Artificial Intelligence · Profile from AI Research Companion
Meta
The Menlo Park owner of Facebook, Instagram, WhatsApp and Messenger (~3.5bn daily users across its apps, company), which funds AI from the most profitable advertising machine outside Google, built one of the field's great research labs (FAIR, 2013) and its most-used framework (PyTorch), and between 2023 and 2025 made open-weight models (Llama) the industry's default alternative to closed labs. After Llama 4 disappointed in April 2025, Mark Zuckerberg changed course at enormous cost: a ~$14.3bn payment for 49% of Scale_AI to bring in Alexandr Wang, a hiring spree with reported nine-figure packages, a new Meta Superintelligence Labs, a turn toward closed models (Muse Spark, April 2026) and capex of $135–145bn in 2026, enough to push its free cash flow negative in Q3 2026 (company). Meta is the purest test in the industry of whether AI spending can be paid for by a product the company already owns.
At a glance
- Founded
- February 2004 as TheFacebook (Mark Zuckerberg, Eduardo Saverin, Dustin Moskovitz, Andrew McCollum, Chris Hughes, Harvard); renamed Meta Platforms October 28, 2021
- Headquarters
- Menlo Park, California
- Ownership / structure
- Public (Nasdaq: META) since May 2012. Dual-class shares: Mark Zuckerberg controls a majority of votes through Class B stock (reported ~61%, verify) while owning ~13% of economic interest (reported, verify)
- Leadership (2026)
- Mark Zuckerberg (founder, chair, CEO); Susan Li (CFO); Javier Olivan (COO); Andrew Bosworth (CTO, Reality Labs); Chris Cox (chief product officer); Alexandr Wang (Chief AI Officer, leads Meta Superintelligence Labs); Nat Friedman (AI products, verify current role); Shengjia Zhao (chief scientist, MSL)
- AI products
- Meta AI assistant (in WhatsApp, Instagram, Facebook, Messenger, the Meta AI app and Ray-Ban/Oakley glasses); Llama open-weight models (Llama 2–4); Muse Spark (closed, April 2026); ad-ranking and recommendation systems (Andromeda, GEM, Lattice); Advantage+ automated ad products; generative ad creative; PyTorch (now under the PyTorch Foundation)
- Scale
- Revenue ~$164.5bn (2024); ~$64.6bn in Q3 2026 alone (+26%), with Q4 2026 guided at $73–76bn (company). Meta AI ~1bn monthly users (company claim, May 2025). Llama downloads passed ~1bn (company, March 2025)
- Capex
- ~$39bn (2024); ~$72bn (2025); $135–145bn for 2026 including finance-lease principal (narrowed October 2026); Q3 2026 capex ~$34.9bn; Q3 2026 free cash flow ~−$2.1bn (company)
- Key AI deals
- Scale AI 49% (~$14.3bn, June 2025); Hyperion campus SPV with Blue Owl (~$27bn, October 2025); ~$30bn bond (October 2025); Google Cloud (>$10bn over six years, August 2025, reported) and TPU rental (February 2026, reported); CoreWeave (~$14.2bn, September 2025, expanded ~$21bn April 2026, reported); Nebius (~$3bn, November 2025; up to ~$27bn more, March 16, 2026); Crusoe (~1.6 GW at Childress and Warrenton, June 2026, reported); AMD (6 GW with a warrant over up to 160m shares, February 2026)
- Signature moves (incl. failures)
- FAIR and open research (2013–); the LLaMA leak and Llama 2 open weights (2023); the "year of efficiency" turned AI capex race (2023–26); Llama 3.1 and the open-source manifesto (2024); Llama 4's reception and the LMArena controversy (2025); the Scale AI stake (2025); Superintelligence Labs and the hiring spree (2025); Hyperion and off-balance-sheet financing (2025); Muse Spark, the Avocado delay and the closed-model shift (2026)
9 strategic moves analysed
Each one gets a full strategy file in the companion: context, the decision, the financing, what happened, and the lesson.
2013
FAIR and Open Research
Founding (research lab) / launch (PyTorch)
2023
LLaMA Leak and Llama 2 Open Weights
Model release
2023
Year of Efficiency to AI Capex
Restructuring / compute deal (capital allocation)
2024
Llama 3 1 and the Open Source Manifesto
Model release
2025
Hyperion Blue Owl Financing
Compute deal (project financing)
2025
Llama 4 Reception and LMArena Controversy
Model release (failure)
2025
Scale AI 49pct Stake
Partnership / licence-and-hire (minority investment plus hire)
2025
Superintelligence Labs and the Hiring Spree
Restructuring (new lab, talent acquisition, reorganisations)
2026
Muse Spark Avocado Delay and Closed Model Shift
Model release / pivot (open weights to closed)