"We want AI to be widely and evenly distributed. We think the best way to ensure that is to be open."
-- Sam Altman, OpenAI founding announcement, December 2015
The first lie was the billion dollars.
On the evening of December 11, 2015, a group of men sat down to dinner in a private room at the Rosewood Hotel on Sand Hill Road -- the most expensive half mile of real estate in Silicon Valley, where venture capitalists kept offices that smelled of leather and inevitability. Outside, the California winter was mild and cloudless, the kind of night that made people forget the rest of the country existed. Inside, the men at the table were about to announce something designed to make the rest of the world pay attention.
Sam Altman, at thirty years old the president of Y Combinator, the most powerful startup accelerator on earth, sat at the center of the table with the relaxed posture of a man who had been running rooms since before he could legally drink. He'd dropped out of Stanford at nineteen to found a mobile app called Loopt that had failed slowly enough to teach him everything about how Silicon Valley actually worked. By twenty-nine, he was running YC, handpicked by Paul Graham, and had invested in companies -- Airbnb, Stripe, Reddit -- that were remaking the economy. He was slight, soft-spoken, and had the particular quality shared by the most dangerous people in technology: when he talked, you found yourself agreeing before you'd fully processed what he'd said.
Across from him sat Elon Musk, who needed no introduction and preferred it that way. Musk was forty-four, already running both Tesla and SpaceX, and had spent the past two years growing increasingly alarmed about a problem that most of the public hadn't yet learned to worry about. Artificial intelligence. Not the kind in movies -- not some chrome-skinned robot kicking down a door. The real thing. The software kind. The kind that Google was building in a nondescript office complex in London, where a company called DeepMind had been quietly assembling the most talented AI researchers on the planet.
Earlier that year, DeepMind had demonstrated internally that its system could defeat any human at Go, the ancient Chinese board game that was supposed to be decades beyond the reach of computers. Musk had heard about the results before they were made public. The achievement itself didn't alarm him -- it was a board game. What alarmed him was the trajectory. If you plotted the curve of what these systems could do and projected it forward five years, ten years, twenty -- the implications were either the best thing that had ever happened to humanity or the worst. And Google, a single corporation answerable to its shareholders and addicted to its advertising revenue, was going to be the one to decide which.
Musk had been talking about this at dinner parties for months. He'd cornered Larry Page, Google's co-founder, at a party and argued until he was hoarse that concentrating this much power in one company was dangerous. Page had waved him off, called him a "speciesist" for caring more about humans than about whatever digital minds might come next. The exchange had shaken Musk. If the people building the most powerful technology in history couldn't even agree that human survival was a priority, someone else needed to step in.
That someone, Musk had decided, was him. And Altman was going to help.
The plan they'd hatched was elegant in the way that Silicon Valley plans always were -- simple on the surface, labyrinthine underneath. They would create a nonprofit research lab dedicated to building artificial general intelligence -- the kind of AI that could match or exceed human cognition -- and then give it away. Open-source it. Make it available to everyone, so no single company could monopolize the technology. They would call it OpenAI, because the name said everything: this would be AI for the people, not for Google's shareholders.
Around the table sat the men and women who would make it real. Greg Brockman, twenty-seven, a prodigiously gifted engineer who had been CTO of Stripe before burning out and looking for something that mattered more than payment processing. He would become OpenAI's president and chief builder, the person who translated the vision into code. Ilya Sutskever, twenty-nine, a Russian-born Israeli-Canadian who had studied under Geoffrey Hinton -- the man they called the Godfather of AI -- at the University of Toronto. Sutskever was quiet in the way that genuinely brilliant people sometimes are, not from shyness but from the understanding that most conversations were happening several levels below where his mind operated. He had been at Google Brain, Google's internal AI research team, and Musk had personally recruited him away with a pitch about saving the world. Sutskever would become OpenAI's chief scientist -- its intellectual engine.
And then there were the funders. Peter Thiel, the PayPal co-founder turned libertarian venture capitalist. Reid Hoffman, the LinkedIn founder who collected philanthropic causes the way other billionaires collected vintage cars. Jessica Livingston, the co-founder of Y Combinator and Altman's mentor. Amazon Web Services, the cloud computing juggernaut, had signed on to provide the infrastructure. The pledges rolled in -- a million here, several million there, the kind of money that felt enormous to normal people and was essentially a rounding error to the men writing the checks.
But the headline number -- the one that would dominate the press coverage the following morning -- was a billion dollars. One billion dollars, pledged to build AI that belonged to everyone. It was an astonishing figure, breathtaking in its ambition, world-historical in its implication. And it was, from the very beginning, more aspiration than commitment.
The truth about the billion dollars emerged slowly, the way truths in Silicon Valley always do -- through tax filings, court documents, and the diligent work of reporters who knew where to look.
Brockman and Altman had initially planned to raise one hundred million. A hundred million was real money, enough to hire a world-class research team and buy the computing power they'd need to train the models that were still just theoretical sketches on whiteboards. But sometime in the weeks before the announcement, the figure had been revised upward -- not because the pledges had grown, but because the ambition of the announcement demanded a number that would make the front page.
A billion dollars was that number.
By the time OpenAI filed its first tax returns as a 501(c)(3) nonprofit, the reality was already diverging from the press release. Of the billion pledged, only about one hundred and thirty million dollars would actually be collected by 2019. Elon Musk, the largest donor by far, contributed somewhere between thirty-eight and forty-five million -- roughly sixty percent of the early funding. Sam Altman's contribution came in the form of a loan: $3,784,637, lent to OpenAI to get operations started, later forgiven as a charitable gift. Reid Hoffman put in about a million. Peter Thiel contributed several million. YC Research -- the arm of Y Combinator that Altman ran -- contributed nothing at all.
The gap between the billion-dollar announcement and the hundred-and-thirty-million-dollar reality would become, years later, Exhibit A in a federal courtroom. But on the night of December 11, 2015, that courtroom was still a decade away, and the gap felt like a technicality. In Silicon Valley, aspiration and reality were always running on different clocks. The important thing was the story you were telling, and the story Musk and Altman were telling was irresistible: two of the most powerful men in technology, putting their money and their reputations behind the idea that the most transformative technology in human history should belong to everyone.
The following morning, OpenAI's founding was announced in a blog post and covered by every major outlet on earth. The message was clear and the framing was noble. "OpenAI is a non-profit artificial intelligence research company," the announcement read. "Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return."
In Redmond, Washington, Satya Nadella read the coverage with interest. The Microsoft CEO had been in the job for less than two years, having taken over from Steve Ballmer in 2014, and was still in the process of remaking the company from a stodgy Windows-and-Office machine into something more agile and forward-looking. He noted that OpenAI's announcement mentioned Amazon Web Services as a technology partner. He turned to his team and asked a question that would prove prescient.
"Did we get called to participate?"
AWS, he noted, had "sneaked in there." Microsoft had not.
Nadella filed the name OpenAI away. He would come back to it.
To understand what happened at that dinner -- and everything that followed -- you have to understand who these people were. Not the press-release versions. The real ones.
Sam Altman was born in Chicago in 1985 and raised in St. Louis, Missouri, the eldest of four brothers. He came out as gay at sixteen, attended Stanford for two years, and dropped out to start Loopt, a location-sharing app that raised thirty million dollars, attracted five million users, and was eventually sold to Green Dot Corporation for forty-three million -- a mediocre return that nonetheless taught Altman how the startup machinery worked from the inside. He was better at understanding systems than at building products, a distinction that matters enormously in Silicon Valley. When Paul Graham tapped him to lead Y Combinator in 2014, it was a recognition that Altman's real talent was not building companies but choosing them -- and, more importantly, choosing the people who ran them.
By thirty, he'd built a network that stretched across every significant technology company in the Bay Area. He invested early. He invested often. His personal portfolio would eventually include stakes in Airbnb, Stripe, Reddit, Instacart, and dozens of others. His official salary at OpenAI, when he eventually became CEO, would be $76,001 -- "enough for health insurance," as he'd later tell Congress. His net worth, built entirely from outside investments, would exceed two billion dollars. He would buy a twenty-seven-million-dollar mansion in San Francisco. The monk's vow of nonprofit poverty was, like the billion-dollar pledge, a story that was true in the narrowest possible sense and misleading in every other.
The people who worked with Altman described his intelligence with genuine awe. He could absorb complex technical arguments faster than the engineers making them. He could walk into a room of skeptics and leave with allies. He was, by all accounts, an extraordinary strategic thinker. But the same people who praised his brilliance often added a qualifier, sometimes spoken carefully, sometimes not. They said he was not always candid. They said he had a gift for telling different people different things. They said that when you tried to pin him down on a commitment, the words had a way of shifting beneath your feet.
"For any individual case," one former colleague would later say, "Sam could always come up with some kind of innocuous-sounding explanation."
Elon Musk was a different species of problem. Born in 1971 in Pretoria, South Africa, he'd emigrated to Canada at seventeen and eventually made his way to Silicon Valley, where he co-founded PayPal, survived a brutal boardroom coup, and used his fortune to start two companies that everyone said were impossible: SpaceX, which would land rockets on barges in the middle of the ocean, and Tesla, which would prove that electric cars didn't have to be ugly and slow. Musk was a man who ran on fear and ambition in roughly equal measure. He feared AI the way a general fears a weapon he can't control -- not abstractly, but viscerally, as a tactical threat to everything he was building. His public warnings about artificial intelligence had grown increasingly urgent through 2014 and 2015. "With artificial intelligence, we are summoning the demon," he told an audience at MIT. "You know all those stories where there's the guy with the pentagram and the holy water, and he's like, 'Yeah, I'm sure I can control the demon.' It doesn't work out."
But fear was only half of it. Musk wanted control. He always wanted control. At Tesla, he was CEO and chairman and product architect. At SpaceX, he was CEO and chief engineer. When he invested in something, he didn't write a check and walk away -- he walked in, sat down, and started making decisions. OpenAI would be no different. Before the founding announcement, Musk had made his expectations clear. He wanted majority equity in any for-profit entity that might eventually be created. He wanted initial board control. And he wanted to be CEO.
Altman, Brockman, and Sutskever said no.
"Concerns about one person having absolute control," was how the refusal was characterized in documents that would later surface in court filings. It was a polite way of saying that the three of them looked at Musk -- a man who had fired executives mid-sentence, who had publicly berated employees on Twitter, who had once told a Tesla engineer that he'd rather the company go bankrupt than surrender decision-making authority -- and decided that putting him in charge of the most powerful technology in history was exactly the kind of thing OpenAI was supposed to prevent.
Musk agreed to the nonprofit structure. He agreed to be co-chair of the board alongside Altman. He agreed, outwardly, that the mission came first and that no individual -- including himself -- should have the power to steer it alone.
He agreed. But he didn't forget.
Ilya Sutskever arrived at OpenAI with the quiet conviction of a convert.
Born in 1986 in what was then the Soviet Union, he had grown up in Israel before his family emigrated to Canada. He enrolled at the University of Toronto and fell under the tutelage of Geoffrey Hinton, the British-Canadian computer scientist who had spent decades toiling in obscurity on a branch of artificial intelligence called neural networks while the rest of the field dismissed it as a dead end. Hinton believed that if you built artificial neural networks that loosely mimicked the structure of the human brain -- layers of interconnected nodes, each adjusting its behavior based on feedback -- and fed them enough data, they would eventually learn to do things that no one had explicitly programmed them to do.
In 2012, Hinton and two of his graduate students -- Sutskever and Alex Krizhevsky -- proved him right. Their neural network, called AlexNet, won the ImageNet competition by a margin so large it was like watching a Formula One car enter a bicycle race. The field of artificial intelligence pivoted overnight. Companies that had ignored neural networks for decades suddenly couldn't hire enough people who understood them. Google hired Hinton. Google hired Sutskever. Facebook hired Yann LeCun. The deep learning revolution had begun, and the people who had been laughed at were now the most sought-after researchers on the planet.
Sutskever had been at Google Brain -- Google's internal AI research lab -- when Musk came calling. The pitch was not about money, though the money was good. It was about mission. Google was building AI to sell advertising and improve search results. OpenAI would build AI to benefit all of humanity. For Sutskever, who had spent his career in the intellectual tradition of Hinton -- a man who believed science should serve the public good -- the choice was clear.
He became OpenAI's chief scientist. In that role, he was the person who decided what to research, how to train the models, and when a result was significant enough to publish. He was, in a sense, the conscience of the organization -- the person whose judgment determined whether OpenAI was living up to its founding ideals or quietly abandoning them.
Greg Brockman, the president, was the operational counterpart to Sutskever's intellectual leadership. A self-taught programmer from small-town North Dakota who had dropped out of Harvard after one year, Brockman had been CTO of Stripe by twenty-two and had spent his career building things that worked at scale. If Sutskever was the brain, Brockman was the nervous system -- the person who turned theoretical breakthroughs into functioning software and managed the growing team of researchers who were beginning to converge on San Francisco from the best AI labs in the world.
In a diary entry from this period -- a document that would later become a key exhibit in federal court -- Brockman wrote with unmistakable idealism about what they were building. He described his commitment to the nonprofit mission. He wrote about the importance of keeping OpenAI free from commercial pressure. He described the project in the language of moral obligation, not financial opportunity.
"Committed to non-profit," he wrote.
What Brockman did not know -- what none of them knew, sitting around that table in December 2015 -- was how quickly that commitment would be tested. Within two years, the pressures of competing with Google, of attracting talent, of paying for the staggering and ever-growing cost of computing power would begin to erode the idealism of the founding. Within three years, Musk would be gone from the board, furious at being denied the control he'd expected. Within four years, OpenAI would take a billion dollars from Microsoft and create a for-profit subsidiary that would eventually swallow the nonprofit whole. Within five, the co-inventor of the technique that made ChatGPT possible -- a former OpenAI vice president named Dario Amodei -- would leave in alarm, taking a dozen researchers with him to found a rival company built on the premise that OpenAI had lost its way.
And within eight years, Greg Brockman's diary entry about being "committed to non-profit" would be read aloud in a courtroom, where a federal judge would hold it up as evidence that the founding promise was broken almost before the ink was dry. Brockman would, by then, have written other things in that diary. Things like: "I cannot believe that we committed to non-profit if three months later we're doing b-corp then it was a lie." And: "Financially, what will take me to $1B?"
The first crack appeared not at OpenAI but at Microsoft.
In August 2016, eight months after OpenAI's founding, an internal Microsoft slide deck circulated among senior executives. Titled "OpenAI on Azure Big Compute," it laid out the situation with the clinical precision of a company that understood leverage. OpenAI, the deck noted, was running its research on Amazon Web Services as part of a deal giving it fifty million dollars in computing power for ten million in committed funds. Microsoft was being shut out of the most interesting AI research project in the world -- by Amazon.
The solution, the slide deck proposed, was straightforward. Microsoft would offer OpenAI sixty million dollars' worth of Azure computing credits -- a larger package than Amazon's, at a steep discount from commercial rates. In exchange, OpenAI would "evangelize" Azure, helping Microsoft market its cloud platform to the developer and startup audience that had historically gravitated toward AWS. The pitch emphasized the "halo effect" of associating Azure with cutting-edge AI research, the "thought leadership" opportunity, and the chance to recruit a "net-new audience" of developers.
When the offer reached OpenAI, Musk -- still co-chair of the board -- reacted viscerally.
"This actually made me feel nauseous," he wrote in an email that would later surface as a court exhibit. "It sucks and is exactly what I would expect from them. Would be worth way more than $50M not to seem like Microsoft's marketing bitch."
It was a revealing moment. Musk understood, even then, that accepting computing credits from a technology giant came with strings -- that the nonprofit would be trading its independence for infrastructure, its purity for processing power. He was right about the dynamic, even if his language was characteristically blunt.
But he was also pragmatic. In the same email chain, he assessed the alternatives. "I think Jeff is a bit of a tool and Satya is not," he wrote, comparing Amazon's Bezos to Microsoft's Nadella, "so I slightly prefer Microsoft, but I hate their marketing dept."
Altman agreed that Amazon was the worse option. "Amazon started really dicking us around on the T+C," he replied, referring to Amazon's increasingly aggressive terms and conditions, "especially on marketing commits. And their offering wasn't that good technically anyway."
The partnership with Microsoft was announced in November 2016. It was, at the time, a minor deal -- a cloud computing arrangement, not a strategic alliance. Nobody at that dinner table in December 2015 would have predicted that within three years, Microsoft would invest a billion dollars in OpenAI. Or that within seven, the total would reach thirteen billion. Or that the nonprofit founded to prevent any single corporation from controlling artificial intelligence would find itself dependent on the most valuable corporation on earth, paying Microsoft twenty percent of its revenue, committed to purchasing two hundred and fifty billion dollars in Azure cloud services, and giving Microsoft's CEO the power to reshape its board in the middle of the most dramatic corporate crisis in Silicon Valley history.
But the seed was planted in that August 2016 slide deck. And it grew.
The structure OpenAI chose for itself at the founding was both its greatest strength and the instrument of its eventual transformation.
As a 501(c)(3) nonprofit, OpenAI was legally bound to serve the public interest. Its articles of incorporation, filed in January 2016, stated that its purpose was to develop artificial intelligence "in the way that is most likely to benefit humanity as a whole." The board of directors had ultimate authority over the organization. A majority of board members could remove any director at any time, without prior warning. No investor controlled the board. No single individual -- not Musk, not Altman, not Sutskever -- had veto power over the organization's direction.
This structure was designed to prevent exactly what Musk feared: the concentration of AI power in the hands of people motivated by profit rather than principle. The bylaws were explicit. If a conflict arose between the organization's mission and the financial interests of any stakeholder, the mission came first. Everyone who participated in OpenAI -- founders, donors, employees -- did so with the understanding that they were not building a company. They were building a public trust.
The problem was computing power.
Training an AI model in 2015 was expensive. Training a frontier model -- the kind that could compete with what Google was building at DeepMind -- was ruinously expensive. It required thousands of specialized processors called GPUs, manufactured almost exclusively by a company called NVIDIA, running twenty-four hours a day in data centers that consumed enough electricity to power a small city. The cost of a single training run was measured in millions of dollars. And the cost was growing exponentially. Every year, the models got bigger, the data sets got larger, and the computing requirements doubled and then doubled again.
A nonprofit funded by philanthropic donations could not compete in this arms race. Not at scale. Not for long. The hundred and thirty million that actually materialized from the billion-dollar pledge was enough to get started -- to hire a research team, to rent computing time, to publish papers that made the wider AI community sit up and take notice. But it was not enough to keep pace with Google, which could throw the resources of a trillion-dollar corporation at the problem. It was not enough to match Facebook, where Yann LeCun was building the AI research lab that would eventually produce LLaMA. It was not enough to stay ahead of the curve as the curve itself accelerated toward a destination that no one fully understood.
By 2017, the pressure was already building. Altman and Brockman began discussing alternatives to the pure nonprofit model -- ways to attract investment capital without abandoning the mission. They talked about benefit corporations. They talked about hybrid structures. They talked about the thing that every nonprofit eventually talks about when the money runs out and the competition doesn't sleep: how to make the mission sustainable without turning it into a business.
In November 2017, Brockman wrote in his diary: "I cannot believe that we committed to non-profit if three months later we're doing b-corp then it was a lie."
It was a remarkable confession from the president of the organization. The nonprofit was less than two years old, and already the people who built it were questioning whether its structure could survive contact with reality. The idealism of the founding dinner was colliding with the economics of artificial intelligence, and the economics were winning.
What none of them could see, sitting at that table in December 2015, was the shape of the story they were starting.
They thought they were founding a research lab. They were founding a powder keg.
Within three years, Musk would propose to take over as CEO of a for-profit entity spun out of the nonprofit. Altman, Brockman, and Sutskever would refuse him -- again citing concerns about absolute control. Musk would leave the board in February 2018, officially citing a conflict of interest with Tesla's own AI ambitions. Unofficially, according to court filings and the testimony of people who were in the room, he left in cold fury at being denied what he believed was rightfully his.
Within four years, OpenAI would create a "capped-profit" subsidiary -- a legal structure that allowed investors to earn returns capped at one hundred times their initial investment, with any value beyond the cap flowing back to the nonprofit. It was an ingenious solution and a Faustian bargain. The cap sounded prudent. In practice, a ten-million-dollar investment would not be capped until it generated a billion dollars in returns, leading critics to call the limit "limited in name only." Microsoft would invest a billion dollars in this structure in July 2019. Bill Gates, who had built the company that was writing the check, would tell Nadella: "Yeah, you're going to burn this billion dollars." Nadella would invest it anyway.
Within eight years, the nonprofit founded to prevent any single entity from controlling AI would be valued at eighty-six billion dollars, then one hundred and fifty-seven billion, then three hundred billion, then five hundred billion. It would have eight hundred million weekly active users. Ninety-two percent of Fortune 500 companies would use its flagship product, ChatGPT. Its CEO would be fired by his own board on a Friday afternoon, hired by Microsoft on a Sunday evening, and reinstated by Tuesday night in one of the most extraordinary corporate crises in American history. Seven hundred and thirty-seven of its seven hundred and seventy employees would sign a letter threatening to resign unless the board that fired the CEO stepped down -- and the board would step down.
Within ten years, two of the men who sat at that dinner table would face each other in a federal courtroom in Oakland, California. Elon Musk would be the plaintiff. Sam Altman would be the defendant. The damages sought would range from seventy-nine billion to one hundred and thirty-four billion dollars. The central allegation: that the nonprofit founded to benefit all of humanity had been systematically converted into a for-profit corporation, enriching its executives and its largest investor at the expense of the mission that had justified its existence.
The judge would allow the case to proceed to a jury trial, citing "ample evidence."
Greg Brockman's diary -- the one where he had written "committed to non-profit" -- would be entered into evidence. So would the emails about Microsoft's marketing. So would the tax filings showing that the billion-dollar pledge had produced a hundred and thirty million. So would the internal documents showing that the co-founders had rejected Musk's bid for control and then, step by step, concentrated control in their own hands.
Ilya Sutskever, the quiet genius recruited from Google to be the conscience of the organization, would not be in that courtroom. By then, he would have left OpenAI entirely -- after leading a boardroom coup against Altman, then signing a letter demanding the resignation of the very board he'd led, then publicly apologizing, then departing to found a new company called Safe Superintelligence Inc., dedicated to the mission that OpenAI was supposed to serve.
His shares, at the time of his departure, would be worth approximately four billion dollars.
But all of that was still in the future. On the night of December 11, 2015, the men at the table raised their glasses and believed they were doing something good. They believed the billion dollars was real. They believed the mission would hold. They believed that the structure they had built -- the nonprofit board, the open-source commitment, the bylaws that put humanity's interest above any individual's ambition -- would be strong enough to contain whatever came next.
They were wrong about almost everything.
But they were right about one thing: what they were building would change the world. Just not in the way they promised.