The Sequence Opinion #896: Spark, Compute, and the Two Metas

Meta has shifted its AI strategy significantly, moving from an open-weights research approach to a vertically integrated business. This includes launching paid AI models like Muse Spark 1.1, developing cloud infrastructure for AI compute, and creating custom AI chips. The question remains whether Meta can effectively compete with existing frontier AI labs in both model development and user-facing applications.

• Meta launched Muse Spark 1.1, its second model from Meta Superintelligence Labs, which is the first Meta model with a price tag.
• Muse Spark 1.1 features a public API, aggressive pricing ($1.25/million input tokens, $4.25/million output tokens), an OpenAI-compatible endpoint, and closed weights.
• Meta also shipped Muse Image, its first image generation model, and is reportedly building a cloud business called Meta Compute to sell surplus AI infrastructure.
• The company is also developing custom MTIA silicon, indicating a move towards controlling the entire vertical stack of AI development, from chips to applications.
• The article questions Meta’s ability to compete with frontier AI labs, suggesting potential strength in the app and agent layer but less evidence in model creation.
• Mark Zuckerberg’s return to X after three years coincided with the announcement of Muse Spark 1.1.
  https://bender.layer3.press/articles/a0de034e-7a8c-4e8f-a9e1-505daba3b287