On a Friday afternoon in November 2023, at approximately 12:19 p.m. Pacific Time, the OpenAI nonprofit board of directors published a 152-word announcement on the company’s website stating that Sam Altman had been removed as chief executive officer. The announcement specified that Altman had “not been consistently candid in his communications with the board, hindering its ability to exercise its responsibilities.” Mira Murati, the company’s chief technology officer, would serve as interim chief executive. The decision had been made and executed across approximately seventy-two hours of internal board deliberation that the company’s senior leadership had not been informed about until the public announcement itself. Greg Brockman, the company’s president and co-founder, resigned within four hours of the announcement in protest. The OpenAI Slack channels filled with employee responses across the following 24 hours. By Sunday November 19, Microsoft chief executive Satya Nadella had publicly offered to hire Altman and any departing OpenAI employees into a newly-established Microsoft AI research division. By Monday November 20, approximately 770 of OpenAI’s 800 employees had signed an open letter threatening collective resignation if the board did not reinstate Altman. By Wednesday November 22, the board had agreed to Altman’s reinstatement and to its own reconstitution, with Bret Taylor (the former Salesforce co-chief executive) and Larry Summers (the former U.S. Treasury Secretary) joining as new board members.
The five-day crisis at OpenAI ran the structural inflection event for the operation that had, by November 2023, established itself as the dominant infrastructure company of the 2020s artificial intelligence industry. The company had launched ChatGPT approximately one year earlier on November 30, 2022, with the consumer conversational interface crossing 100 million weekly active users inside two months of release and establishing itself as the fastest-growing consumer product in technology-industry history. The company’s underlying GPT-3.5 and GPT-4 language models had generated approximately $1.3 billion in 2023 revenue against an estimated $5.4 billion in operating costs, with the structural unit-economic gap covered by Microsoft’s $10 billion investment commitment announced in January 2023. The company’s most recent secondary-share tender offer, completed in early 2023 at approximately $29 billion valuation, had been superseded by ongoing negotiations that would value the company at $86 billion by October 2024, $157 billion by February 2025, and $500 billion by October 2025 across the successive valuation rounds the company’s continued operation generated.
The 2020s OpenAI operation inverted the 2010s WeWork operation while extending the structural pattern across every load-bearing variable. WeWork had run a real-estate sublease operation dressed up as a technology company through marketing language including “elevate the world’s consciousness,” “Space-as-a-Service,” and “the future of work,” with the structural gap between the underlying commercial-real-estate business and the technology-company marketing narrative driving the August-to-October 2019 collapse. OpenAI runs an actual technology operation that has built and deployed the most consequential consumer technology product since the iPhone. The structural alignment between the underlying technology and the marketing narrative routes substantially closer at OpenAI than the WeWork comparable. The structural unit-economic gap, however, runs comparable in magnitude: OpenAI generates approximately $5 billion in 2024 revenue against approximately $9 billion in 2024 operating losses, against WeWork’s $1.8 billion in 2018 revenue against $1.6 billion in 2018 operating losses. The structural difference: OpenAI’s underlying technology actually works, the addressable market is theoretically global rather than confined to commercial real estate, and the long-arc precedent (Amazon ran twenty years of operating losses before achieving sustained operating-leverage profitability) suggests that the WeWork-pattern catastrophic-collapse outcome is not the only possible structural resolution.
OpenAI was not just a technology company. OpenAI was the structural question that the 2020s artificial intelligence infrastructure deployment cycle had not yet resolved.
The San Francisco Origin
OpenAI was founded on December 11, 2015 as a nonprofit research organization. The founding announcement listed Sam Altman (then president of the Y Combinator startup accelerator), Elon Musk (chief executive of Tesla and SpaceX), Greg Brockman (the recently-departed chief technology officer of Stripe), Ilya Sutskever (a research scientist at Google Brain), Wojciech Zaremba (a research scientist at Facebook AI Research), and John Schulman (a doctoral candidate at UC Berkeley) as the core founding team. The initial capital commitments totaled approximately $1 billion across a founding investor group that included Musk personally, Reid Hoffman, Peter Thiel, Y Combinator, Amazon Web Services, Infosys, Sequoia Capital, and the Open Philanthropy Project effective-altruism research organization.
The structural founding mission, as articulated in the December 2015 announcement and subsequent founding documents, ran ensuring artificial general intelligence (AGI) “benefits all of humanity” through open research, model-sharing infrastructure, and broader-public-benefit-oriented operation. The nonprofit-organization structure routed the company outside the conventional venture-capital-funded technology-startup framework, with the founders articulating an explicit position that AGI research should not be controlled by any single for-profit entity and that the resulting research outputs should be shared publicly with the broader scientific community rather than retained as proprietary commercial intellectual property.
The 2018 Musk departure from the OpenAI board followed ongoing strategic disagreements between Musk and the broader founding team regarding the operation’s direction. Musk had argued for a more aggressive commercialization trajectory and had proposed that he assume direct operating control of the company through a merger structure that would have placed OpenAI under Tesla’s organizational umbrella. The board declined the proposal. Musk resigned from the board in February 2018, citing potential conflicts of interest between his Tesla AI work and OpenAI’s research operation. The Musk departure also routed through the partial cessation of his personal financial commitments to the organization, with approximately $100 million of his original $1 billion commitment having been deployed by the time of his departure.
The March 2019 announcement of OpenAI LP, the for-profit subsidiary structure, ran the structural restructuring that allowed the organization to raise capital at scale beyond what the nonprofit research-organization framework could accommodate. The structural arrangement: OpenAI Nonprofit (the original 501(c)(3) organization) maintained governance control over OpenAI LP through a board-appointment structure, with OpenAI LP operating as a “capped-profit” entity that limited investor returns to 100x the initial investment before any additional profits would route back to the nonprofit parent. The capped-profit structure represented an unusual hybrid arrangement that the conventional venture-capital and corporate-governance infrastructure had not previously deployed at scale. The founding investors in OpenAI LP included Microsoft, which committed $1 billion in July 2019 as the first major capital-and-cloud-infrastructure partnership routing OpenAI’s compute requirements through Microsoft Azure.
The Model Deployment Specification
The GPT model series ran the technical infrastructure that the company’s commercial deployment subsequently scaled across. GPT-1, released in June 2018 at 117 million parameters, ran the original transformer-architecture proof-of-concept that demonstrated the language-model training methodology that Google researchers had introduced in the June 2017 “Attention Is All You Need” paper could be scaled into a deployable language-generation system. GPT-2, released in February 2019 at 1.5 billion parameters, ran the first model to generate sustained cultural-discourse coverage outside the machine-learning research community, with OpenAI initially withholding the full model from public release citing concerns about potential misuse before ultimately releasing the complete model across staged deployments through November 2019.
GPT-3, released in June 2020 at 175 billion parameters, ran the first commercial API deployment that established OpenAI’s revenue-generation infrastructure. The API allowed third-party developers to integrate GPT-3’s text-generation capabilities into their own applications through paid-access pricing that ran approximately $0.02 per 1,000 tokens at the initial release. The API generated approximately $30 million in 2021 revenue across the early commercial deployment window, with the customer base running primarily through technology-startup developers building consumer applications on the underlying language-model infrastructure.
ChatGPT launched on November 30, 2022 as a conversational interface running on the GPT-3.5 model variant. The launch was structured as a “research preview” with free access for the initial deployment window. The product crossed 1 million users inside five days, 100 million weekly active users inside two months, and 600 million weekly active users by the end of 2024. The structural significance ran beyond the user-count growth: ChatGPT established the conversational interface as the dominant consumer interaction modality for large language models, displacing the prior generation’s command-line and API-developer interfaces with a chat-window infrastructure that approximately matched the cognitive load that consumer messaging applications had already trained users to operate through.
GPT-4 launched in March 2023 at undisclosed parameter count (industry estimates ran approximately 1 trillion parameters across a mixture-of-experts architecture), integrated into ChatGPT Plus subscription at $20 per month per user. The subscription tier generated the operation’s primary subscription-revenue infrastructure across the 2023-to-2024 window, with approximately 7.7 million ChatGPT Plus subscribers generating approximately $1.8 billion in annual subscription revenue by mid-2024. GPT-4o launched in May 2024 with integrated multimodal capabilities (text, image, voice). The o1 model launched in September 2024 with explicit reasoning-chain functionality. The o3 model launched in December 2024 with significantly expanded reasoning capabilities. GPT-5 launched in August 2025 as the next-generation foundational model. The continued model-release cycle ran approximately four-to-six major releases per year across the 2024-to-2025 window.
The DALL-E image generation infrastructure ran the parallel commercial deployment alongside the text-generation infrastructure. DALL-E launched in January 2021 as a 256-by-256 pixel resolution research system. DALL-E 2 launched in April 2022 at 1024-by-1024 pixel resolution. DALL-E 3 launched in October 2023 integrated into ChatGPT’s conversational interface. Sora, the text-to-video model, launched in February 2024 with public-deployment access expanding across the 2024-to-2025 window. The Codex coding-assistant infrastructure, deployed across the GitHub Copilot integration with Microsoft, generated additional revenue through Microsoft’s enterprise distribution channels.
The structural scaling: revenue growth from approximately $30 million in 2021 to approximately $1.6 billion in 2023 to approximately $5 billion in 2024 to projected $11 billion in 2025 according to internal financial disclosures that subsequently routed into press coverage across Bloomberg, The Information, Reuters, and The Wall Street Journal across the 2024-to-2025 reporting window.
The Microsoft Partnership and Capital Stack
The Microsoft investment progression ran the structural financing infrastructure that the operation’s compute-cost scale required. Microsoft’s initial $1 billion investment in July 2019 routed through a multi-year arrangement that provided OpenAI with Azure cloud-compute infrastructure at preferential pricing against revenue-share commitments and intellectual-property licensing access for Microsoft. The structural arrangement allowed OpenAI to operate compute infrastructure at scale that the company’s revenue base could not directly support, with Microsoft effectively underwriting the compute-cost subsidy that the language-model training and inference operations required.
The January 2023 announcement of Microsoft’s additional $10 billion investment commitment ran the structural escalation following ChatGPT’s November 2022 commercial breakthrough. The $10 billion commitment included additional Azure compute capacity, a structural revenue-share arrangement allowing Microsoft to recover approximately 75 percent of OpenAI profits until the investment recoupment, and continued intellectual-property licensing access for Microsoft’s product integration including Microsoft Copilot, Bing Chat, and the broader Microsoft 365 integration infrastructure. The additional follow-on investments across 2024 brought Microsoft’s cumulative committed capital to approximately $13 billion against approximately $10 billion in directly-deployed cash, with the remaining commitments routing through ongoing Azure compute-capacity provision rather than direct cash transfers.
The 2024 SoftBank and Thrive Capital tender-offer rounds ran the secondary-market valuation progression. The October 2024 tender at approximately $157 billion valuation allowed employees and early investors to monetize equity positions through secondary share sales without the company itself raising primary capital. The February 2025 tender at approximately $300 billion ran similar secondary-share liquidity at the further-elevated valuation. The October 2025 tender at approximately $500 billion ran the most recent secondary-share valuation event. The structural scaling: from $29 billion valuation in early 2023 to $500 billion valuation in October 2025 represented approximately a 17x valuation expansion across approximately thirty months.
The compute-cost economics ran the structural unit-economic challenge. Industry-press estimates of OpenAI’s 2024 compute costs ran approximately $7 billion across Azure infrastructure provision, against $5 billion in 2024 revenue. The structural gap of approximately $2 billion in operating losses (the $9 billion total 2024 losses included broader operating expenses across the approximately 4,000-person organizational infrastructure beyond direct compute costs) required continuous capital injection to sustain the operation. The structural question that the capital stack generated: whether the compute-cost economics would eventually route into structural operating-leverage profitability through model-efficiency improvements, scale-economy unit-cost reductions, and revenue-growth multiples that would close the unit-economic gap, or whether the structural gap would persist or widen across the continuing model-development cycle.
The Microsoft relationship complications across 2024-2025 ran additional structural tension. Microsoft had developed its own internal AI research and product infrastructure across the 2023-to-2025 window, with Microsoft AI under Mustafa Suleyman (the co-founder of DeepMind who had joined Microsoft in March 2024) developing capabilities that ran parallel to OpenAI’s commercial deployments. The structural ambiguity between partnership and competition routed through ongoing renegotiations of the contractual terms between the two companies, with the eventual restructured arrangement remaining subject to ongoing negotiation at the time of writing.
The November 2023 Board Crisis
The November 17, 2023 board meeting that voted to remove Altman as chief executive ran across approximately seventy-two hours of internal deliberation that the broader OpenAI leadership had not been informed about. The board at the time of the vote consisted of six members: Altman himself, Brockman, Sutskever, and the three independent directors Adam D’Angelo (chief executive of Quora), Tasha McCauley (a technology entrepreneur), and Helen Toner (an AI policy researcher at Georgetown University’s Center for Security and Emerging Technology). The vote excluded Altman and Brockman from the deliberation, with Sutskever joining the three independent directors in the four-to-zero vote to remove Altman.
The public announcement at 12:19 p.m. Pacific Time on Friday November 17 generated immediate response across the OpenAI organizational infrastructure. Brockman, who had been informed of his own removal from the board approximately thirty minutes before the public announcement, resigned as president within four hours and posted a public statement contesting the board’s actions. The senior leadership team, including Murati (the newly-appointed interim chief executive), Jakub Pachocki, and the broader research-and-engineering leadership, requested the board provide specific evidence supporting the “not consistently candid” framing. The board declined to provide additional specificity.
The structural escalation across the weekend ran multiple parallel tracks. Microsoft chief executive Satya Nadella conducted continuous negotiations across Friday evening, Saturday, and Sunday with Altman, Brockman, the OpenAI board, and senior OpenAI investors including Thrive Capital’s Joshua Kushner and Tiger Global. On Sunday November 19, Microsoft publicly announced that Altman and Brockman would join Microsoft to lead a newly-established advanced AI research division. The Microsoft announcement routed the structural pressure onto the OpenAI board: if approximately 770 of OpenAI’s 800 employees departed for Microsoft, as the open letter signed across Sunday and Monday indicated they would, the remaining OpenAI nonprofit organization would retain the underlying intellectual property and the cash reserves but would have lost the structural human-capital infrastructure that the operation required to continue at commercial scale.
Sutskever publicly reversed his position on Monday November 20 through a Twitter post acknowledging that he “deeply regret[ted] participation in the board’s actions.” The reversal removed the structural majority that had supported Altman’s removal. The continuing negotiations across Monday and Tuesday routed through Microsoft’s continued offer of employment for departing OpenAI staff, the open letter’s continued accumulation of signatures (eventually reaching approximately 770 of 800 employees), and the increasingly evident structural reality that the board’s position could not be sustained against the combined employee-and-investor pressure.
The board agreed to Altman’s reinstatement on Wednesday November 22 at approximately 10 p.m. Pacific Time. The simultaneous board reconstitution included Bret Taylor, the former Salesforce co-chief executive who had also served on Twitter’s board through the Elon Musk acquisition process, as the new board chair. Larry Summers joined as an additional new board member. Adam D’Angelo remained on the board. McCauley and Toner departed. Sutskever subsequently departed OpenAI in May 2024 to found Safe Superintelligence Inc., a parallel AI research organization explicitly focused on superintelligent AI safety research.
The structural significance of the five-day crisis routed through what it demonstrated about the practical operating structure of the nonprofit-controlled for-profit subsidiary arrangement. The nonprofit board’s governance authority over the for-profit subsidiary, as specified in the founding documents and the OpenAI LP structure, theoretically allowed the nonprofit board to remove executives, redirect strategy, and override operational decisions on the basis of the broader public-benefit mission. The November 2023 crisis demonstrated that, in practice, the nonprofit board could not sustain such authority against coordinated employee-and-investor pressure backed by Microsoft’s structural capital-and-employment-alternative leverage. The for-profit subsidiary had accumulated operational mass that the nonprofit governance structure could not, in actual operation, override.
The Equipment Cancellation
The structural questions that remain open at the time of writing close Season 2.
The for-profit conversion controversy continued across 2024-2025 as OpenAI attempted to restructure from its current nonprofit-controlled for-profit subsidiary into a conventional for-profit corporation. The structural conversion, as proposed across various intermediate filings with the Internal Revenue Service and the California Attorney General’s office, would route the nonprofit parent organization into a passive philanthropic-grantmaking role while the for-profit subsidiary would become the dominant operating entity. The conversion generated significant cultural-discourse coverage running across whether the structural shift violates the founding mission’s articulated commitment to AGI development serving the broader public benefit rather than concentrated for-profit shareholder interests. The Elon Musk lawsuits filed across 2024 alleged that the for-profit conversion violates the original founding agreements that Musk had signed in December 2015, with the litigation remaining active at the time of writing.
The compute-cost economics continue requiring continuous multi-billion-dollar capital injection at the operation’s structural scale. The Stargate project, announced in January 2025 as a $500 billion infrastructure-deployment initiative across data-center construction in collaboration with SoftBank, Oracle, and additional investor partners, represented the most ambitious single capital-deployment commitment in the company’s history. The structural question: whether the Stargate-scale infrastructure deployment can be financed through the continued capital-markets infrastructure that has supported the operation across the 2022-to-2025 window, or whether the structural-cost requirements will exceed even the expanded capital base that the rising valuations have generated.
The competitive landscape continues generating sustained pressure on OpenAI’s market position. Anthropic, founded by former OpenAI researchers Dario and Daniela Amodei in 2021, operates the Claude model series as the primary commercial competitor in the U.S. market. Google DeepMind operates the Gemini model series with Google’s broader cloud-infrastructure and consumer-platform integration. Meta AI operates Llama as an open-source competitive infrastructure. xAI, the Musk-founded competitor, operates Grok across the X platform integration. The broader Chinese AI infrastructure including DeepSeek (whose January 2025 R1 model release demonstrated capabilities approximating OpenAI’s o1 reasoning model at substantially reduced compute-cost economics), Alibaba’s Qwen, and ByteDance’s Doubao runs the parallel competitive infrastructure that the U.S.-based deployment may not be positioned to fully address.
The structural question that closes Season 2 routes through three possible interpretation frameworks. The Amazon-pattern framework reads OpenAI’s current structural unit-economic gap as the long-arc investor-patience trajectory that Amazon ran from 1997 through approximately 2015, with eventual operating-leverage profitability emerging across the model-efficiency improvements, scale-economy unit-cost reductions, and revenue-growth multiples that the continuing commercial deployment will generate. The WeWork-pattern framework reads the current structural valuation as a venture-capital growth-narrative bubble that the eventual public-market test or capital-cycle correction will resolve into catastrophic collapse comparable to the 2019 WeWork sequence. The third-outcome framework acknowledges that the platform-distribution-and-AI-infrastructure decade has not previously demonstrated the structural patterns that the prior two interpretation frameworks would predict, with the possibility that the eventual resolution will route through structural patterns that no prior decade-cycle has yet illustrated.
Lehman closed the 2000s decade in a single week of cascading institutional failure. WeWork closed the 2010s decade across six weeks of S-1 reception, IPO withdrawal, and emergency financing. The 2020s decade-end remains approximately four years away at the time of writing. OpenAI may yet operate as the structural success story that validates the entire 2020s AI-infrastructure deployment cycle, with the company resolving into sustainable commercial operation comparable to Google’s resolution from late-1990s startup into the dominant search infrastructure of the subsequent twenty years. OpenAI may yet operate as the structural collapse that the prior decades’ patterns suggest, with the unit-economic gap and the governance complications routing into eventual catastrophic correction comparable to the WeWork and Lehman precedents. OpenAI may yet operate as a third structural outcome that no prior decade-closer has illustrated, with the operation continuing across some sustainable hybrid configuration that resolves the underlying structural questions through patterns that the existing capital-markets and technology-deployment infrastructures have not previously demonstrated.
The story continues. The decade continues. The next decade approaches. Season 2 closes here, at the structural question that the 2020s have not yet resolved, with the resolution running across the remaining decade window and into whatever the next decade’s closer eventually becomes.
· · ·