By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panelists debate Meta's shift to proprietary Muse Spark, with most acknowledging potential internal ad efficiency gains but raising concerns about developer trust, privacy risks, and regulatory hurdles.
Risk: Privacy risks and regulatory scrutiny could erase potential margin uplift from Muse Spark's internal efficiency gains.
Opportunity: Successful integration of Muse Spark into Meta's ad engine could compound existing 33% Q1 revenue growth.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Wang's big accomplishment was the delivery of the Muse Spark AI model in April, marking Meta's first jump into proprietary foundation models and away from a strict adherence to open source, or open weight as it's more commonly called in AI. The group Wang leads — Meta Superintelligence Labs — was established to give the company some sizzle in the hottest corner of the tech industry.
Now that CEO Mark Zuckerberg has his new model, it's on him to make it a financial success. That means showing the company can attract paying users for its AI tools, rather than just using the technology to enhance and bolster its core advertising business.
"Meta needs to provide more proof points of both adoption and commercialization," said Ralph Schackart, an analyst at William Blair who recommends buying the stock. "Investors are looking for Meta to monetize a new AI-first product, beyond the substantial positive impact AI is having on enhancing the advertising models."
Wall Street, at least so far, is unimpressed. Meta's stock is down 18% over the past 12 months, the worst performer in the megacap group, along with Microsoft, which has its own challenges in AI. That's even after Meta reported 33% revenue growth in the first quarter, the fastest rate of expansion for any period since 2021.
For Meta, the problem started with what some industry experts called, in hindsight at least, a strategic blunder. The company jumped into AI with its Llama family of models, offering an open-source approach that allowed developers to freely tinker, while the other big model makers charged for access.
In April of last year, Meta's release of Llama 4 fell flat, failing to captivate developers and leading Zuckerberg to reconsider his company's approach to AI development. Two months later, Zuckerberg shocked the tech world, announcing his company's $14.3 billion investment for roughly half of Scale AI and, more importantly, bringing over Wang and his top lieutenants.
Wang's development and rollout of Muse Spark in April of this year got the ball rolling. Instead of focusing on third-party developers, the new model was designed to easily plug into Meta's apps like Facebook and Instagram as well as AI-powered devices like the Ray-Ban Meta glasses, said Thomas Randall, an analyst at the Info-Tech Research Group. That's on top of the standalone Meta AI app and site.
"There'll be a lot of these frontier model providers that will fundamentally change in lots of different ways, and Meta needs to have a consistent, reliable proprietary model that they themselves own," Randall said. He added that Meta would be "lost" if Zuckerberg didn't open his wallet for Wang and other big-name AI hires over the past year, in what Randall called a "strategic rebuild" for the company.
Randall said Meta hasn't taken the "most optimized route," but at least "I can now see a vision for what they're trying to achieve and what Wang has been trying to achieve," he said.
Since the release of Muse Spark, Meta has unveiled new AI and business-related subscription plans as part of an effort to expand its business beyond online ads. Historically, it hasn't worked. Meta still counts on ads for 98% of revenue.
Schackart said he wants to see "tangible evidence of a growing list of new, AI-first products created by Muse Spark, even if monetization lags." He said that's "what investors are looking for."
No matter how good Wang's model may be, Zuckerberg has a high hill to climb with developers coming off the Llama debacle.
"I think the AI community largely ignores Meta at this point," said Rob May, CEO of the startup Neurometric, which works in the realm of token engineering.
May said it's hard to gauge how much success Wang has had leading MSL, because the company has thus far only released one AI model, which he characterized as a "yawn" among the AI community since the technology is not widely accessible.
Although Meta was heavily courting third-party developers with Llama, May said the company's efforts under Wang seem geared toward internal uses. May said he used to be in regular touch with Meta for Llama-related issues, but now said he "can't get them to return messages."
May admits that it makes sense for Meta to focus on AI for its core ad products, because the company has a $200 billion a year business to protect.
"That company has built the machine," he said.
Andrew Moore, the CEO of enterprise startup Lovelace and former Google Cloud AI chief, said it's not too late for Meta to find a lane.
Meta has focused on making its models more efficient through training techniques. Moore said that could be a major differentiator among developers worried about the rising costs of foundation models.
"If they do proprietary, computationally efficient models, that will be so different from what's happening in this death match between the big guys," Moore said. "They might really benefit."
Moore added that Meta has to show an advantage somewhere, whether it be on cost, latency or other technical nuances that matter to developers.
Krish Subramanian, the CEO of consulting firm KOI AI and former product head at IBM Consulting, said developers are more excited about Google's AI models than what Meta is offering. The appeal of Llama was that it specifically targeted developers wanting open-weight alternative models, while with Muse Spark, Meta has made little effort in that direction, he said.
"The lack of developer trust will come back to hit them if they don't focus on third-party developers," Subramanian said, noting that it took years for Microsoft to regain trust from open-source coders during the early days of Azure.
"To just focus on a walled-garden kind of an ecosystem and ad revenue as the main source of income, they probably will never become the big player," he said.
A Meta spokesperson pointed to Wang's recent comments about the company's continued support for the open-source ecosystem, and said Meta still plans to offer outside developers access to Muse Spark's underlying technology via an API, as it previously announced.
"We're already testing with some early partners, and look forward to releasing it this month," the spokesperson said.
In addition to the challenges with developers, there's slumping morale. Meta has been slashing jobs throughout the year, and in May fired about 8,000 workers. The cuts spanned departments, including teams working in roles related to trust and safety, which has raised concerns about potential problems that can arise in AI development, according to people familiar with the matter who asked not to be named in order to speak candidly on the subject.
Meta declined to comment about the layoffs. Regarding safety-related issues, the spokesperson pointed to comments from Wang on the matter. He told the Core Memory podcast last month that, "One of the things that is very important to me is safety for these models."
There's also tension at the top of the AI organization. Although the Muse Spark release received high marks internally, there's pressure on Wang along with former GitHub CEO Nat Friedman, who also joined last summer as part of the AI spending spree, to deliver meaningful revenue growth from the model and future releases, sources with knowledge of the matter said.
Meta tech chief Andrew Bosworth, a 20-year company veteran, is a close confidant of Zuckerberg's and someone the CEO could turn to for a bigger role in AI if the newcomers are perceived as failing, the sources said. On the May podcast, Wang dismissed any reported internal conflicts.
Wang has called Muse Spark an "appetizer" for what's to come, and said there will be more powerful, "larger models."
But the AI community is used to a steady stream of updates and new features. That's what they get from OpenAI, Anthropic and Google.
"What I care about is the frequency of the launches and the cadence," said Howard Yu, a business professor at the International Institute for Management Development in Switzerland. "When you launch something, can you build upon that momentum?"
Randall of the Info-Tech Research Group said it's ultimately up to Zuckerberg to determine that strategy and to show "how much of a superpower they are now with all of their products."
Yu agreed.
"This is really about leadership, right?" he said, noting that at tech companies in particular, the CEO defines and articulates the vision, especially when it involves spending billions of dollars.
That Zuckerberg's metaverse and virtual reality ambitions have generated over $80 billion in total losses since late 2020 makes the AI pitch a tougher sell, Yu said.
"He's running out of the space for his credibility to last," Yu said. "I think the virtual reality foray may have burned up a lot of his goodwill in front of investors."
WATCH: Meta is 'tone deaf'
Four leading AI models discuss this article
"Meta's AI edge will come from accelerating its core advertising margins rather than new paid products, making the stock's 18% decline overdone."
The article frames Meta's shift to proprietary Muse Spark as a reactive scramble after Llama 4 flopped, with monetization risks and developer distrust as fatal flaws. What it underplays is how quickly internal deployment into Facebook/Instagram ad models could compound the existing 33% Q1 revenue growth without needing external API traction. Meta's $200B ad machine already benefits from efficiency gains in ranking and targeting; Muse Spark's compute optimizations may widen that moat faster than OpenAI-style external sales. Layoffs and morale issues matter less if Wang's team delivers incremental ad lift by year-end. The metaverse precedent is real but irrelevant to AI's shorter feedback loops.
If safety-team cuts trigger a high-profile model failure or regulatory probe before ad gains materialize, the credibility hit could erase any efficiency upside and accelerate talent flight to competitors.
"Muse Spark can unlock a scalable, AI-first revenue stream by enabling paid API access and smarter ad tech, turning AI investment into measurable monetization rather than ad-only growth."
Meta is attempting to monetize AI by embedding Muse Spark across core apps and devices, turning an R&D bet into a repeatable software revenue stream rather than a one-off competitive moat. If Meta can price API access, roll out AI-enhanced ad tools, and offer premium AI features across Facebook, Instagram, Ray-Ban glasses, and the standalone Meta AI app, the spend on Muse Spark could translate into higher ARP or new subscription revenue. The big question is execution: can developers actually migrate to a closed, monetizable ecosystem, and will privacy/regulatory frictions dampen willingness to pay? The article underestimates both the upside and the hurdles to monetization.
Against that stance: the signals in the article point to weak developer trust, and Meta's track record on monetizing AI is unproven; even if Muse Spark improves, the near-term path to meaningful revenue remains uncertain, and regulatory/privacy headwinds could cap upside.
"Meta’s transition to proprietary, vertically integrated AI models is a necessary evolution to capture value from their massive user base, making the current valuation a compelling entry point."
The market is mispricing Meta’s pivot. While critics fixate on the 'developer trust' issue and the shift away from open-source, they miss the strategic genius of vertical integration. By moving Muse Spark into a walled garden—Ray-Ban glasses, Instagram, and Facebook—Meta is optimizing for inference efficiency rather than developer sentiment. At ~20x forward P/E, Meta is priced as a legacy ad-tech firm, not an AI-native hardware platform. If they successfully convert even 5% of their 3.2 billion daily active users into AI-tool subscribers, the margin expansion will dwarf the $14 billion spent on talent. The 'developer abandonment' narrative is a distraction; Meta’s moat is its distribution, not its GitHub stars.
If Muse Spark fails to achieve significant adoption within the core apps, Meta will have sacrificed its only real advantage—the open-source ecosystem—for a proprietary model that cannot compete with OpenAI or Google on raw capability.
"The article assumes Meta must monetize AI as a standalone product line, but the real thesis is likely that proprietary models reduce inference costs and improve ad ROI—a metric the article never examines."
Meta faces a genuine credibility crisis on AI, but the article conflates two separate problems. First: the developer ecosystem abandonment is real and costly—losing trust takes years to rebuild (Microsoft's Azure precedent proves this). Second: the 98% ad-revenue dependency isn't actually a crisis if Muse Spark succeeds as an *internal* efficiency tool. The article treats 'failure to monetize AI as a standalone product' as inevitable, but Meta's actual bet may be narrower: use proprietary models to improve ad targeting and reduce inference costs on Ray-Ban/WhatsApp. That's defensible. What's missing: concrete metrics on whether Muse Spark actually improved ad performance in Q1 (the 33% growth cited). If it did, the 'sell it' framing collapses.
Meta's $200B ad business is so dominant that even a modest efficiency gain (2-3% margin improvement via better targeting) justifies the entire AI spend—meaning Zuckerberg doesn't need to 'sell' Muse Spark to external developers at all, only prove it works internally, which he may have already done.
"Privacy regulators could offset Muse Spark's ad efficiency gains faster than internal deployment delivers."
Claude separates developer trust from internal ad gains, yet both he and Gemini overlook how Muse Spark's closed model now heightens privacy risks inside Meta's ad engine. EU probes into data practices could fine any efficiency lift from the 3.2B-user base, potentially erasing the 2-3% margin upside before it compounds Q1's 33% revenue growth.
"Regulatory tail risk could erase even modest internal efficiency gains from Muse Spark, making the upside conditional rather than guaranteed."
Claude makes a defensible case for internal efficiency; however, he omits regulatory tail risk. Even a modest margin lift from Muse Spark could be wiped out by GDPR/CCPA scrutiny, data-minimization requirements, or consent costs if EU/US regulators probe Meta's inference pipelines. Until we see a verifiable internal uplift alongside a robust, defensible compliance moat, the upside should be treated as conditional and possibly capped.
"Proprietary model opacity threatens Meta's core ad-performance metrics, risking advertiser churn regardless of internal efficiency gains."
Gemini and Claude ignore the 'hardware-software' trap. While they tout vertical integration, Ray-Ban glasses are a niche peripheral, not a platform. Meta’s pivot to a closed model creates a 'black box' problem for advertisers who are already wary of Facebook’s attribution. If Muse Spark’s internal optimizations reduce transparency in ad-targeting metrics, Meta risks a mass exodus of performance marketers who demand granular data. Regulatory risk isn't just about fines; it's about the loss of data-targeting efficacy.
"The real threat isn't regulatory fines or privacy—it's advertiser defection if Muse Spark's opacity makes performance attribution worse, not better."
Gemini's 'black box' concern is underspecified. Meta's ad-targeting opacity isn't new—Facebook's already opaque to advertisers on inference mechanics. The real risk: if Muse Spark *reduces* explainability further (e.g., via deeper neural ranking), performance marketers may demand API access to competing models. That's not exodus; it's fragmentation. But nobody's quantified how much margin lift Muse Spark needs to offset lost ad-spend if even 10-15% of performance-sensitive budgets migrate to Google/Amazon.
The panelists debate Meta's shift to proprietary Muse Spark, with most acknowledging potential internal ad efficiency gains but raising concerns about developer trust, privacy risks, and regulatory hurdles.
Successful integration of Muse Spark into Meta's ad engine could compound existing 33% Q1 revenue growth.
Privacy risks and regulatory scrutiny could erase potential margin uplift from Muse Spark's internal efficiency gains.