150 billion dollar gamble "turn in the test paper"! Meta's first AI model Muse Spark debuts, with ETFs becoming the focus of fund chasing.

After Meta (META.US) officially released its self-developed AI model Muse Spark, exchange-traded funds (ETFs) heavily invested in the stock became the focus of the market. The AI model named Muse Spark aims to enhance Meta's overall ecosystem by achieving faster and more advanced AI-driven capabilities within its platform. Meta's stock price closed up 6.5% on Wednesday, prompting the market to refocus on ETFs with a high allocation to the stock. Overall, more than 650 ETFs collectively hold 337 million shares of Meta stock, highlighting its significant influence in the fund sector. Currently, those funds with the highest allocation to Meta in their portfolios are attracting special attention from investors seeking targeted exposure. After more than a year of organizational restructuring and substantial investment, Meta has finally delivered a phase report. On Wednesday, Meta officially released its self-developed AI model Muse Spark. Muse Spark is the first core achievement since Meta announced the establishment of the "Meta Super Intelligence Lab" (MSL) last June and the first major AI model released since Meta spent $15 billion to poach Scale AI founder Alexandr Wang. In June 2025, Meta invested approximately $15 billion in Scale AI and lured its founder Alexandr Wang to the company to lead the newly established "Super Intelligence Lab" (MSL). This move was seen as a key step for Meta to "restart its strategy" in the AI competition. At that time, Meta lagged far behind in large model capabilities compared to OpenAI, Alphabet Inc. Class C, and Anthropic. Zuckerberg then initiated a series of radical measures - massive poaching of top AI talent, streamlined organizational structure to improve research and development efficiency, and committed to investing hundreds of billions of dollars to build AI computing infrastructure. Muse Spark is the first "experiment" in this system. Meta states that the model is "specially created" for use in all its products. Meta's CEO Zuckerberg previously stated that the model is not yet capable of matching the most advanced systems of its competitors but will provide "smarter and faster" capabilities for the virtual assistant Meta AI. The company claims that the model will use content from Instagram, Facebook, and Threads to support more personalized and visual responses. Meta also stated that its new model has strong potential for applications in the medical field. The company stated that it has collaborated with over 1,000 doctors to train the model to generate more detailed responses on topics such as nutrition and exercise. With the model, Meta AI will also introduce a "shopping mode" to help users compare prices. According to Meta's disclosure evaluation results, Muse Spark has surpassed leading models from Alphabet Inc. Class C, OpenAI, and Anthropic in some reasoning and multi-modal capability benchmarks. It is worth mentioning that this model deviates from Meta's previous open-source route with Llama and shifts to a closed-source model - the model weights and architectures are not publicly disclosed. This shift is significant. In the past, Meta built its ecosystem influence on open-source models, but now it is moving towards commercialization and competition-oriented approaches. In the future, it may monetize through API fees or even subscription models. It is widely believed in the industry that this marks Meta's transition from "AI evangelist" to "AI competitor" status.