OpenAI

Overview

OpenAI is an American artificial intelligence research company headquartered in San Francisco, California. Founded in December 2015, OpenAI develops and deploys large language models, image generation systems, and multimodal AI products that serve hundreds of millions of users globally. Its flagship product, ChatGPT, launched in November 2022 and became the fastest consumer application to reach 100 million users in history, doing so in two months.

OpenAI operates under a "capped-profit" structure, meaning investors receive returns up to a defined multiple of their investment before surplus profits flow to the company's nonprofit parent entity. In a major structural shift, OpenAI announced in late 2024 that it would convert to a Delaware-incorporated for-profit public benefit corporation (PBC), separating the operational company from its original nonprofit governance layer. The nonprofit retains a significant equity stake in the new structure.

As of May 2026, OpenAI is valued at approximately $300 billion following a secondary share sale, making it one of the most valuable private technology companies in history. The company employs over 3,500 people and processes billions of API requests daily across its consumer and enterprise product lines. Microsoft has invested more than $13 billion in OpenAI and integrated its models across Office 365, Azure, GitHub Copilot, and Bing.

History and Founding

OpenAI was founded in December 2015 by a group that included Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba, and John Schulman. The founding thesis was that advanced AI posed an existential risk to humanity if developed solely by profit-driven corporations, and that a nonprofit research lab with an open-publication mandate could serve as a counterweight to closed commercial development at Google and Facebook.

The early years were marked by foundational research in reinforcement learning and language modeling. OpenAI's GPT-1 (2018) and GPT-2 (2019) demonstrated that scale in unsupervised pretraining produced qualitatively new language capabilities. GPT-2's release was deliberately staged over several months over concerns that the model was capable enough to be misused for disinformation at scale, a decision that generated significant debate in the AI research community about when capability disclosure obligations override open science norms.

Elon Musk departed the OpenAI board in 2018, citing a conflict of interest with Tesla's AI programs. In 2024, he filed a lawsuit alleging that OpenAI had violated the terms of its founding mission by pursuing profit and forming a close commercial relationship with Microsoft. The lawsuit was dismissed in 2025 after Musk filed a competing AI company, xAI, and the courts found that Musk's standing to challenge the nonprofit's governance was limited.

The November 2023 board crisis remains the most consequential internal event in OpenAI's history. The board fired CEO Sam Altman on November 17, 2023, citing a lack of candor with the board, without specifying the nature of the concern. Within 96 hours, more than 700 of OpenAI's approximately 770 employees signed an open letter threatening to resign and join Altman at Microsoft if he was not reinstated. Altman returned as CEO on November 21, 2023. Three of the five board members who voted to remove him departed in the aftermath.

Core Products | ChatGPT, Sora, and the API Platform

ChatGPT is OpenAI's primary consumer interface, a conversational AI assistant available via web, iOS, and Android. It uses the latest available GPT or o-series model depending on subscription tier. The free tier provides access to GPT-4o mini; ChatGPT Plus ($20/month) provides access to GPT-4o, the o3 and o4-mini reasoning models, image generation via DALL-E 3, and priority access during peak demand. ChatGPT Team and Enterprise tiers provide higher usage limits, admin controls, and data privacy guarantees. As of early 2026, ChatGPT has over 400 million weekly active users.

Sora is OpenAI's text-to-video generation system, announced in February 2024 and released publicly in December 2024. Sora generates video up to one minute in length from text prompts, using a diffusion transformer architecture trained on a large proprietary dataset of video content. It is available to ChatGPT Plus and Pro subscribers. Professional film and advertising productions began integrating Sora into pre-visualization workflows in 2025, and the system has attracted significant attention from the entertainment industry as a cost reduction tool for concept visualization. For the latest on Sora and AI video, see the OpenAI energy infrastructure coverage.

The OpenAI API is the developer-facing product that generates the majority of OpenAI's commercial revenue. It provides programmatic access to GPT-4o, o3, o4-mini, embeddings models, DALL-E, Whisper (speech recognition), and TTS (text-to-speech). Enterprise customers building AI products on the API include Microsoft, Salesforce, Adobe, Morgan Stanley, and hundreds of thousands of smaller software companies. The API also supports fine-tuning, function calling, structured output, and the Assistants API, which allows developers to build persistent AI agents with tool use, code execution, and file retrieval capabilities.

ChatGPT in CarPlay launched in May 2026, making OpenAI the first AI assistant to reach Apple's in-car platform following Apple's decision to open CarPlay to third-party AI chatbots. Full coverage at ChatGPT CarPlay launch.

GPT Models | Architecture, Scale, and the o-Series

The Generative Pre-trained Transformer (GPT) model family is the technical foundation of OpenAI's products. Each generation represented a step-change in capability driven by increases in model parameters, training compute, and data quality. GPT-3 (2020) introduced the concept of few-shot learning at scale, demonstrating that a sufficiently large language model could perform novel tasks from a small number of examples without task-specific fine-tuning. This property, now called in-context learning, fundamentally altered the trajectory of AI application development.

GPT-4, released in March 2023, was OpenAI's first publicly acknowledged multimodal model, capable of processing both text and image inputs. GPT-4 scored in the 90th percentile on the Uniform Bar Exam, the 99th percentile on the GRE verbal reasoning section, and passed the USMLE medical licensing exam. GPT-4o, released in May 2024, unified text, audio, and vision processing in a single model architecture with dramatically reduced latency, enabling real-time voice conversation with emotional tone detection.

The o-series models represent a distinct product line from the GPT series, optimized for complex multi-step reasoning through extended chain-of-thought processing. o1 launched in September 2024, followed by o3 and o4-mini in April 2025. The o-series models scored higher than GPT-4o on mathematical competition benchmarks, PhD-level science questions, and software engineering tasks but generate responses more slowly and at higher computational cost. OpenAI positions the o-series for tasks requiring deliberate reasoning, such as scientific analysis, complex coding, and legal document review, and GPT-4o for tasks requiring speed and multimodal fluency.

OpenAI launched Symphony, its open-source autonomous coding agent framework, in early 2026. The release marked a significant shift in OpenAI's historically closed-model posture. Full coverage at Symphony open-source coding agents.

Safety and Governance | Alignment Research and Policy

OpenAI was explicitly founded on the premise that advanced AI poses existential and near-term societal risks if developed without adequate safety research. The company maintains a dedicated Safety team working on alignment, interpretability, and policy. However, the relationship between safety priorities and commercial deployment timelines has been a persistent source of internal tension, most acutely visible during the November 2023 board crisis, which multiple former employees characterized as partly rooted in disagreements about deployment pace.

The company's safety framework centers on Reinforcement Learning from Human Feedback (RLHF), the technique that produces models that follow instructions and decline harmful requests. RLHF was the core technical contribution that differentiated ChatGPT from the raw GPT-3 base model and enabled its consumer deployment. OpenAI also conducts pre-deployment red-teaming, in which internal and external researchers attempt to elicit harmful outputs before a model is released.

OpenAI launched a Safety Fellowship program in 2026 following a New Yorker investigation that documented concerns raised by former safety team members about the company's risk evaluation processes. The fellowship funds independent AI safety researchers. Full coverage at OpenAI Safety Fellowship launch.

The most serious governance incident of 2026 involved ChatGPT's role in the Tumbler Ridge shooting case. OpenAI's automated systems had flagged and banned the shooter's account eight months before the attack, but company leadership determined the activity did not meet the threshold for a law enforcement referral. Sam Altman issued a public apology and the case became the center of a negligence lawsuit and a Florida criminal investigation. Full coverage at Altman apology, Tumbler Ridge.

Business Model | Revenue, Microsoft, and Enterprise

OpenAI's primary revenue streams are ChatGPT subscriptions and API usage fees. The company reported annualized revenue of approximately $3.4 billion in early 2024 and has not disclosed more recent figures publicly, though analyst estimates based on disclosed user growth and pricing suggest annualized revenue exceeded $7 billion by late 2025. OpenAI is not profitable; compute costs associated with training and running large-scale models at consumer and enterprise volumes are substantial, and the company has continued to raise capital to fund both infrastructure and research.

The Microsoft partnership, formalized in a multi-year, multi-billion dollar investment deal in 2019 and expanded significantly in January 2023, is the structural pillar of OpenAI's commercial infrastructure. Microsoft provides exclusive Azure cloud compute for OpenAI's training and inference workloads in exchange for preferred licensing rights. Microsoft deploys OpenAI models across GitHub Copilot, Microsoft 365 Copilot, Azure OpenAI Service, and the AI-enhanced Bing search product. This partnership gives OpenAI distribution at enterprise scale while giving Microsoft access to frontier models it could not build independently.

OpenAI signed a landmark deal in March 2026 to provide AI services to the Pentagon on AWS infrastructure, the company's first major U.S. government contract. The deal signaled OpenAI's willingness to serve national security applications, a position that generated internal debate and external scrutiny. Coverage at OpenAI AWS Pentagon classified AI deal and the subsequent governance debate at Altman, government oversight, and the Pentagon deal.

Brand apps inside ChatGPT, including integrations from Sephora and Realtor.com, launched in May 2026 as OpenAI expanded its in-platform commerce layer. Coverage at ChatGPT brand apps.

Competition | Google Gemini, Anthropic Claude, and Meta AI

OpenAI operates in a competitive landscape that has become significantly more crowded since ChatGPT's launch in late 2022. Its primary competitors are Google DeepMind (Gemini), Anthropic (Claude), Meta AI (Llama), and xAI (Grok). Each competitor has distinct structural advantages: Google has search distribution and proprietary data at scale; Anthropic has a research team that includes many of OpenAI's former alignment researchers; Meta distributes Llama as open weights, capturing developer adoption; xAI benefits from Elon Musk's public platform and distribution through X.

Google's Gemini family, particularly Gemini 1.5 Pro and Gemini 2.0, closed the benchmark gap with GPT-4o significantly in 2024 and 2025. Gemini's integration into Google Search, Gmail, Docs, and Android gives it a distribution advantage that no standalone AI company can replicate. The competitive tension between Google and OpenAI is particularly acute in the enterprise market, where customers must choose between Google's vertically integrated AI stack and OpenAI's API with Microsoft's Azure infrastructure layer.

Anthropic's Claude 3.5 Sonnet became the preferred model for software engineering tasks among professional developers in late 2024, temporarily displacing GPT-4o in coding benchmarks and user preference surveys. Anthropic has positioned Claude as the safety-first alternative to OpenAI, a differentiation that carries credibility given Anthropic was founded by former OpenAI safety researchers including Dario Amodei and Daniela Amodei.

Meta's Llama series represents a structurally different competitive threat: open-weight models that any developer, enterprise, or nation-state can download and run without API fees. Llama 3.1 405B, released in July 2024, approached GPT-4 performance at zero marginal API cost, challenging OpenAI's pricing model for latency-tolerant, high-volume enterprise use cases. For the latest AI and tech coverage, see the OzoneNews Tech desk.

OpenAI in 2026 | Platform Expansion and Key Developments

OpenAI's 2026 trajectory has been defined by three parallel pressures: product expansion into new surfaces (CarPlay, brand apps, government), governance pressure from the Tumbler Ridge case and safety fellowship scrutiny, and infrastructure scaling driven by compute demand that now exceeds what even the Microsoft Azure partnership can absorb at projected growth rates.

The company's Stargate initiative, announced in January 2026 as a joint venture with SoftBank and Oracle to deploy $500 billion in AI infrastructure over four years, is the clearest signal of the scale OpenAI believes will be required to train and run next-generation models. The first Stargate data centers are under construction in Texas. OpenAI's energy infrastructure commitments, including a deal with Helion Energy, are covered at OpenAI and Helion Energy.

On the model side, OpenAI has not publicly announced a "GPT-5" release date, but the o3 and o4-mini reasoning models released in April 2025 demonstrated that the o-series roadmap is the near-term frontier priority. Internal communications leaked in early 2026 suggested that OpenAI's next flagship model may unify the GPT and o-series architectures, producing a single model capable of both fast conversational responses and extended reasoning on demand.

Sam Altman's public profile has continued to expand beyond the company itself, including a podcast circuit focused on AI policy and a vocal position on the need for government oversight frameworks. His stated view that AI development requires international coordination and domestic regulatory structure, combined with the Pentagon deal, positions OpenAI as the most government-engaged major AI lab. Coverage at Altman and government oversight.

Follow all OpenAI and AI industry developments at the OzoneNews Tech desk and breaking AI news at OzoneNews.