From Altruism to Empire: The Full OpenAI Origin Story

TL;DR: OpenAI was incorporated on December 8, 2015, as a nonprofit backed by a $1 billion pledge from some of Silicon Valley's biggest names; only $130 million of that pledge was actually received by 2019.^[1] What followed was a decade of structural contradictions: a capped-profit pivot in 2019, a $13 billion Microsoft partnership, a five-day boardroom coup in 2023 that nearly destroyed the company, the departure of co-founder Ilya Sutskever, and a completed for-profit restructuring in October 2025 that left OpenAI Group PBC valued at $300 billion and the nonprofit foundation controlling equity worth $130 billion.^[2] The mission was always to benefit humanity. The execution looked a lot like building a tech empire.

---

The dinner was at Sam Altman's house in San Francisco. The guests included Elon Musk, Greg Brockman, Ilya Sutskever, and a handful of researchers who were, at that moment, some of the most sought-after people in artificial intelligence. The pitch Altman and Musk made was simple, almost absurd in hindsight: pool resources, publish everything, and race to build artificial general intelligence in the open so that no single corporation could control it. They wanted to save the world by building the thing most likely to end it, and they wanted to do it as a nonprofit.

That dinner, and the conversations that surrounded it through 2015, produced what is arguably the most consequential organization in the history of technology. Not because OpenAI built AGI. But because it built ChatGPT, and ChatGPT changed what the public believed computers could do. Everything else followed from that.

What the founding mythology tends to skip is the structural reality. The $1 billion pledge was a pledge. By 2019, only $130 million had actually arrived. The "open" in OpenAI had been quietly dropped by the time GPT-3 launched as a closed API in 2020. And the "nonprofit" was restructured into a capped-profit entity just four years after founding, with Sam Altman as CEO and Microsoft as the anchor investor.^[3]

The real story of OpenAI is a decade of tension between what the founders said they were building and what they actually built. Both things are interesting. But only one of them became a $300 billion company.

The Founding: A Dinner Party That Built a Lab

OpenAI was incorporated on December 8, 2015, with an unusual founding team for a Silicon Valley AI company: it had eleven co-founders, a nonprofit structure, and no product.

The eleven founders were Sam Altman, Elon Musk (as co-chairs), Ilya Sutskever, Greg Brockman, Trevor Blackwell, Vicki Cheung, Andrej Karpathy, Durk Kingma, John Schulman, Pamela Vagata, and Wojciech Zaremba.^[1] The roster reads like a draft pick of the 2015 AI research world. Sutskever had just helped found Google Brain after co-authoring AlexNet with Geoffrey Hinton. Karpathy was finishing his Stanford PhD on deep learning for vision. Schulman would go on to co-develop the reinforcement learning from human feedback (RLHF) techniques that made ChatGPT possible. Brockman had been CTO of Stripe. This was not a group short on ambition or credential.

The financial backing was equally high-profile. Sam Altman, Greg Brockman, Elon Musk, Reid Hoffman, Jessica Livingston, Peter Thiel, Amazon Web Services, and Infosys pledged a combined $1 billion. The operative word is "pledged." By 2019, company disclosures showed only $130 million had actually been received. The gap between the announcement and the reality would become a recurring theme.

The stated mission was to "advance digital intelligence in the way that is most likely to benefit humanity as a whole." The key phrase was that because OpenAI received no financial return from its work, it would have no conflicting interests. This was the founding promise: an AI lab that could be trusted because it had nothing to gain.

The early years were genuinely research-forward. In 2016, OpenAI released OpenAI Gym, an open-source toolkit for reinforcement learning, and Universe, a platform for training AI agents. In 2017, the team spent $7.9 million on cloud compute, 25% of its total functional expenses, a number that sounds modest until you compare it to DeepMind's $442 million in total expenses that same year. The resource gap was already visible.^[5]

The research outputs were real. OpenAI Five, the Dota 2 playing system, demonstrated that reinforcement learning could master extraordinarily complex multi-agent games. Dactyl, a robotic hand trained entirely in simulation, showed that sim-to-real transfer was further along than the field believed. These were not demo projects. They were genuine research milestones that shaped how the field thought about AI capability.

The First Crack: Musk Leaves, GPT-1 Arrives (2018)

The year 2018 was the first in which the founding narrative started visibly fracturing.

In February, Elon Musk resigned from OpenAI's board. The stated reason was potential conflict of interest with Tesla's own AI development for autonomous vehicles. Musk remained a donor and supporter, publicly. Privately, a different story was already taking shape. Musk had pushed hard to take operational control of OpenAI, proposing to merge it with Tesla or to become CEO himself. When the board declined, he left.^[6] The rift would not become public for years, but the relationship was already broken in 2018.

The same year, OpenAI published GPT-1: a language model with 117 million parameters trained on 7,000 unpublished books. It was the first model in what became the most consequential model family in AI history. GPT-1 could generate coherent text, complete sentences, and answer basic questions. Nobody outside the field noticed. That was about to change.

The GPT-2 Controversy and the Capped-Profit Pivot (2019)

February 2019 was the moment OpenAI became a subject of public debate rather than just academic interest.

The company announced GPT-2, a language model with 1.5 billion parameters capable of generating remarkably convincing fake text. OpenAI's decision was to not release it. The stated reason: fear of misuse, specifically fake news generation. The actual release would be staged over nine months, with progressively larger model sizes published as the research community demonstrated limited evidence of actual misuse.

The AI community's reaction was split. Some praised the caution as responsible disclosure. Others accused OpenAI of using "safety" as cover for competitive advantage and publicity. The debate prefigured arguments that would play out repeatedly over the next decade: who decides what AI capabilities are too dangerous to publish, and what interests drive those decisions?

One month later, in March 2019, OpenAI made a far more consequential announcement with far less fanfare: the creation of OpenAI LP, a new "capped-profit" entity.^[3] The structure was novel. Investors could receive returns of up to 100 times their investment. The nonprofit board retained control. Profits beyond the 100x cap would go to the nonprofit mission. The pitch to investors was that frontier AI required billions in compute and talent that a nonprofit could never attract.

Sam Altman stepped down from Y Combinator to become CEO of OpenAI. Three months later, Microsoft announced a $1 billion investment in OpenAI LP, the beginning of what became the most important bilateral partnership in AI history.^[7]

GPT-3 and the Closed Garden (2020)

June 2020. OpenAI released GPT-3 with 175 billion parameters, nearly 100 times larger than GPT-2. It was the most capable language model the world had seen, and OpenAI did not open-source it.

GPT-3 was available only through an API with a waitlist. The "open" in OpenAI had been functionally abandoned. The justification was safety and responsible deployment. The practical effect was a competitive moat. Every capability demonstration, every integration, every research paper built on GPT-3 ran through OpenAI's servers and generated revenue for OpenAI's bottom line.

GPT-3 convinced the technology industry that large language models were something qualitatively different from previous AI systems. The emergent capabilities, code generation, translation, summarization, few-shot learning across arbitrary domains, were not things the field had predicted at that scale. OpenAI had discovered something real, and they had the only copy.^[8]

The same year, DALL-E was announced (January 2021), Codex launched as the foundation for GitHub Copilot, and Microsoft's investment relationship deepened into exclusive licensing of GPT-3's underlying technology. By 2021, OpenAI's valuation had reached $14-20 billion according to VC investment rounds. The nonprofit phase was operationally over.

The Year of Reckoning: 2022

The financial picture in 2022 was alarming. OpenAI reported $540 million in losses on $28 million in revenue. The compute costs to train increasingly large models were scaling faster than any revenue model could cover. Microsoft's $1 billion was not going to last. The company needed either a product that could generate billions in revenue or another massive investment round.^[9]

It got both.

On November 30, 2022, OpenAI launched ChatGPT as a free research preview. It was built on GPT-3.5, a more efficient model than the full GPT-3, and it had a conversational interface that made the underlying capability accessible to anyone with an internet connection. No API key. No prompt engineering. Just a chat box.

Within two months, ChatGPT had 100 million users. It was the fastest consumer application in history to reach that milestone, surpassing TikTok. Every boardroom in the world ran a demo. Every government started a working group. Every AI competitor accelerated their roadmap. The race that the founding founders had theoretically been running since 2015 became a sprint that everyone could see.

January 2023: Microsoft Goes All In

In January 2023, Microsoft announced a multi-year investment of $10 billion in OpenAI, bringing its total commitment to more than $13 billion. The deal structure was reported to give Microsoft 49% of OpenAI LP's for-profit profits until the investment was recouped, and up to a 75% revenue share during that period.^[7]

The partnership had been building since 2019, but the $10 billion announcement was categorically different. This was not a bet on AI in general. It was Microsoft integrating OpenAI's models into every product it shipped: Bing, Azure, Office, GitHub Copilot, Windows Copilot. The Copilot branding became the Microsoft AI identity. OpenAI's technology became the substrate of Microsoft's entire AI strategy.

For OpenAI, the arrangement was existential: without Microsoft's Azure compute infrastructure, there was no viable path to training frontier models. The cloud costs to train GPT-4, released in March 2023, ran into the hundreds of millions of dollars. Microsoft was not just an investor. It was the infrastructure provider on which OpenAI's existence depended.

GPT-4 landed in March 2023 with multimodal capabilities (text and images), significantly improved reasoning, and a 32,000-token context window. It outperformed GPT-3.5 on virtually every benchmark and passed bar exams, medical licensing tests, and graduate-level reasoning evaluations at human-expert level. The system that had started with 117 million parameters in 2018 now had parameters measured in the hundreds of billions and capabilities that the founders of OpenAI had not imagined when they had dinner in San Francisco in 2015.

Dev Day and the Week from Hell: November 2023

On November 6, 2023, OpenAI hosted its first developer conference in San Francisco. Sam Altman unveiled GPT-4 Turbo (128K context, cheaper, faster), custom GPTs (mini-apps built on ChatGPT), the Assistants API for building AI agents, and DALL-E 3. It was the company at peak momentum: 100 million users, a dominant market position, a product suite that covered text, code, and image generation, and a CEO who was on the cover of every business magazine. Eleven days later, the board fired him.

The timeline of what happened between November 17 and November 21, 2023 remains the most analyzed five days in Silicon Valley history.

The post-mortem that emerged over the following months was complicated. The board's stated reason, lack of candor, was vague enough to suggest deeper disagreements. Subsequent reporting indicated that some employees had raised concerns about how Altman handled the safety implications of an internal AI capability discovery shortly before the firing. Others pointed to disagreements about the pace of commercialization versus safety research. The board had believed it was acting within its governance authority. It had not accounted for the fact that its governance authority meant nothing if every employee and every investor chose to override it.

What the crisis proved, conclusively, was that the nonprofit board structure Altman and Musk had designed in 2015 was not actually in control of OpenAI in 2023. The for-profit entity, its investors, and its employees were. The board could theoretically fire the CEO. It could not survive the attempt.

The Brain Drain: Ilya, Jan Leike, and the Safety Exodus (2024)

May 2024. Ilya Sutskever announced he was leaving OpenAI. The statement was brief and characteristically understated. He had "a big new vision," he said. It was more suitable for a new company.

What Sutskever did not say was the more important part. In a deposition taken in October 2025 as part of Elon Musk's lawsuit against OpenAI, Sutskever described years of accumulating tension inside the company: doubts about Altman's judgment, eroding trust in the leadership's priorities, and a growing sense that the direction OpenAI was heading was incompatible with the research agenda he believed mattered most.^[10]

Hours after Sutskever's departure announcement, Jan Leike, the head of OpenAI's superalignment team (the group tasked with ensuring advanced AI remained safe and aligned), also resigned. Leike was less circumspect. He posted that OpenAI had "drifted from its core mission" and that safety culture and processes had "taken a back seat to shiny products." It was as direct a public indictment of a company's priorities as a senior researcher had ever published about an employer in real time.^[11]

The departures were not isolated. John Schulman, one of the original eleven founders and a key architect of the RLHF training methods that enabled ChatGPT, also left in 2024. Greg Brockman announced a sabbatical in September 2024 and did not return. By mid-2024, only 4 of OpenAI's original 11 co-founders remained with the company.

In June 2024, Sutskever announced his new company: Safe Superintelligence Inc. (SSI), co-founded with Daniel Gross and Daniel Levy, with offices in Palo Alto and Tel Aviv. The mission statement was deliberately maximal and product-free: SSI would build "safe superintelligence" in a "straight shot," with no other products and no revenue-generating distractions until that goal was achieved.^[12]

SSI raised $1 billion in September 2024 from Andreessen Horowitz, Sequoia, DST Global, and SV Angel. By March 2025, it had raised a further $2 billion at a $32 billion valuation. In June 2025, Meta attempted to acquire SSI. Sutskever declined. His co-founder and CEO Daniel Gross left for Meta. Sutskever became CEO of SSI himself. The lab that Ilya built, the one that was supposed to be OpenAI's soul, had been reconstituted as a separate company that Meta failed to buy.

The Musk War: Lawsuits, xAI, and the Long Grudge

Elon Musk and OpenAI's relationship was never fully repaired after his 2018 board departure. In March 2024, it became a legal matter.

Musk filed suit against OpenAI and Sam Altman, arguing that the company's conversion to a for-profit structure violated the founding agreement he had helped create and the charitable contributions he had made in reliance on the nonprofit structure.^[6] The lawsuit was expansive: it alleged breach of contract, breach of fiduciary duty, and violations of California's unfair competition law.

OpenAI's response was pointed. The company released documents purporting to show that Musk himself had agreed to and pushed for a for-profit structure when it became clear that the nonprofit could not attract the capital needed for frontier AI development. OpenAI's lawyers argued the lawsuit was "yet another front in Mr. Musk's ongoing campaign of harassment" rather than a legitimate legal grievance.

A parallel xAI trade-secrets lawsuit, accusing OpenAI of receiving stolen technical information from a former xAI engineer, was dismissed by a US federal judge on February 24, 2026. The core lawsuit over the for-profit conversion remains in litigation.

The Musk-Altman dynamic is perhaps the defining interpersonal feud of the AI era. Two people who shared a founding vision in 2015 are now running competing AI companies and suing each other in federal court. Musk's Grok, built by xAI, competes directly with ChatGPT. The acrimony between them generates more coverage than most actual AI research.

Stargate and the $300 Billion Reality (2025)

January 21, 2025. At the White House, Sam Altman, SoftBank CEO Masayoshi Son, and Oracle Chairman Larry Ellison announced Project Stargate: a $500 billion initiative to build AI infrastructure across the United States over four years, backed by SoftBank, Oracle, and OpenAI as the primary partners. The first $100 billion was committed immediately. The announcement came the day after Donald Trump's inauguration.

Three months later, in April 2025, OpenAI closed a $40 billion funding round led by SoftBank. The post-money valuation: $300 billion. It was the largest private technology deal in recorded history, surpassing the previous record by a substantial margin.^[4]

By July 2025, OpenAI reported annualized revenue of $12 billion, up from $3.7 billion in all of 2024. ChatGPT had 20 million paid subscribers. Five million businesses were paying customers. The free preview that launched on November 30, 2022 had become a multi-billion-dollar subscription and enterprise software business in under three years.

On October 28, 2025, OpenAI completed its for-profit restructuring. The company became OpenAI Group PBC, a public benefit corporation. The OpenAI Foundation, the surviving nonprofit entity, retained equity in the for-profit currently valued at approximately $130 billion, making it one of the most valuable philanthropic organizations in the world by asset size. Microsoft, which had been the company's anchor investor and infrastructure partner since 2019, received equity as part of the restructuring, valued at approximately $135 billion at closing.^[2]

What OpenAI Actually Is in 2026

The company that emerged from a decade of structural contradiction is something its founders would recognize in some respects and not at all in others.

On the product side, OpenAI is the dominant consumer AI brand globally. ChatGPT has no close competitor for mindshare. The GPT model family, now extending through GPT-4, GPT-4o, o1, o3, GPT-4.5, GPT-4.1, o4-mini, GPT-5, GPT-5.1, and GPT-5.2, spans reasoning, coding, vision, and multimodal tasks. The Sora video generation model, Whisper for speech recognition, and DALL-E for image generation extend the product surface well beyond text. Operator, their autonomous agent product, competes in the agentic AI market that every major lab is racing to define.

On the research side, the picture is more contested. The departures of Sutskever, Leike, and Schulman removed three of the architects of OpenAI's most important research contributions. The superalignment team that Sutskever and Leike built, tasked with figuring out how to ensure advanced AI remains safe, has been restructured multiple times. The publication record for fundamental research has declined relative to the company's early years, when it freely published work on reinforcement learning, robotics, and language models.

On the mission side, the distance from the founding vision is hardest to measure. The nonprofit board that fired Sam Altman in November 2023 was acting on a sincere belief that the for-profit pressure was overriding the safety-first mandate. The investor-employee-market coalition that reinstated him believed the mission required the company to survive and grow. Both positions contain truth. The restructuring that completed in October 2025 chose a side.

Last updated: February 26, 2026