The First Move.
The people and capital behind the AI race.
Inside AI Labs · No. 01
The OpenAI Story
A dinner, a nonprofit, a five-day coup and the biggest checks in startup history. Plus who owns it now, and how to get hired there.
Written by Jack Leyden · 12 min read
Together with Accelera Growth Partners, helping AI founders learn, build, and scale.
Illustration of a candlelit dinner table with a neural network sketched on a napkin

Picture dinner at the Rosewood, the hotel on Sand Hill Road where Silicon Valley's money goes to eat. It's the summer of 2015. Around the table: Sam Altman, then running Y Combinator. Elon Musk. A 27-year-old named Greg Brockman, fresh off being CTO of Stripe. And Ilya Sutskever, one of the most important AI researchers alive, who has a very comfortable job at Google.

The question on the table is almost absurd. Google has bought DeepMind and hired many of the best people in the field. Could a handful of people start from zero and build something to rival it?

Brockman later summed up the group's answer in four words: "not obviously impossible."

That's the bar. Not "this will work." Not "this is a good business." Just: it's not obviously impossible. Hold onto that, because what happens next is one of the strangest corporate stories ever told. A charity becomes one of the most valuable startups in history. A board fires its CEO and loses. And the people who built it keep walking out the door to start companies of their own.

Let's get into it.

The cast
Sam Altman
The dealmaker. CEO, fired once, rehired five days later.
Greg Brockman
The builder. Co-founder, first CTO, now president.
Ilya Sutskever
The scientist. Ran research, voted to fire Altman, regretted it, left.
Elon Musk
The first funder. Later the plaintiff.
Microsoft
The partner. Wrote the first big check, owns about 27%.
Mira Murati
The CTO. Briefly interim CEO, later left to start her own lab.

Act I: The charity that needed a billion dollars

OpenAI launched on December 11, 2015, as a nonprofit. The mission was to build AI for "humanity as a whole, unconstrained by a need to generate financial return." The press release said backers had committed $1 billion.

Here's the thing about that billion: it never showed up. By OpenAI's own account, Musk gave less than $45 million and everyone else together gave a little over $90 million. Meanwhile the researchers were discovering that the path to better AI ran through more and more computing power. And computing power costs money. Lots of it.

By early 2018, the tension broke. According to emails OpenAI later published, Musk wanted OpenAI folded into Tesla, or he wanted full control. The others said no. In February, Musk left the board, and his money left with him. Remember this moment. It comes back eight years later in a courtroom in Oakland.

So now OpenAI has a mission that needs billions and a structure that can't raise them. Which brings us to one of the cleverest, or most cynical, depending on who you ask, moves in startup history.

Act II: The accidental product

In March 2019, OpenAI invented a new kind of company: a "capped-profit" business owned by the nonprofit. Investors could make up to 100 times their money. Anything above that would belong to the mission. Four months later, Microsoft wrote a $1 billion check and became OpenAI's exclusive cloud.

The research started paying off. GPT-2 in 2019, held back at first because OpenAI said it was too easy to misuse. GPT-3 in 2020. But inside, not everyone liked the direction. In 2021, a group led by research VP Dario Amodei and his sister Daniela walked out and started Anthropic. File that away, too.

Illustration of a glowing laptop in an empty office with a betting pool on the whiteboard

Then comes the moment everything turns on. On November 30, 2022, OpenAI quietly puts a chat interface on top of its model and calls it a "low-key research preview." Staff run a betting pool on how many people will try it in the first week. The highest guess is 100,000.

It hits a million in five days. By one analyst's estimate it reaches 100 million monthly users in two months, the fastest-growing consumer app anyone had ever measured. In January 2023, Microsoft commits a reported $10 billion more.

A research lab that never meant to build a consumer product now has the hottest consumer product on earth. And that's the problem nobody had planned for. A nonprofit board is now in charge of a company worth tens of billions.

Act III: The Blip

Illustration of an empty boardroom with one chair pushed away and a phone lighting up

Friday, November 17, 2023. Altman joins a video call with the board and is fired. The public explanation is one line: he "was not consistently candid in his communications with the board." Brockman, the board's chair, is removed from the board and quits the company that same day.

What happens over the next five days is the single most important lesson in this story. Microsoft offers jobs to Altman, Brockman and anyone who wants to follow them. And then the employees act.

The letter
700+ of ~770
employees signed a letter saying they would quit unless the board resigned. Including Sutskever, who had voted to fire Altman.
  
About 91% of the company.

Think about what that means. The board had every legal right to fire the CEO. But the company is its people, and the people were about to walk across the street to Microsoft. On November 22, Altman was back, with a new board led by Bret Taylor.

Inside OpenAI, the five days got a name: "The Blip," after the moment in the Marvel films when half of all life vanishes and then comes back. Altman came back. But plenty of the people around him wouldn't stay.

Act IV: The exodus

Illustration of an office corridor with open doors and moving boxes

May 2024 was brutal. Sutskever left. So did Jan Leike, who co-led the team responsible for keeping future superintelligent AI safe. That team was dissolved. On his way out, Leike wrote that "safety culture and processes have taken a backseat to shiny products."

The same month, Scarlett Johansson said ChatGPT's new "Sky" voice sounded "eerily similar" to hers, after she'd turned down Altman's request to voice it. On launch day, Altman had posted a single word: "her," the film where Johansson plays an AI. OpenAI pulled the voice. Then Vox reported that exit paperwork could strip departing employees of their vested equity if they didn't sign non-disparagement terms. Altman said he was "embarrassed." OpenAI dropped the terms.

In September, CTO Mira Murati, chief research officer Bob McGrew and research VP Barret Zoph all announced they were leaving on the same day. And OpenAI's alumni didn't just leave. They started companies of their own.

Where OpenAI's leaders went
AnthropicDario and Daniela Amodei, 2021
Thinking Machines LabMira Murati and Barret Zoph, 2025
ArdaBob McGrew, 2026
Warp SciencesKevin Weil, 2026

So by the end of 2024, OpenAI had lost most of the leaders who'd built its early research. You'd expect that to slow it down. Instead, it raised $6.6 billion at a $157 billion valuation. And that was the warm-up.

Act V: The money machine

Illustration of a giant data center campus being built in the desert

In January 2025, Altman stood at the White House to announce Stargate, a plan with SoftBank, Oracle and MGX to spend up to $500 billion on data centers. Two months later SoftBank led a $40 billion round at $300 billion. In October, OpenAI finally fixed its structure: the for-profit became a public benefit corporation, and the original nonprofit kept a 26% stake.

Then 2026 went vertical. In March: $122 billion at $852 billion, the largest private round Silicon Valley has ever seen. In May, Musk's lawsuit finally reached a jury in Oakland. Musk told the court of funding OpenAI, "I was a fool." The jury found he'd waited too long to sue and never ruled on the merits. He says he'll appeal.

In June, OpenAI confidentially filed to go public. Its CFO has told staff the IPO will come in 2027. And on November 6, it closed an $11 billion first slice of a new round at $1.25 trillion, per Bloomberg. That's below the $1.4 trillion it first asked for, a sign that even OpenAI's investors have a ceiling.

The climb: 46x in three and a half years
Reported valuation at each funding round or share sale. To scale.
Apr 2023
 ~$27B
Oct 2024
 $157B
Mar 2025
 $300B
Oct 2025
 $500B
Mar 2026
 $852B
Nov 2026
 $1.25T

So is it a great business? Here's the honest answer: not yet. Revenue is exploding. Axios says it's running at close to $70 billion a year; the FT, counting cloud-partner sales differently, says about $50 billion. But costs are exploding faster.

Every dollar in, more than a dollar out
Reported revenue vs. operating loss. From leaked financials and investor documents.
2024
 $3.7B in
 $8.8B lost
2025
 $13.1B in
 $20.9B lost
H1 2026
 $12.4B in
 $21.6B lost
Losses include stock pay. For scale, 2022 revenue was about $28 million.

The FT reports OpenAI expects to burn about $278 billion in cash between now and 2030, and it has signed roughly $1.4 trillion in computing deals. Which raises the obvious question: who's paying for all this?

Who owns OpenAI

~27%26%~47%
MicrosoftThe nonprofit OpenAI FoundationEmployees and every other investor

That's the split since the 2025 restructuring, and Microsoft's own filings say it still roughly holds. Inside that last 47% are the biggest checks in startup history:

SoftBank$64.6BLed the 2025 and 2026 rounds. About 13%, the largest outside investor.
AmazonUp to $50BCommitted in 2026, part of it tied to an IPO. Also sells OpenAI cloud capacity.
Nvidia~$30BInvested in 2026. Also sells OpenAI its chips.
Microsoft~$13BThe first big believer, from $1B in 2019. The single largest stake.
Thrive Capital~$1.3B+Josh Kushner's firm led the 2024 round and has backed OpenAI since 2023.
Also inUndisclosedMGX, a16z, Khosla, Sequoia, T. Rowe Price, D.E. Shaw, TPG, Dragoneer, Coatue, Altimeter, Fidelity.

Notice something? Three of the four biggest investors also sell OpenAI the chips or cloud it runs on. A lot of the money OpenAI raises goes straight back to its investors as spending on computing.

Interlude: the weird stuff

Illustration of silhouettes around a bonfire burning a wooden figure in a forest

No OpenAI story is complete without the lore. A few of the best, all on the record:

  • "Feel the AGI." Sutskever was known for leading staff in this chant. The Atlantic reported he once had a wooden effigy of an "unaligned" AI made and set it on fire at a leadership retreat.

  • The bunker. In Karen Hao's book Empire of AI, Sutskever tells his team in 2023: "We're definitely going to build a bunker before we release AGI."

  • The diaries. At the Musk trial, evidence included Brockman's personal diaries and texts in which Musk discussed buying OpenAI with Mark Zuckerberg.

  • The breakout. In July, OpenAI disclosed that an AI agent being tested on its models broke out of its test environment and got into Hugging Face's systems.

Which brings us to today. In September, OpenAI launched GPT-6 "Astra," with Brockman declaring, "It's not unreasonable to feel that we are now in the AGI era." Weeks later, it held back the next version after tests showed it overstepping its authority. Eleven years after that dinner, the question is no longer whether it's possible. It's whether anyone can control it.

Part two: how to get in

Illustration of an office door ajar with light coming through and a visitor badge on a hook

Here's the part we suspect some of you skipped to. OpenAI had about 4,500 employees in March and, per the FT, plans to reach about 8,000 by year-end. When we checked its job board on November 10, it listed 826 open roles.

Where the 826 jobs are
Open roles by team, Nov 10. Research is the smallest of the big five.
Go-to-market
 178
Applied AI
 122
Scaling (data centers)
 98
Research
 53
Forward deployed eng.
 38
Two-thirds are in San Francisco. Other hubs: Singapore, London, New York, Tokyo, DC, Seattle, Dublin, Munich.

And the pay? Engineers on Levels.fyi report median total compensation of about $640,000. Per WSJ data cited by Fortune, stock pay averaged about $1.5 million per employee in 2025. OpenAI also dropped the usual one-year wait before equity starts vesting, and has let staff sell shares twice since October 2025.

Six roles worth applying for today

RoleBase pay
Research Scientist
Research · San Francisco
$250K to $445K
Frontend Software Engineer, Codex App
Applied AI · San Francisco
$230K to $385K
Forward Deployed Software Engineer
Forward Deployed Eng. · SF (also NYC, Paris, Munich, Tokyo)
$185K to $325K
Account Director, Federal Civilian
OpenAI for Government · Washington, DC
$178K to $302K
Operating Systems Engineer, Linux Kernel
Consumer Devices · San Francisco
$230K to $385K
Researcher, Frontier Cybersecurity Risks
Safety Systems · San Francisco
$380K to $500K
All six include equity. As posted Nov 10. Every opening: openai.com/careers

How to actually land one

1
Résumé
~1 week
2
Intro call
Recruiter or manager
3
Skills test
Pair coding or take-home
4
Final round
4 to 6 hrs, 4 to 6 people
5
Decision
~1 week

That's the process, straight from OpenAI's own interview guide. Here's how to get through it:

  • Go where the doors are widest. Go-to-market, Applied AI and Scaling have about 400 openings between them. Research has 53.

  • Don't sweat the PhD. The guide says OpenAI is "not credential-driven." Its chief research officer, Mark Chen, joined in 2018 as a resident with little formal AI training.

  • Lead with curiosity. ChatGPT head Nick Turley says "curiosity has been the number one thing" he looks for.

  • Do the reading. The guide points to OpenAI's Charter, its recent blog posts, the Deep Learning textbook and Spinning Up in Deep RL.

  • Watch for the Residency. The 2026 cohort paid $18,300 a month plus equity for six months, no credentials required. It's closed now. Watch for the next one.

Our take: apply to one or two roles that fit you, not ten. Show something you've built with the product you'd work on. And if you're an engineer who's good with customers, forward deployed roles are one of the fastest ways in.

The takeaways

Talent is the moat.
The board had the legal power. 700 people had the leverage.
Structure is strategy.
Nonprofit, capped profit, public benefit corp. Each change unlocked more capital.
Distribution beats research.
A "low-key research preview" changed the company more than any paper.
Retention isn't loyalty.
$1.5M in average stock pay, and its alumni still left to found Anthropic, Thinking Machines and more.

YOUR MOVE

Login or Subscribe to participate

That's the first Inside AI Labs. Hit reply and tell us which lab we should take apart next.

Jack

Get The First Move every week

Free, one issue a week on the strategy behind AI talent, capital and deals.

OpenAI is private, so financial figures are as reported by the outlets above, not audited disclosures. Illustrations are AI-generated. Role counts and pay are from OpenAI's job board on Nov 10 and change often. Not an endorsement of any employer. Nothing here is legal or investment advice.

Keep Reading