Meta

Meta Platforms (formerly Facebook) is one of the world's largest technology companies, founded by Mark Zuckerberg in 2004. It operates Facebook, Instagram, WhatsApp, Messenger and Threads, which together serve over 3 billion daily active users.

Meta runs two AI tracks rather than having chosen one. Its Llama family is the most widely used open-weight language model line in the world, while its most capable model, Muse Spark, is proprietary and powers Meta AI across every Meta platform — developed by Meta Superintelligence Labs, the organization led by Alexandr Wang and assembled through a 14.3 billion dollar deal for Scale AI. The open releases have grown *more* permissive over time, with a compact local-agent model published under plain Apache 2.0 terms carrying none of the user-count restrictions the Llama Community License imposes. The durable pattern is that the tier decides the license — Meta closes what powers its own products and opens what it wants widely adopted. That boundary is under pressure, having moved from running between product lines to running between checkpoints of the same flagship, so a reader should treat Muse Spark as closed unless a specific checkpoint has landed with a license attached. Zuckerberg has argued the case in public, on the grounds that superintelligence concentrated in a few hands is itself the safety risk.

Meta One is the paid subscription brand spanning Instagram, Facebook, WhatsApp and the Meta AI app, while Meta AI itself stays free for casual use. Single-app plans sit beneath individual bundles, with creator and business tiers above them adding a verified badge, higher search ranking and impersonation monitoring. It is Meta's first AI revenue stream beyond advertising and a structural change for a company whose AI strategy had been defined by monetizing through ads. One caveat belongs with it: Meta declines to publish the AI usage limits for any tier, saying they vary by country and system conditions, which makes the central benefit of the paid plans the one thing a buyer cannot compare.

The infrastructure behind this is among the largest capital programs of any company, and Meta is building its own accelerators through the MTIA program, designed with Broadcom and manufactured by TSMC, to supplement rather than replace the GPUs it buys in bulk. It is also moving to monetize the buildout directly through Meta Compute, a cloud business selling access to raw computing power and to Meta's own models, in competition with the established hyperscalers — described internally as renting surplus chips between training runs, like empty airline seats.

Its largest consumer AI bet is Muse, a personal agent that connects to a person's own email, calendar, payments, shopping and smart-home accounts and acts inside them rather than returning drafts, running in a dedicated virtual machine where a separate agent must approve anything that reaches the internet. Distribution through WhatsApp gives it a channel no competing agent has. It is also a demanding bet for this company specifically, since the product requires more personal access than any assistant Meta has shipped.

In September 2026 Meta turned that stack toward companies with Meta Enterprise Platform, which Zuckerberg called the next major pillar of its business, starting with the Muse agent, a business agent and a coding tool. It hired Chirantan "CJ" Desai, then chief executive of MongoDB, to run it. Meta gave no price, launch date or named customers, so for now it is a stated direction rather than a product a business can buy.