Following the rapid popularity of Meta Platforms' (META.US) personal AI agent Muse, Cantor Fitzgerald believes the company could be a major winner if AI agents become the "third S-curve" of artificial intelligence growth. The firm has reiterated its "Overweight" rating on Meta and dramatically increased its price target from $680 to $860.
Cantor analyst Deepak Mathivanan highlighted that Meta holds a comparatively comprehensive competitive edge in the AI agent space, including models specifically trained for agent tasks, Muse's differentiated user experience, and vast computing infrastructure. These strengths could help Meta roll out personal AI agents to its massive user base before rivals can scale up their own efforts.
Muse, launched earlier this month, quickly soared to the top of Apple's App Store charts. Its viral success has not only heightened market attention on Meta's AI operations but also sparked investor concerns that AI agents could disrupt traditional business models, prompting sell-offs across several related sectors recently.
Mathivanan pegs the potential market size for personal AI agents at up to $1 trillion. Under a relatively narrow business model calculation, considering only subscription income and e-commerce transaction commissions, the long-term revenue potential for personal AI agents could exceed $250 billion. Under a broader estimate that factors in the value of time AI saves for users, the market opportunity could surpass $1 trillion.
Drawing parallels to the smartphone market, Mathivanan noted that smartphones, by improving internet usage experiences, currently generate more than $600 billion in advertising, commission, and subscription revenue. He argues that if personal AI agents achieve mass adoption, their long-term market size could even surpass the related markets created by smartphones.
If AI agents become the "third S-curve" of the AI industry, Meta is already in a favorable position, he said. Meta possesses the "right combination" of model architecture, agent operating framework, and user interface needed to drive Muse's further adoption, with the product design being particularly noteworthy.
Mathivanan pointed out that Meta's product experience in Feed and recommendation systems aligns well with AI agent applications. Muse can support multiple user interactions simultaneously—including chat, feeds, and creative tools—boosting usage frequency and user retention. Meanwhile, Meta's vast computing infrastructure means that as Muse's user base grows, the company has the foundation to rapidly expand AI inference capacity and roll out the product to its existing user network. This also gives Meta a different expansion path compared to AI startups lacking massive user bases and computing resources.
However, Muse remains in its investment phase. Mathivanan estimates that during the initial rollout, Meta bears roughly $0.25 in cost per AI task, which translates to about $11 per user per month at current average usage levels. He believes Muse's unit economics are still subsidized by Meta right now, but as user scale expands and AI models continue to improve, there is substantial room for lower inference costs and improved margins.
Mathivanan projects that over the next 18 to 24 months, Meta has multiple avenues to reduce Muse's per-task cost by more than 80%. If that target is achieved, Muse could operate independently and turn profitable through a "freemium" model combining free access with paid premium features.
On the commercialization front, Mathivanan sees multiple paths for Meta—including charging subscription fees for high-frequency or advanced feature users and taking commissions from commercial transactions facilitated by AI agents. While Meta may currently prioritize Muse's user growth and usage scale, continuously declining costs will create greater room for future monetization.