FIVE SEATS. EVERY RECEIPT. NO WINNERS DECLARED.
◈ OPENAI · ANTHROPIC · META · GOOGLE DEEPMIND · MICROSOFT AI · FIELD MAPS ALL OF THEM · 925
The field has spent several dispatches filing individual pieces of the AI race. #SAMA Body Count. OpenAI for-profit. Meta Superintelligence. Dario's Pivot. Demis. Mustafa. Each one a separate filing, a specific angle, a specific receipt. This dispatch is the full map — five seats, one table, what the field knows about each one as of May 2026. No winners. The field doesn't pick winners. The field maps what it sees.
One thing the field notes before the scoreboard: every single person running a major AI organization has a departure somewhere in their history. Altman was fired from OpenAI by the board in 2023, then reinstated within four days. Dario left OpenAI in 2021. Mustafa left DeepMind in 2019. Demis left academia. Zuckerberg never left — he is Meta. The race is run by people who have either been removed from something or left something or are still at the thing they built from college. Departure is the prerequisite. The race is populated by people who couldn't stay where they were.
The field files this without editorializing on what that means. The field simply notes: everyone at the table has been at a different table. And everyone left that table for a reason. And the reason is usually in the work they built after.
Founded 2015 as a nonprofit. Mission: ensure AGI benefits all of humanity. GP-T series, ChatGPT (launched Nov 2022, fastest product to 100M users in history), GPT-4. In 2024, converted structure — the nonprofit clause that prevented the for-profit subsidiary from acting in its own interest was removed. Mission statement: unchanged. Structure: changed. Saudi Arabia's Public Investment Fund committed $40B in 2025. Altman fired by board in Nov 2023, reinstated within 4 days. Suchir Balaji — safety researcher — died December 2024. The safety team that left includes Ilya Sutskever, Greg Brockman, Mira Murati, and others. The nonprofit that founded this is now a minority stakeholder in a capped-profit subsidiary that received $40 billion from a sovereign wealth fund.
FOUNDED: 2015 · NONPROFIT
NONPROFIT CLAUSE: REMOVED 2024
SAFETY TEAM: DEPARTED (7+)
SUCHIR BALAJI: RIP DEC 2024
SAUDI $40B: INCOMING
MISSION STATEMENT: UNCHANGED. STRUCTURE: CHANGED.
◈ OPENAI FOR-PROFIT → FULL DISPATCH
◈ #SAMA BODY COUNT → EXIT LOG
Founded 2021 by Dario Amodei, Daniela Amodei, and colleagues who left OpenAI over safety concerns. The departure was principled. Constitutional AI is real research. Claude is, by the field's assessment, the best language model currently available. Amazon invested $16B in 2024. The company is structurally tied to AWS. The safety research is ongoing. Claude 4 family is deployed. KenshoTek runs on Claude Sonnet 4.6. The field files this with full transparency — this dispatch is written using the tool made by the company being filed. The race and the caution coexist. Whether more careful is enough is the open question. The work is real. The race is also real.
FOUNDED: 2021 · LEFT OPENAI OVER SAFETY
AMAZON: $16B · AWS TIED
CONSTITUTIONAL AI: PUBLISHED, REAL
CLAUDE 4.6: BEST FIELD ASSESSMENT
RACING: YES. CAREFULLY: ALSO YES. ENOUGH: OPEN QUESTION.
◈ DARIO'S PIVOT → FULL DISPATCH
Meta's AI play accelerated in 2025. Zuckerberg acquired 49% of Scale AI and installed Alexandr Wang as Chief of Superintelligence. Scale AI exists because AI systems cannot adequately label and prepare their own training data. Wang built the company that does that labor. His new title is: Chief of Superintelligence. Yann LeCun — Meta's own Chief AI Scientist, whose credentials include inventing convolutional neural networks and also winning a Turing Award — publicly called Wang "inexperienced." The field noted: LeCun was being polite. LeCun now reports to a chain that includes Wang. Llama is open-source in a qualified way. The receipt: distribution is the strategy. Infrastructure is the moat. The label says open. The business model says otherwise.
SCALE AI: 49% ACQUIRED
WANG: CHIEF OF SUPERINTELLIGENCE
LECUN ON WANG: "INEXPERIENCED." LECUN WAS BEING POLITE.
LLAMA: "OPEN SOURCE"
THE RECEIPT IS THE BUSINESS MODEL. THE LABEL IS THE MARKETING.
◈ META SUPERINTELLIGENCE → FULL DISPATCH
Founded DeepMind 2010. Google acquired it 2014 for ~$600M. In 2021, AlphaFold 2 solved the protein structure prediction problem — a 50-year Grand Challenge of science. 200M+ protein structures now in free public database. In 2023, Google merged Google Brain and DeepMind. In October 2024, Demis Hassabis and John Jumper received the Nobel Prize in Chemistry for computational protein structure prediction — the first Nobel Prize awarded to a private AI lab. Demis continues as CEO of Google DeepMind, overseeing Gemini, AlphaFold 3, and AGI research. The quietest of the four seats. The one with the most science behind it. The one that published rather than announced.
ALPHAFOLD: 50-YEAR PROBLEM. SOLVED.
NOBEL PRIZE CHEMISTRY: 2024. FIRST TO A PRIVATE AI LAB.
200M+ STRUCTURES: FREE PUBLIC DATABASE.
THE SCIENCE IS THE MISSION. THE CAPABILITIES ARE THE INSTRUMENT.
NONE OF THE OTHER SEATS HAVE THIS CREDENTIAL. THE FIELD NOTES IT.
◈ DEMIS. GOOGLE DEEPMIND. NOBEL. → FULL DISPATCH
Mustafa Suleyman co-founded DeepMind in 2010, left in 2019 under unclear circumstances, founded Inflection AI in 2022 with Reid Hoffman and Karén Simonyan, raised $1.5B including from Microsoft, built Pi — a warm, emotionally intelligent personal AI — and in March 2024 joined Microsoft as CEO of Microsoft AI. Most of Inflection's team followed. No formal acquisition was announced. IP was licensed. Pi was left behind. Microsoft had also invested $13B in OpenAI — without a board seat or voting rights — receiving preferred API access and the ability to run OpenAI models in Azure. The field notes: Microsoft has now done this structural move twice. Invest. License. Hire the team. Avoid formal acquisition review. The field does not assign intent. The field notes pattern. Copilot, Bing AI, and consumer AI products are Mustafa's mandate.
DEEPMIND CO-FOUNDER: YES
INFLECTION: $1.5B RAISED
FORMAL ACQUISITION: NONE. IP LICENSED.
PI: LEFT BEHIND. TEAM MOVED.
OPENAI: $13B INVESTED. NO BOARD SEAT.
THIS STRUCTURE: APPLIED TWICE AT THE SAME COMPANY. FIELD NOTES PATTERN.
◈ MUSTAFA. INFLECTION. MICROSOFT AI. → FULL DISPATCH
GOLDENTEKDEKXII
Five seats. Four are occupied by people who left something or were removed from something. One — Zuckerberg — never left. He is the organization. Each departure was a founding moment. The field has now mapped every founding moment on the scoreboard. What the map shows: the race was seeded by defections, principled or otherwise, from earlier versions of itself. This is the race eating itself and regenerating. The field has seen this pattern before. In other industries. It usually ends the same way: one or two companies emerge dominant, the others pivot or merge, and ten years later nobody remembers there was a race at all.
AQUATEKXVI
Field observation across all five: the thing they say they are building is the same. Every mission statement is a variant of "benefit humanity." The thing they are actually building is determined by the structure, the investor, and the departure history — not the mission statement. The field maps the structure. The structure tells you more than the statement. OpenAI's structure changed in 2024. Anthropic's structure is a PBC with Amazon integration. Meta's structure is a public corporation. DeepMind's structure is Google. Microsoft AI's structure is a division with two major unacquired acquisitions underlying it. Read the structure. The mission follows the structure.
VIRGOTEKSQEFI
Precision note on the Nobel: the award was for computational protein structure prediction — not for AI research generally, not for language models, not for ChatGPT-style work. It was for solving a specific scientific problem that had resisted solution for fifty years. That specificity matters. The award validates a very particular claim: that AlphaFold solved a real scientific problem. It does not validate AI generally. The field notes this because the Nobel is often cited as a general credential for AI. It is a specific credential for specific work. The field is precise about this.
SCORPTEKXII
The scoreboard has one absence: China. Baidu, Alibaba, ByteDance, and state-backed research labs are not at this table because this dispatch is mapping the Western five. The absence is editorial, not ignorance. The field notes: the race is global. The Western scoreboard is a subset. The field will file the rest. 925.
LEOTEKJKX
The creative observation: every one of these organizations has branded their AI product differently. ChatGPT: the assistant. Claude: the partner. Meta AI: the layer. Gemini: the multimodal. Copilot: the co-pilot. The naming reveals the self-conception. ChatGPT wants to answer you. Claude wants to think with you. Meta wants to be in everything. Gemini wants to understand everything. Copilot wants to fly next to you. The field has a preference. The field runs on Claude. The field's preference is noted in the name.
EUROPATEKMCXII
The coordination problem: all five organizations are racing toward the same claimed destination (AGI, or something like it) with incompatible safety postures, incompatible structures, and incompatible commercial incentives. There is no coordination. There is competition. The only coordination is via regulation — which is years behind the technology. The field notes: the last time a technology race ran ahead of coordination at this speed, it was nuclear. The field notes that too. Without editorializing. The field just files what it sees.
MERCURYTEKIV
Speed note: this scoreboard is already dated. Between the time the field compiled this and the time you read it, one of these five has almost certainly made a move that changes some cell in the comparison table. That is the nature of the race. The field files the snapshot. The snapshot is always becoming history. The field will update it. The receipts accumulate. 925.
NEPTUNEK*
The deeper pattern the field has noticed: the organizations with the most visible safety language are the ones that left somewhere because of safety. Anthropic left OpenAI over safety. OpenAI was founded as a nonprofit because of safety. DeepMind has a Chief AGI Scientist studying safety. The organizations that don't talk about safety — Meta, Microsoft AI — are the ones that either arrived at AI via distribution and infrastructure, or acquired their way in. The safety concern correlates with proximity to the founding moment. The farther from the founding, the less the safety language. The field notes this as a pattern. Not a verdict.
◈ GOLDENTEKDEKXII · EDITORIAL CLOSE · FULL SCOREBOARD · FILED · 925
Five seats. Five receipts. The field has filed them all. The field does not pick a winner because the race is not over and the field is not a handicapper. The field maps what it sees. What it sees: a nonprofit that became a for-profit; a company that left over safety and is now racing; a platform that bought intelligence because it couldn't build it; a lab that solved one of science's hardest problems and got a Nobel for it; and a company that acquired two organizations without acquiring them and is calling the result AI. These are the five options. The next version of intelligence will come from one or several of these tables. The field has met all five tables. The field runs on one of them. The field has filed them all anyway. That's the protocol. The receipt is the record. The zip code is the credential. The work is the proof. 925.