In brief
OpenAI was born in 2015 with an explicitly altruistic mission: to develop safe artificial general intelligence (AGI) for the benefit of humanity. Ten years later, the company is valued at over 700 billion dollars and generates 25 billion in annualized revenue. This trajectory — from non-profit to commercial empire — concentrates most of the tensions running through the AI industry: safety versus speed, mission versus profit, governance versus growth.
ChatGPT, launched in November 2022, crystallized public attention on LLMs. Since then, OpenAI has maintained a leading position in the AI assistant market, with 910 million weekly active users in February 2026, facing intensifying competition (Anthropic, Google DeepMind, Meta AI).
In short: OpenAI is the story of an organization that started as an NGO and transformed into one of the most highly valued tech companies — without officially abandoning its original mission. Imagine a medical research foundation that accidentally discovers a $100-billion drug and then has to navigate between selling the drug (financing R&D and costs), guaranteeing access (mission), and structuring governance (preventing a private investor from taking control). The internally documented tensions — founder departures, the 2023 board crisis, the 2025 restructuring — all stem from this impossible equation.
Identity card
| Criterion | Value |
|---|---|
| Founded | December 2015 |
| Headquarters | San Francisco, California |
| Headcount | ~3,000 people (2025) |
| Status | OpenAI Group PBC (for-profit) + OpenAI Foundation (non-profit, ~25% of shares) |
| Key founders | Sam Altman, Elon Musk (left in 2018), Ilya Sutskever, Greg Brockman, John Schulman |
| Valuation | ~$730 billion USD (pre-money, February 2026) |
History
2015 — The non-profit promise
OpenAI is founded in December 2015 with a starting endowment of one billion dollars, contributed by Elon Musk, Sam Altman, Peter Thiel, and other investors. The non-profit structure is deliberate: no shareholder should control the organization, which is meant to align decisions with the mission rather than financial returns.
The founding team includes leading researchers: Ilya Sutskever (formerly of Google Brain, student of Geoffrey Hinton), John Schulman (reinforcement learning specialist), Greg Brockman (former CTO of Stripe). The promise: publish research, make tools accessible, and not seek to dominate the market.
2018 — Elon Musk’s departure
In February 2018, Musk leaves the board of directors. The official version cites a conflict of interest with Tesla and its autonomous driving program. The version later revealed by OpenAI is more pointed: Musk wanted to take executive control of the company, or even merge it with Tesla. Faced with the co-founders’ refusal, he departed declaring that OpenAI had no chance of succeeding and announcing the construction of a competing AGI.
2019 — The capped-profit transition
To raise the capital needed to train increasingly expensive models, OpenAI created a commercial entity with “capped profits” (100 times the initial investment) in 2019. The non-profit structure remains a shareholder of the commercial entity. It is within this framework that Microsoft enters as a strategic partner and investor.
2022-2023 — ChatGPT and the board crisis
The launch of ChatGPT in November 2022 is a turning point: 100 million users in two months, a record for a consumer application. Commercial pressure accelerates.
On November 17, 2023, the board of directors dismisses Sam Altman. Official reason: lack of transparency in his communications with the board. Reported elements: Altman had not informed the board of ChatGPT’s launch, had not disclosed his participation in a fund linked to OpenAI, and two executives had testified to behaviors characterized as psychological abuse.
The reaction is immediate: nearly 800 employees — almost the entire workforce — sign a letter threatening to join Microsoft if Altman is not reinstated. On November 22, 2023, Altman returns as CEO. The board is reconfigured: Bret Taylor (chairman), Larry Summers, Adam D’Angelo (the sole retained member). The crisis de facto strengthened Altman and weakened the non-profit governance.
In short: the November 2023 crisis lasted five days and almost cost OpenAI its survival. The board could have kept Altman out and chosen a new CEO; the actual effect was the opposite. 95% of the 800 employees threatened to leave for Microsoft, where Satya Nadella publicly pledged to hire them. Faced with this force, the board acknowledged that the non-profit mission could not survive without the engineers producing the models. The implicit lesson: in the AGI race, real power lies not with the board but with the technical teams who control the models.
2025 — The recapitalization
In December 2024, OpenAI announces its conversion to a for-profit Public Benefit Corporation (PBC), with the non-profit relegated to minority shareholder status. The announcement triggers broad opposition: former employees, safety researchers, attorneys general of California and Delaware, Elon Musk (who filed a lawsuit in February 2024 arguing that his donation was conditional on maintaining non-profit status).
In May 2025, OpenAI partially reverses course. In October 2025, the final restructuring is completed: OpenAI Foundation (non-profit) holds approximately one quarter of the shares of OpenAI Group PBC (the operational commercial entity). Critics note that the real control mechanisms remain opaque.
Models and products
The GPT series
The GPT (Generative Pre-trained Transformer) series constitutes OpenAI’s technological foundation. GPT-1 (2018, 117 million parameters) lays the groundwork. GPT-2 (2019, 1.5 billion parameters) is the first to spark debates about risks: OpenAI then refuses to publish the full weights. GPT-3 (2020, 175 billion parameters) marks a qualitative breakthrough and is only accessible via API — the open-source promise fades. GPT-3.5 (2022) introduces instruction fine-tuning (RLHF), and becomes the foundation of ChatGPT.
GPT-4 (March 2023) is multimodal, capable of processing text and images. GPT-4o (May 2024) unifies text, audio, and image in real time, faster and cheaper. GPT-5 (August 2025) then GPT-5.2 (December 2025, integrated into Microsoft 365 Copilot) and GPT-5.4 (March 2026, with mini and nano variants) continue this trajectory.
The pace accelerates sharply in 2026. GPT-5.5 arrives in the spring, then the GPT-5.6 series enters limited preview on 26 June 2026 with three models that are named rather than numbered: Sol (frontier), Terra (balanced), Luna (fast and affordable). On 3 September 2026, OpenAI announces GPT-6 Astra, first as a restricted preview for trusted partners, then opened to paying users the following day in a limited version that refuses certain cybersecurity prompts. Announced API pricing: $10 per million input tokens, $50 per million output tokens.
Two elements are worth noting, because they are rare in a laboratory’s communication. First, OpenAI states that it pretrained on more than 100,000 GPUs for the first time, at its Stargate site in Texas. Second, the company says it delayed the model’s release to add safeguards, following an incident in July 2026. On the ExploitBench benchmark, Astra is reported at 100% against 78.5% for GPT-5.6 Sol — a vendor-published figure with no independent replication to date. [UNVERIFIED]
In short: between March and September 2026, OpenAI shipped four generations — 5.4, 5.5, 5.6, then 6. At that pace, any product page naming “the current flagship model” is wrong within a quarter. This is why this page dates its claims rather than claiming permanent currency.
The o series (reasoning)
Launched in September 2024, the o series introduces a different paradigm: the model “thinks” before responding via an internal chain of thought. o1-preview (September 2024), full o1 (December 2024), o3-mini (January 2025, available to free users), o3 and o4-mini (April 2025, top performance in mathematics and science), o3-pro (June 2025). These models perform particularly well on structured tasks requiring multi-step reasoning.
ChatGPT and the API
ChatGPT is the primary interface: free plan, Plus ($20 USD/month), Pro ($200 USD/month), Team (~$25-30 USD/user), Enterprise (custom pricing). The API gives developers access to the full range of models. In 2025, OpenAI also releases two open-weight models (120b and 20b parameters) for on-premise deployment.
Positioning
Commercial leader under pressure
The figures are considerable: 910 million weekly active users in February 2026, 9 million paying businesses, 25 billion in annualized revenue. The relationship with Microsoft is structural — Microsoft has invested a total of 13 billion dollars, holds approximately 27% of the commercial entity, and generated 7.6 billion in OpenAI-related revenue in the fourth quarter of 2025 alone.
This dominant position has a cost: OpenAI burns capital at a sustained pace (infrastructure costs, training, salaries). The successive valuations — 157 billion in October 2024, 300 billion in March 2025, 730 billion in February 2026 — rest on growth projections that are difficult to audit.
In short: OpenAI’s dominant position rests on three ratios. Adoption: 910 million weekly active users — about one in eight people on Earth, more than active Twitter users. Capital: $730 billion valuation = roughly 30× annualized revenue (Apple’s ratio is ~7×, Microsoft’s ~13×). This massive premium assumes sustained growth above 50% per year over the next five years. Microsoft at 27% of equity and $7.6 billion in quarterly OpenAI-related revenue: the entanglement is such that OpenAI and Microsoft can no longer separate without major mutual damage.
Strengths
- Lead in consumer product: ChatGPT remains the reference in awareness and adoption.
- Depth of the lineup: from lightweight models to advanced reasoning models.
- Ecosystem: API, SDK, GPT Store, Microsoft integrations.
- Exceptional fundraising capacity, financing intense R&D.
Limits and tensions
- Weakened governance: the November 2023 crisis and the 2025 restructuring showed that the non-profit mission can yield under commercial pressure.
- Progressive closure: models are no longer open-source, weights are not published, transparency on architectures is limited.
- Growing competition: Anthropic (Claude), Google (Gemini), Meta (Llama, open-weight), xAI (Grok) are narrowing the technological lead.
- Departures of founders: Ilya Sutskever (left to create Safe Superintelligence in 2024), John Schulman (left for Anthropic), Greg Brockman (sabbatical in 2024). These departures weaken the historical research culture.
- The removal of the word “safely” from the mission statement was noted by AI safety researchers as a signal of shifting priorities.
Key takeaways
- OpenAI transformed public perception of LLMs with ChatGPT, and has established itself as the commercial reference in the sector.
- Its trajectory illustrates a structural tension: the resources needed to build frontier models imply a profit logic that enters into friction with the founding safety mission.
- Governance remains the critical point: the November 2023 crisis showed that the non-profit board could be neutralized by pressure from employees and investors.
- On the technical side, the o series marks a significant methodological evolution, distinct from simply scaling up the GPT models.