The DeepMind, OpenAI, Anthropic, and xAI saga

2010–2026
Sources are linked under each event. Primary sources appear first when available.
Origins: The fear takes shape
Sep 2010
DeepMind founded in London
Three founders: Demis Hassabis, Shane Legg, and Mustafa Suleyman. Mission: "solve intelligence." Early investors include Peter Thiel's Founders Fund, Elon Musk, and Jaan Tallinn. The lab quickly begins training software to play Atari games from raw pixels and reward signals alone, no rules given, a concrete early demonstration that AI agents could learn to master unfamiliar environments on their own.
DeepMind blog · Google Research · TechCrunch · The Guardian
2012
AlexNet crushes ImageNet
Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton publish their landmark CNN paper. Trained on two GPUs in Krizhevsky's bedroom at his parents' house. Hinton later: "Ilya thought we should do it, Alex made it work, and I got the Nobel Prize." Deep learning's modern era begins.
NeurIPS paper · Computer History Museum · IEEE Spectrum
landmark paper
2012–13
The SpaceX cafeteria warning
Peter Thiel's Founders Fund connects Hassabis to Musk via Luke Nosek. Hassabis tours SpaceX and hears Musk describe Mars as civilization's backup plan in the event of war, asteroids, or collapse. Hassabis replies that a sufficiently advanced AI could follow humans to Mars and destroy the colony too, that there is no escape hatch. Musk goes silent for a full minute processing the idea. He then invests $5M in DeepMind to keep tabs on the technology.
Vanity Fair · TIME / Isaacson
pivotal momentMars thread
Late 2013
Musk tries to block Google's DeepMind acquisition
Musk and Nosek scramble to counter-bid. At a party in LA, they do an hour-long Skype call with Hassabis from an upstairs closet. "The future of AI should not be controlled by Larry," Musk tells him. It fails. The leading AI agent lab, the one already training software to learn from scratch, is now inside Google.
TIME / Isaacson · The Guardian
Jan 2014
Google acquires DeepMind (approximately $500M)
Google wins the bid. Larry Page agrees to a "safety council" with Musk as a member. It holds one meeting at SpaceX, with Page, Hassabis, Eric Schmidt, and Reid Hoffman. Musk concludes it's "basically bullshit." He starts hosting his own AI safety dinner discussions and reaches out to President Obama for a one-on-one meeting in May 2015.
TechCrunch · The Guardian · TIME / Isaacson
pivotal moment
OpenAI: The counterweight to Google
~2015
Larry Page calls Musk a "speciesist"
During a debate about AI risk, Google co-founder Larry Page accuses Musk of being a "speciesist" for prioritizing humans over digital minds. Musk later testifies this exchange helped motivate OpenAI as a direct counterweight to Google.
Reuters trial takeaways · TIME / Isaacson
Dec 11, 2015
OpenAI founded as a nonprofit
$1B pledged (Musk ultimately contributed approximately $38M). Musk and Altman serve as co-chairs. Musk personally recruits Ilya Sutskever from Google, calling him "the linchpin for OpenAI being successful."
The 11 cofounders (and where they ended up)
Sam Altman — CEO. Leading OpenAI through the Musk trial and restructuring push.
Elon Musk — Left the board in 2018. Founded xAI in 2023. Now suing OpenAI while SpaceX partners with Anthropic for compute.
Ilya Sutskever — Chief scientist. AlexNet co-author with Krizhevsky and Hinton. Led the failed Altman ouster Nov 2023. Left mid-2024 to found Safe Superintelligence Inc.
Greg Brockman — President. Still at OpenAI. Testified during the 2026 Musk trial. Took extended leave in 2024 and later returned.
Andrej Karpathy — Created Stanford CS231n. Joined Tesla in 2017 (see below), returned to OpenAI in 2023, left again in 2024. Coined "vibe coding" in 2025. Now runs Eureka Labs. nanoGPT and minGPT repos are canonical.
John Schulman — Key RLHF and ChatGPT researcher. Left OpenAI Aug 2024 for Anthropic, then left Anthropic Feb 2025 for Thinking Machines Lab.
Wojciech Zaremba — One of three original cofounders still at OpenAI (with Altman and Brockman). Turned down 2–3x market rate offers to join.
Durk Kingma — Left in 2018, went to Google Brain, then joined Anthropic in Oct 2024.
Vicki Cheung — Left 2017 for Lyft.
Trevor Blackwell — YC / robotics background.
Pamela Vagata — Engineering background.
OpenAI founding post · OpenAI on Musk · Reuters, Schulman · Reuters, Schulman exits Anthropic · TechCrunch, Kingma · Karpathy
pivotal momentdiaspora begins
May 2016
Dario Amodei visits, then joins OpenAI
Then a Google researcher, Amodei visits Altman and Brockman and tells them "no one understood what they were doing." He joins anyway. His sister Daniela follows about two years later. He later becomes VP of Research and the person most responsible for GPT-2 and GPT-3's development.
The New Yorker · Anthropic
Jun 2017
"Attention Is All You Need" published
Vaswani et al. at Google Brain introduce the Transformer architecture. Eight authors, six of whom would later leave Google to found AI companies. The paper that makes everything that follows possible.
arXiv · NeurIPS
landmark paper
2017
The control fight, the poach, and the exit
Three events that belong together. First, Musk pushes for majority equity in OpenAI and the right to choose most board members. Brockman later testifies that Musk said he needed $80 billion to build a self-sustaining Mars city, and that controlling OpenAI's AI could help him raise it. The Mars escape-hatch fear from the SpaceX cafeteria has become a control problem. When the equity discussion collapses, Musk "storms around the table so aggressively that Brockman thought Musk was going to physically hit him," tears a painting off the wall, and threatens to withhold funding.
Reuters, Brockman testimony · Wired · OpenAI on Musk
Mars thread
2017
Musk poaches Karpathy while still on the board
Part of the same rupture. While still an OpenAI board member, Musk secretly recruits cofounder Andrej Karpathy to lead Tesla Autopilot Vision. This is the person who created Stanford's CS231n and helped build OpenAI's early research culture. Musk later comes to Brockman with "an apology and a confession." An email to a Tesla VP reads: "The OpenAI guys are gonna want to kill me." Karpathy goes on to build the neural network architecture powering millions of Teslas.
TechCrunch, 2017 · TechCrunch, trial
irony alertdiaspora
Feb 2018
Musk leaves OpenAI's board
Cites "potential future conflict of interest" with Tesla's AI work. He also pitches having Tesla acquire OpenAI; that fails too. OpenAI later restructures into a "capped-profit" entity, the very structure Musk will sue them for creating.
The Verge · OpenAI on Musk
The fracture: Safety exoduses
2019
GPT-2 released (initially withheld)
Dario Amodei helps lead development. OpenAI stages a limited release citing misuse risks, the first major "too dangerous to release" moment. The tension between scaling and safety that will fracture the lab is already visible.
OpenAI GPT-2 post · OpenAI follow-up
2020
GPT-3 arrives
Amodei's team scales dramatically. He and a group become convinced of two things: scaling will keep working, and safety needs to be architectural from the start.
arXiv · NeurIPS
2021
Anthropic founded
Dario Amodei (VP of Research) and Daniela Amodei leave with five other OpenAI colleagues, seven cofounders total. The group had worked on GPT-2, GPT-3, RLHF, scaling laws, and interpretability. Dario later: "I had a particular vision of how AI should be brought out in the world... It is incredibly unproductive to try and argue with someone else's vision." He clarifies he wasn't opposed to commercialization (he helped commercialize GPT-3) but wanted safety baked in at the architecture level. They raise $124M.
Anthropic · Lex Fridman transcript · Inc.
pivotal momentdiaspora
Nov 2022
ChatGPT launches
Everything changes. OpenAI goes from research lab to household name. John Schulman, cofounder and the researcher most credited with ChatGPT's RLHF training, will eventually leave too. Musk says he has "lost trust" in Altman by this point.
OpenAI · UC Berkeley
Jan 2023
Microsoft makes its reported $10B OpenAI investment
Microsoft announces a "multiyear, multibillion-dollar" deal; reporting puts the figure at $10 billion. For Musk, this is the "tipping point." He later testifies it convinced him Altman and Brockman were "stealing the charity."
Microsoft · OpenAI · Reuters trial
Feb 2023
Karpathy returns to OpenAI
After leaving Tesla in mid-2022, Karpathy rejoins OpenAI to work on GPT-4, synthetic data, and midtraining research. His second stint lasts about a year.
Karpathy · The Verge
diaspora
The chaos: Moratoriums, lawsuits, and Grok
Mar 2023
Musk signs AI moratorium letter
Co-signs open letter calling for a six-month pause on AI systems more powerful than GPT-4. Signed by 1,100+ people. Cites existential risk. He later testifies he signed because "it seemed like a good idea."
Future of Life Institute · Reuters
Mar 2023
Musk incorporates xAI
The same month he calls for an AI moratorium. Officially launches July 2023 with engineers recruited from Google, DeepMind, and OpenAI. Will build the Colossus data center in Memphis at record speed.
Reuters
irony alert
Nov 2023
The board crisis / Grok launches
OpenAI's board (including Sutskever) fires Altman. 738 of 770 employees threaten to quit. Altman returns days later. The board reportedly approaches Dario Amodei about replacing Altman and potentially merging the two companies; he declines both. Meanwhile, xAI ships Grok on X. Sutskever signs the letter demanding Altman's return, then leaves for good in mid-2024 to found Safe Superintelligence Inc.
Wired · Reuters · xAI · AP
diaspora
Feb 2024
Karpathy leaves OpenAI (again)
"Nothing 'happened'... it's not a result of any particular event, issue, or drama." Goes on to build Eureka Labs, an AI-native education company. His YouTube "Zero to Hero" series on LLMs becomes canonical. Coins "vibe coding" in Feb 2025, a term that immediately enters the professional lexicon.
Karpathy on X · TechCrunch · Karpathy, vibe coding
diaspora
2024
Musk's OpenAI lawsuit escalates
Musk first sues in March 2024, alleging OpenAI abandoned its founding mission. Withdraws the suit in June. Refiles in August with expanded claims and later amends to add Microsoft as a defendant. His attorneys seek up to $150 billion in damages and removal of Altman and Brockman. The same month he refiles, cofounder John Schulman leaves for Anthropic. Brockman takes extended leave through November, leaving Altman and Zaremba as the two active original cofounders at that point.
Reuters, June · TechCrunch · Reuters, Schulman
lawsuitdiaspora
Jun 2024
Colossus 1 powers up in Memphis
xAI builds one of the world's largest data centers in a former Electrolux factory, deployed at record speed. Runs initially on 35 unpermitted gas turbines. NAACP will later sue over Clean Air Act violations in the majority-Black neighborhoods surrounding the site.
Data Center Dynamics · AP · NAACP
2026: The trial and the twist
Feb 2026
SpaceX acquires xAI
xAI becomes a SpaceX subsidiary ahead of a planned record IPO, with reporting tying the merger to ambitions for space-based data centers. Musk calls Anthropic "misanthropic," "evil," and claims it "hates Western civilization."
Reuters merger · Reuters space compute
Mars thread
Apr 28, 2026
Musk v. Altman trial begins in Oakland
Musk testifies for three days. Under oath, admits xAI "partly" trained Grok on OpenAI model output via distillation. He ranks current AI providers: Anthropic first, OpenAI second, Google third, xAI trailing. Brockman rebuts Musk on the stand: Musk was never interested in open-sourcing, the Karpathy poach wasn't voluntary, and Musk was the one who pushed to go for-profit and sought control to fund Mars.
Reuters kickoff · Reuters, Brockman · TechCrunch, distillation
irony alertongoing trial
May 6, 2026
SpaceX announces Anthropic compute deal
The company Musk called "evil" gets full access to Colossus 1: over 220,000 Nvidia GPUs, more than 300 MW of capacity. The data center xAI built to compete with Anthropic is now leased to Anthropic. As part of the deal, Anthropic expresses interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity, bringing the story full circle: the fear that began in a SpaceX cafeteria with Hassabis warning that AI could follow humans into space now ends with Musk's rocket company and the AI safety lab he despises exploring how to put data centers in orbit together.
Anthropic · xAI · Reuters space compute
irony alertMars thread
◈
The cofounder diaspora, as of today
Of the 11 original OpenAI cofounders, 3 remain in active roles: Altman, Brockman, and Zaremba. One is suing the company (Musk). One tried to fire the CEO, then left to found a new company (Sutskever → Safe Superintelligence). One was poached by Musk, returned, left again, and now teaches AI on YouTube (Karpathy → Eureka Labs). One left for Anthropic, then left Anthropic for Thinking Machines Lab (Schulman). Another left for Google Brain, then joined Anthropic (Kingma). The rest departed between 2017 and 2024. Meanwhile, Dario Amodei, VP of Research and outside the cofounder group, took six colleagues and built what Musk himself ranks under oath as the top AI provider in the world.
OpenAI founding post · Reuters, Schulman · Reuters, Schulman exits Anthropic · TechCrunch, Kingma