks the initial step.”

Khan further elaborated that Meta’s subsequent critical step involves making the model functional and devising effective monetization strategies.

The newly introduced model, dubbed Muse Spark, represents a departure from Meta’s previous Llama family of models, which were primarily open-source offerings. While Meta has stated its intention to eventually release some open-source versions of Muse Spark, the initial iteration is proprietary. This strategic shift followed the April release of Llama 4, which reportedly failed to garner significant traction among developers.

Arun Chandrasekaran, a Gartner analyst, characterized this move as a “major shift” and suggested it “signals an intention to move away” from the Llama brand entirely.

Following the industry trend set by other leading AI labs, Meta intends to offer third parties paid API access to Muse Spark, following an initial “private API preview” with select partners.

However, Meta enters this market considerably later than its rivals. OpenAI and Anthropic have collectively achieved valuations exceeding $1 trillion, driven by the immense popularity of their models and services. Google, meanwhile, has integrated its Gemini models across its extensive portfolio of applications and products, while also offering access to these models through its cloud services.

For Meta’s AI technology to succeed, it must not only rival the performance of leading models but also present a compelling and novel business opportunity.

Andrew Boone, an analyst at Citizens, highlighted Meta’s significant advantage: its user base of over 3 billion individuals who engage with Facebook, Instagram, and WhatsApp monthly. The primary business opportunity for Meta, he argues, lies not in attracting developers, who are currently flocking to OpenAI, Anthropic, Gemini, and various Chinese AI models, but rather in enhancing its core revenue stream: advertising.

“That’s the crown jewel, and it’s what needs to continue to improve,” Boone stated, recommending a buy rating for Meta’s stock.

Khan shares this perspective.

“I believe that would be the killer use case from Meta’s perspective,” Khan commented, emphasizing the goal of “making ads more engaging and improving targeting.”

Advertising constituted 98% of Meta’s $200 billion in revenue last year. While the company has made considerable efforts to diversify its business, most notably through its multi-billion dollar investment in the metaverse, its advertising model remains its most consistently successful revenue driver. Meta’s AI investments have historically served to refine ad targeting capabilities and provide enhanced tools for marketers.

Khan posited that as advertisers witness a strong return on investment from their Meta advertising spend, they are likely to reinvest further into the platform. Consequently, they would likely be receptive to paying for AI services that promise even more superior results.

Meta declined to provide further comment on its API strategy beyond its initial announcement.

Based on the technical benchmarks Meta has released, comparing Muse Spark against its rivals, the new AI model appears to demonstrate particular strengths in image and video processing, according to Doris Xin, CEO of AI startup Disarray. These capabilities are crucial for advertisers aiming to create dynamic campaigns targeting an audience increasingly drawn to short-form video content on Reels and visual platforms like Facebook and Instagram.

“Compared to models like Claude and Gemini, it definitely feels like it has more of a consumer bent,” Xin remarked about Muse Spark.

However, Mark Zuckerberg has consistently harbored ambitions that extend well beyond advertising. His prior approach with Llama was aimed at engaging developers and attracting top AI talent to utilize Meta’s tools, even if direct monetization wasn’t the immediate goal.

With the pivot to proprietary models, the value proposition for developers becomes more complex. Joseph Ott, CEO of AI startup Samu Legal Technologies, expressed uncertainty about the specific advantages Muse Spark might offer him.

“The only reason I would use Llama is that I could fine-tune it,” Ott explained, referring to the process of customizing AI models for specific applications.

Many developers utilize open-weight AI models, including those from Chinese tech companies, as foundational platforms to train AI models tailored to their unique use cases. Ott questioned what would differentiate Meta’s Muse Spark from existing free or more affordable alternatives, as well as leading proprietary AI models.

Ulrik Stig Hansen, co-founder of AI and data training startup Encord, emphasized the strategic importance for Meta to develop its own foundational AI models to mitigate future dependencies on third parties. As one of the few entities possessing the requisite resources and computing infrastructure for creating and maintaining large-scale AI models, Meta aims to remain a significant player in this highly competitive market.

“It is about AI sovereignty and being a player in the game,” Hansen stated. “They want to be perceived and recognized as an AI company.”

Regarding Meta’s substantial investment in Alexandr Wang and his team, Boone indicated that the latest benchmarks suggest Zuckerberg has achieved his objective, and the onus is now on him.

“We’ve delivered a state-of-the-art frontier model,” Boone remarked, referring to the team responsible for Muse Spark. “Now, what will you do with it?”

Correction: Advertising accounted for 98% of Meta’s $200 billion in revenue last year. An earlier version of this report mischaracterized this figure.