Who is building this, and who owns them
A few dozen organizations build the models the world uses: a handful of well-funded labs in San Francisco, the largest technology companies on earth, a fast-moving group in China, a European champion, and specialists in images, video and voice. Below is a profile of each, built from filings, company announcements and major-outlet reporting: who founded it, who runs it, who owns it, what it is worth, what it has promised about safety, and the events that shaped it. Figures are labeled verified when they come from a primary source or a fetched major-outlet report, reported when a major outlet reported them but the page could not be retrieved, and unverified when only secondary compilations carry them.
Headquarters of the major developers
Nearly everything is within a few miles of San Francisco Bay, with clusters in London, Paris, Beijing, Hangzhou and Shanghai. Use the buttons below the map to zoom into a cluster; names appear as you zoom in, and clicking a marker jumps to that company’s profile.
What the market says they are worth
Latest market capitalization or private valuation, US$ billions, log scale
Public companies at Sept 2026 market capitalization (Nvidia, Alphabet and SpaceX include far more than AI). Private companies at their most recent round; “reported” figures rest on major-outlet reporting that could not be retrieved directly and are marked with a lighter bar. See worth double-checking.
Three tiers and one supplier
Frontier labs train the largest models: OpenAI, Anthropic, Google DeepMind, xAI (which SpaceX acquired in 2026, according to reporting), Meta’s Superintelligence Labs, and newer entrants such as Safe Superintelligence, Thinking Machines and Reflection that have raised billions before shipping much. Big Tech supplies the money, the cloud and the distribution: Microsoft owns about 27% of OpenAI; Amazon and Google are Anthropic’s largest outside investors; Apple licensed Google’s Gemini for Siri. Chinese labs (DeepSeek, Alibaba’s Qwen team, Moonshot, Zhipu, ByteDance, MiniMax, Tencent, Baidu) now ship most of the world’s open-weight frontier models, and two of them listed in Hong Kong in January 2026, before any U.S. lab went public.
Above all of them sits Nvidia, whose chips power most training and which has become the most valuable company in history, at more than $5 trillion, while also investing in OpenAI, Safe Superintelligence, CoreWeave and, in August 2026, buying the open-model hub Hugging Face. The same dollars circulate: Nvidia invests in OpenAI, which commits to buy Nvidia chips through Oracle and CoreWeave, which borrow against those contracts. The impacts page takes that up.
The web of stakes, deals and acquisitions
Drag the nodes. Solid blue lines are ownership stakes, teal are investments, red are acquisitions, amber dashed are supply or licensing partnerships.
The companies, one by one
Chips, clouds and the Stargate project
The companies that sell to the labs, briefly.
The models these companies make
Every current model from every developer above, with dates, licenses, prices and benchmark claims, in one filterable table.