Meta Pivots to Proprietary AI with Muse Spark Launch Amid Massive Capex Surge

Meta has officially entered the proprietary AI race with the release of Muse Spark, marking a departure from its previous commitment to open-source models. The launch comes after a year of intensive hiring and infrastructure investment, aimed at proving the company's AI capabilities to investors after a period of relative silence in product releases. The strategic pivot follows the perceived underperformance of Llama 4 among developers. To spearhead this overhaul, Meta spent over $14 billion to recruit Alexandr Wang and a team of elite researchers to form the Meta Superintelligence Labs, signaling a major shift in the company's technical direction. The company's financial commitment to AI is stark. Meta plans to invest between $115 billion and $135 billion in capital expenditures this year, nearly doubling its 2025 spending. This aggressive investment is aimed at building the infrastructure necessary to compete with the likes of OpenAI, Anthropic, and Google. While Meta intends to offer paid API access to third parties via a private preview, analysts suggest the real value lies in its advertising engine. With 98% of its $200 billion annual revenue derived from ads, the primary goal for Muse Spark is to enhance targeting and engagement for its 3 billion monthly users across Facebook, Instagram, and WhatsApp. Early technical benchmarks indicate that Muse Spark excels specifically in image and video processing. By integrating these capabilities into its core platforms, Meta aims to leverage its existing ecosystem to create a sustainable revenue stream that rivals the subscription and cloud models of its competitors.