General · EdgepediaSeptember 17, 202610 min read

OpenAI

OpenAI is an artificial intelligence company that develops frontier language and reasoning models, sold through the ChatGPT chatbot. It began as a nonprofit research laboratory with the stated mission of building safe artificial general intelligence (AGI), reorganized in 2019 around a capped-profit subsidiary to raise the capital its compute-intensive research required, and in October 2025 completed a recapitalization that made the commercial arm a Delaware public benefit corporation, OpenAI Group PBC, still controlled by the nonprofit, now named OpenAI Foundation.¹ Its flagship products included the o-series reasoning models o3 and o4-mini in April 2025, and ChatGPT, which the company said was used by over 300 million people weekly, most of them for free.²

Key fact	Detail
2019 restructuring	Converted to a capped-profit structure in 2019 after estimating it needed on the order of $10 billion to build AGI²
Current structure	OpenAI Foundation (nonprofit) holds 26% of OpenAI Group PBC equity, valued at about $130 billion, and appoints the entire PBC board through Class N stock¹
Valuation	$500 billion after an October 2025 secondary sale of $6.6 billion in employee shares to Thrive Capital, Dragoneer, T. Rowe Price, SoftBank, and MGX³
Revenue	Audited 2025 figures show $13.07 billion in revenue and a $38.53 billion net loss attributable to the company; the company separately confirmed an annualized run rate of $20 billion or more in 2025⁴^•⁵^•²⁵
Compute	Ended 2025 with 1.9 GW of compute, more than triple the start of the year⁶^•⁷
Flagship models	o3 and o4-mini (April 2025), reasoning models that can agentically use ChatGPT tools including web search, Python, and image analysis⁸
Safety framework	Preparedness Framework version 2 (April 15, 2025), with High and Critical capability thresholds per tracked category⁹

History and corporate structure

OpenAI began as a nonprofit research center aiming to build safe artificial general intelligence. Its early funding came in the form the nonprofit sector provides: $137 million in cash donations, less than a third of which came from Elon Musk, plus compute credits and discounts including $1.8 million from Amazon and $50 million or more each from Azure and Google Cloud.² By 2019 the organization needed money for the expensive computing power required to run and test its models, and the founders created a "capped-profit" commercial arm governed by the nonprofit board, with Sam Altman as CEO and excess value reinvested into OpenAI.¹⁰ Co-founder Ilya Sutskever described the intent to VOX in 2019 as a structure that would allow raising more money while adhering to the original mission as much as possible.¹⁰

The capped-profit mechanism. Under the OpenAI LP structure, investors and employees could receive a capped return if the mission succeeded, allowing the lab to raise investment capital and offer startup-like equity; any returns beyond the cap were owned by the original nonprofit entity.¹¹ A Lawfare analysis argues that reliance on donations, even millions of dollars of them, was a poor strategy for such a capital-intensive endeavor, which is what prompted the creation of the for-profit subsidiary.¹² The for-profit raised an initial round of over $100 million, followed by $1 billion from Microsoft.²

The 2024–2025 recapitalization

In May 2025 OpenAI announced it would convert its commercial subsidiary into a Delaware public benefit corporation (PBC), a corporate form with a stated public benefit interest, while maintaining the two-tier governance in which the nonprofit controls commercial operations.¹³ The company said the nonprofit's interest would be valued by independent financial advisors and that the OpenAI mission would become the PBC's public benefit interest.²

The conversion closed in October 2025, after the Delaware and California attorneys general, Kathy Jennings and Rob Bonta, said they would not oppose the restructuring, ending more than a year of negotiations.¹⁴ The nonprofit was renamed OpenAI Foundation and the subsidiary became OpenAI Group PBC.¹ The Foundation holds 26% of the PBC's equity, valued at approximately $130 billion at the time of the recapitalization, and retains control through a special class of stock, Class N, carrying the right to appoint all members of the PBC board.¹ At the close of the restructuring the Foundation held 26% of shares, Microsoft about 27%, and current and former employees and investors the remaining 47%.¹⁵ The Foundation also gains additional ownership at future valuation milestones.¹⁶

Models and products

In April 2025, OpenAI released o3 and o4-mini as the latest in the o-series of reasoning models trained to think for longer before responding. They can agentically use ChatGPT's tools, including web search, Python, image analysis, and image generation.⁸ o3 set a new state of the art on Codeforces, SWE-bench without a custom model-specific scaffold, and MMMU, and in evaluations by external experts makes 20 percent fewer major errors than o1 on difficult, real-world tasks.⁸ The rollout replaced the previous generation: o3, o4-mini, and o4-mini-high went to ChatGPT Plus, Pro, and Team users starting April 2025, replacing o1, o3-mini, and o3-mini-high.⁸

By the numbers

Valuation and funding. In October 2025 current and former employees sold $6.6 billion in shares to investors including Thrive Capital, Dragoneer Investment Group, T. Rowe Price, SoftBank, and the United Arab Emirates' MGX, pushing the privately held company's valuation to $500 billion.³

Revenue and losses. The two available measures of OpenAI's 2025 revenue differ, and the difference is largely one of accounting. Audited figures show $13.07 billion in revenue, $34 billion in costs and expenses, a $20.92 billion operating loss, and a net loss attributable to the company of $38.53 billion; the restructuring itself triggered a $41.55 billion loss from changes in the fair value of convertible equity instruments and warrant liabilities, and total 2025 net loss including noncontrolling interests was $60.35 billion before attribution adjustments.⁴ Separately, OpenAI confirmed an annualized revenue run rate of approximately $50 billion at the end of September, about $20 billion below widely reported figures, with run-rate revenue growing 70 percent in the third quarter.⁵ CFO Sarah Friar described revenue growing 3 times every year and 10 times from 2023: $2 billion annual recurring revenue in 2023, $6 billion in 2024, and more than $20 billion in 2025.⁶

Headcount. In March 2026 The Information reported that OpenAI planned to grow from roughly 4,500 employees to around 8,000 by year-end.¹⁵

Stargate and the compute buildout

In early 2025 OpenAI announced Stargate, a joint venture with Oracle and SoftBank aiming to invest $500 billion in AI data centers in the United States. The initial goal was 20 data centers, with the first project, at Abilene, Texas, already operational.¹⁷ On September 23, 2025 the three partners announced five new US sites, selected from more than 300 proposals across more than 30 states, bringing planned capacity to nearly 7 GW and over $400 billion in investment.¹⁸

A shift toward leasing. After partner disputes, OpenAI has effectively moved away from first-party ownership of Stargate data centers, preferring to lease third-party capacity; SoftBank agreed to own and develop the Texas data center while OpenAI designs and operates it on a long-term lease.¹⁷ Rather than building its own infrastructure, the company reached capacity deals with AWS, Google Cloud, AMD, and Cerebras, and its anticipated compute outlays through 2030 amount to about $600 billion—a steep reduction from a prior commitment of $1.4 trillion by 2033.⁷ OpenAI's computing capacity more than tripled in 2025 to roughly 1.9 GW, and CEO Sam Altman has set a target of eventually building a gigawatt of new AI infrastructure per week.⁷

Safety and evaluations

OpenAI's Preparedness Framework, first published in beta in December 2023, is at version 2 dated April 15, 2025. It defines two capability thresholds per tracked category: at High capability a system must have safeguards that sufficiently minimize the associated risk before deployment, plus security controls; at Critical, safeguards are required during development itself.⁹ Under this framework, o3 and o4-mini were evaluated on biological and chemical capability, cybersecurity, and AI self-improvement, and the company determined both remain below the High threshold in all three categories.⁸

Independent assessment is more mixed. The Future of Life Institute's Summer 2025 AI Safety Index found that OpenAI stood alone in making its whistleblowing policy public, described a stronger method for managing risk within its safety framework, and evaluated risks before mitigation was applied; nevertheless, Anthropic once more received the top overall grade, ranking first in five of the six domains.¹⁹ The December 2025 index places Anthropic, OpenAI, and Google DeepMind at the front of the field on governance and accountability.²⁰

The 2023 board crisis

At about noon on November 17, 2023, CEO Sam Altman was fired by the three independent directors of OpenAI, an event that provoked new questions about the people and companies behind the AI revolution.²¹ Greg Brockman was also removed from the board.²² According to accounts of the five-day turmoil, OpenAI was split between people frightened by what AI might do and colleagues who regarded it as a rare, career-defining chance for wealth and status; as fault lines widened, the company's top figures traded barbs and attacked each other.²³

The review and its aftermath. The law firm WilmerHale reviewed more than 30,000 documents and conducted dozens of interviews, including of members of OpenAI's prior board, regarding the November 17 removals and Altman's termination as CEO.²² The conclusion, more than three months after the event, was characterized as a resounding victory for Altman as he moved to reassert control of the company; he rejoined the board in March 2024.²⁴

Controversies and litigation

The New York Times' copyright case against OpenAI and Microsoft remained ongoing, with summary-judgment briefs filed in September 2026.¹⁵ The crisis-era reporting also documented the internal governance and safety-culture tensions described above, in which leaders openly disagreed about whether the technology was primarily a risk or an opportunity.²³

How it compares with other frontier labs

On safety frameworks, OpenAI's Preparedness Framework centers internal governance through its Safety Advisory Group, whereas Anthropic's Responsible Scaling Policy, at version 3.0 effective February 24, 2026, commits to external review of its Risk Reports, and Google DeepMind's Frontier Safety Framework is at version 3 (September 22, 2025).⁹ A comparative table of safeguards shows OpenAI with SOC 2 Type II attestation and a Preparedness team, Google DeepMind with AI red-teaming and an AGI Safety Council, Anthropic with a PBC plus a Long-Term Benefit Trust that appoints its directors, a Responsible Scaling Policy, and ISO/IEC 42001 certification, and Meta relying on open-sourcing for transparency.¹³ OpenAI and Anthropic both operate as public benefit corporations that balance mission and commercial success, and OpenAI is the only company among those assessed with a public-facing whistleblowing policy.²⁰

Revenue comparisons across labs carry an accounting caveat: OpenAI counts its share of certain revenue from cloud-partner deals, while Anthropic includes gross sales through partners, so headline run-rate figures are not measured on the same basis.⁵

The Microsoft relationship

Microsoft's involvement began with the $1 billion investment in the capped-profit subsidiary in 2019.² Under a renegotiated partnership finalized in May 2026, OpenAI's total revenue-share payments to Microsoft are capped at $38 billion through 2030, a reduction of approximately $97 billion from the prior projected trajectory of about $135 billion.¹⁶ After the October 2025 recapitalization, Microsoft held about 27 percent of OpenAI Group PBC's shares.¹⁵

References

Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › AI companies, people, and products › Frontier AI labs and companies

Initially written Sep 17, 2026 · Reviewed: Sep 30, 2026; Oct 15, 2026 · Edited: Sep 30, 2026; Oct 15, 2026 · Last review: Oct 15, 2026

Notice something wrong?