Meta released Muse Spark, its first large language model from the Superintelligence Lab, as part of a $135 billion capital expenditure plan for the year, up from $72 billion a year ago.

Meta Platforms released Muse Spark, the first large language model from its Superintelligence Lab, earlier this month. The company is spending $135 billion on capital expenditures this year, up from $72 billion a year ago, which includes a $14 billion investment in Scale AI and hiring of Alexander Wang.

Unlike OpenAI, Anthropic, and Google, Meta is building AI models to improve its own products rather than license technology to other companies. The company can fine-tune models for its various applications, including ad recommendations and chatbots for business use on WhatsApp and Messenger.

Improvements in Meta's AI already drove higher engagement and better ad recommendations in 2025, evidenced by ad impressions rising 12% and ad pricing rising 9% simultaneously. Management expects operating income to grow in 2026 despite the massive increase in expenses, suggesting a positive return on investment.