Meta

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

Meta has become one of the most important players in AI through its open-source Llama model family and the Meta AI assistant. Llama models (including Llama 4 Maverick and Llama 4 Scout with a 10 million token context window) are the most widely used open-weight language models in the world. In April 2026, Meta announced Muse Spark — its new proprietary flagship model from Meta Superintelligence Labs, a new organization led by Alexandr Wang (former Scale AI CEO, hired via a $14.3 billion deal). Muse Spark accepts voice, text, and image inputs and powers Meta AI across all Meta platforms. Meta runs both tracks rather than having chosen one. Its most capable model is proprietary, while its open releases have grown *more* permissive: in August 2026 Meta published Muse Glimmer, a 30 billion parameter model for local agents, under a plain Apache 2.0 license with none of the user-count restrictions the Llama Community License carries. Zuckerberg set out the reasoning in a 6,500-word essay, *The Future is for Everyone*, arguing that superintelligence concentrated in a few hands is itself the safety risk, and committing $1 billion to a fund for communities near Meta's data centers. 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 pattern came under pressure almost immediately. On September 2, 2026 Meta shipped Muse Spark 1.3 — closed, through the Meta Model API — and Zuckerberg promised open weights for Muse Spark "soon" without naming a version. Meta has not decided whether to publish 1.3's weights and still plans to release only those for 1.2, so the closed-versus-open boundary now runs between checkpoints of the same flagship rather than cleanly between product lines. Until a specific checkpoint lands with a license attached, treat Muse Spark as a closed model that Meta has expressed an intention to open.

Meta One is Meta's paid subscription brand, spanning Instagram, Facebook, WhatsApp, and the Meta AI app. Meta AI remains free for casual users. It launched in the United States in May 2026 and went global on September 15, 2026, by which point Meta reported more than 50 features and 15 million subscriptions and trials. The naming changed with that rollout, and the earlier labels no longer describe the lineup: "Plus" now refers to the single-product plans for one app — Instagram Plus, Facebook Plus and WhatsApp Plus, from $2.99 a month — while the individual bundles are Meta One Core at $7.99 and Meta One Premium at $19.99, both combining those app subscriptions with heavier use of Meta's most compute-intensive AI, including image and video generation through the Muse models. Creator and business plans run from Essential at $14.99 through Advanced at $49.99 and Expert at $149 to Max at $499 a month, adding a verified badge, a bold follow button, higher search ranking and impersonation monitoring. Meta declines to publish the AI usage limits for any tier, saying they vary by country, surface and system conditions, which makes the central benefit of the paid plans the one thing a buyer cannot compare. The shift marks Meta's first revenue stream for AI beyond advertising, a structural change for a company whose AI strategy had previously been defined by the choice to monetize through ads rather than subscriptions.

Meta has also signaled a move into premium AI agents with Hatch, an announced consumer agent that builds software tools and handles tasks like scheduling and email from plain-language prompts. A consumer version of Meta's OpenClaw agent, Hatch is expected to launch in the US around July 2026 with free and premium tiers, and Meta is reportedly weighing a price of up to $200 a month for the top tier — which would make it the company's first standalone paid AI agent, positioned directly against premium subscriptions from OpenAI and Anthropic.

Meta's 2026 AI capital expenditure is projected at $115 to $135 billion — among the largest infrastructure investments by any company. To control that spend, Meta is building its own AI accelerators: the MTIA program’s fourth-generation chip, code-named Iris — designed with Broadcom and manufactured by TSMC — is slated to enter production in 2026 to supplement, not replace, the Nvidia and AMD GPUs Meta buys in bulk, as the company scales toward roughly 14 gigawatts of compute by 2027. The company's FAIR research lab continues producing influential AI research. Llama 4 Behemoth (288 billion active parameters) has been deprioritized in favor of the Muse Spark roadmap.

Meta is also moving to monetize its enormous data-center buildout directly. The company began developing Meta Compute, a cloud infrastructure business that would sell access to raw AI computing power and to Meta's own models — including Muse Spark — putting it in competition with Amazon Web Services, Microsoft Azure, and Google Cloud. Because Meta earns little direct revenue from its models today, renting out excess capacity is a way to recoup its infrastructure spend, echoing similar compute-leasing moves by SpaceX's xAI unit. The effort is led by infrastructure chief Santosh Janardhan alongside Meta Superintelligence Labs' Daniel Gross and company president Dina Powell McCormick, and marks Meta's push to turn a cost center into a business. That buildout is now crossing borders: in July 2026 Meta announced a roughly 10 billion dollar, one-gigawatt data center north of Edmonton, Alberta — its first in Canada and its 33rd worldwide — expected online within three years. Meta has described the strategy as renting surplus graphics chips to outside customers between its own AI training runs, treating idle compute like empty airline seats.

In September 2026 Meta launched Muse, a personal AI agent and its largest consumer AI bet to date. Unlike the Meta AI assistant embedded across its apps, Muse connects to a person's own email, calendar, payments, shopping and smart-home accounts and acts inside them rather than returning drafts, running on Muse Spark inside a dedicated virtual machine where a separate Sentinel agent must approve anything that reaches the internet. It launched in the United States on iOS, Android, muse.ai and inside WhatsApp — the last of which gives Meta a distribution channel no competing agent has. It is also a demanding bet for this company in particular: the product needs more personal access than any assistant Meta has shipped, and it arrived less than two weeks after Meta settled multistate social-media harm claims for 18 billion dollars.