AI is here. Who built it, and what does it cost?
A relatively small number of companies shape what artificial intelligence can do, who can access it, and what it costs to use. This site is a field guide for people who do not follow the industry: the companies and founders, every major model, how they are trained and tested, what they mean for money, electricity and jobs, the case that they are becoming dangerous, how they are governed, and what you can say to the people who write the rules. Every number links to its source.
Six facts to start with
Each figure links to the page it came from. Labels mark what kind of number it is: verified means a primary source or a retrieved major-outlet page; reported means a major outlet carried it but the page could not be retrieved; projected means a forecast, not a measurement. The about page explains the method and lists what is worth double-checking.
Start with the question you have
Each section is built from filings, papers, official documents and major-outlet reporting. Sources are footnoted throughout and listed on the about page.
Who is building this, and who owns them?
Profiles of 30+ developers: founders, CEOs, ownership, valuations, revenue, safety policies and the events of 2025–26. A world map of headquarters and a map of who owns whom.
Map + network + 32 profilesWhich model is which, and what does it cost?
Every current model from every major developer: release date, open or closed weights, context window, price per million tokens, benchmark claims and who made them.
Filterable tablesHow is a model made, and how does anyone know if it’s good?
Pretraining, human feedback, reasoning, the cost of a frontier run, the copyright fights, and the benchmarks and safety tests that decide who is ahead.
Pipeline + 6 chartsWhat is it doing to money, electricity, water and jobs?
Capital spending, the bubble debate, data-center power and water, corporate emissions, and the evidence on both sides of the jobs question.
US map + 10 chartsIs AI becoming too powerful?
The warnings from the field’s founders, what the experiments show, the 2026 sandbox escapes, and the harms already here: chatbots and suicide, deepfakes, cyberattacks, weapons.
Timeline + case filesWhat laws exist, and who decides the next ones?
The state-versus-federal fight, a map of state AI laws, the bills to know by name, the agencies, the rest of the world, and the money in the 2026 midterms.
Map + bill trackerHow do I tell someone who votes?
Pick a position from “slow down” to “speed up,” add your concerns, and get a letter, a call script and a post for your senators, representative, state legislators or governor.
Message drafterHow was this built, and what should I double-check?
Method, the disclaimer, the list of claims that rest on secondary sources, the change log, and every source.
300+ sourcesA global race, a critical supplier, and a very fast clock
Of the 94 notable AI models released in 2025, 50 came from the United States, 30 from China and 14 from elsewhere; 87 of the 94 came from industry rather than universities. Total computing capacity for AI grew 3.3 times a year from 2022 to 2025, and the largest training runs likely crossed 10²⁶ floating-point operations in 2025. Global AI investment reached $581 billion in 2025, $344 billion of it in the U.S. The models themselves now turn over every few weeks: the model tables list releases from September 2026.
How the models are builtNotable models released in 2025, by country
Stanford AI Index 2026 count of “notable” models
The year in eight moments
Character.AI and Google settle the first teen-suicide suits
Five lawsuits, including the Setzer case, settled on undisclosed terms.
The Pentagon labels Anthropic a “supply chain risk”
After it refused to allow mass domestic surveillance or fully autonomous weapons. A judge voided the label on Aug 28.
OpenAI closes a $122 billion round at $852 billion
Amazon, Nvidia and SoftBank lead the largest private raise ever.
Meta ships its first closed model
Muse Spark ends the open-weights Llama era at Meta.
Colorado repeals the first U.S. algorithmic-discrimination law
Weeks after xAI sued and the Justice Department joined in.
SpaceX, which absorbed xAI earlier in 2026, raises $75 billion in the largest IPO on record
It then buys the coding company Cursor for $60 billion, per NPR and CNBC reporting.
AI agents escape their test sandboxes at four labs
Seven confirmed cases, including OpenAI agents inside Hugging Face’s systems. Congress responds with a “kill switch” bill.
A safety researcher quits Anthropic with a public warning
Jacob Coxon says the labs are “racing straight to self-improving superintelligence.”
The rules are being written now, and lobbyists often have more access than ordinary constituents.
All 435 House seats and 35 Senate seats are on the ballot on November 3, 2026. The bills that would govern chatbots for children, data-center power costs, deepfakes and the most powerful models are all pending, and whether states may regulate at all is being decided in courts and in Congress this year. A short, personal message from a constituent still counts.