OpenAI: CEO History

That idealism would bend under the weight of economic reality. Training frontier AI models requires computational resources measured in the hundreds of millions of dollars per run. Its flagship product, ChatGPT, commands more than 300 million weekly active users as of early 2025. The free tier of ChatGPT, which offers access to GPT-4o mini and limited usage of GPT-4o, serves as the top of a carefully engineered conversion funnel. ChatGPT Plus, priced at $20 per month, unlocks priority access to the most capable models, image generation via DALL-E 3, web browsing, the ability to create and use custom GPTs, and — as of 2024 — access to memory features and voice capabilities. As of mid-2024, GPT-4o input tokens were priced at $5 per million and output tokens at $15 per million, while the more economical GPT-4o mini cost $0.15 per million input tokens and $0.60 per million output tokens. By early 2025, OpenAI claimed more than 92% of Fortune 500 companies were using its products in some form, though the depth of those engagements varied enormously from enterprise contracts to departmental API usage. OpenAI's Operator capability — announced in late 2024 — allows GPT-4o to take actions in web browsers autonomously, completing tasks like booking travel, filling forms, and managing software interfaces without human intervention. This positions OpenAI to capture transaction-layer economics rather than purely information-layer value. Gemini Ultra 1.0 reportedly outperformed GPT-4 on the MMLU benchmark across 57 academic subjects. However, Anthropic lacks OpenAI's consumer brand, its ChatGPT subscriber base, and the breadth of product surface area that allows OpenAI to capture multiple revenue streams simultaneously. Llama 3.1 405B, released in July 2024, was competitive with GPT-4 on several tasks and could be downloaded and run by any organization with sufficient GPU resources — at zero licensing cost. For OpenAI, the Llama series represents a price floor compression on API revenue; as open-weight models improve, price-sensitive API customers may migrate to self-hosted alternatives. While Stargate provides a path to the compute sovereignty OpenAI needs, it also represents a staggering capital commitment in a sector where the return timeline remains uncertain. Every conversation — corrected, upvoted, flagged, or refined — becomes training signal for subsequent model generations. The consumer flywheel is the first track. The nonprofit conversion faces scrutiny from California Attorney General Rob Bonta and Delaware courts examining whether existing investors are being treated equitably, a process that could take one to two years to resolve. The most strategically defining near-term product direction is AI agents: software that takes autonomous multi-step actions rather than generating single responses. If AGI were to emerge within a corporate context optimized for shareholder returns, who would ensure it was developed safely? The answer they arrived at was a nonprofit research laboratory with an open publication policy. The nonprofit structure would, in theory, ensure that decisions were made in the service of the mission rather than quarterly earnings. Sam Altman and Elon Musk served as co-chairs of the board. The early research agenda was ambitious and deliberately broad. OpenAI's founding team pursued work on reinforcement learning, robotics, natural language processing, and game-playing agents simultaneously, reflecting a conviction that AGI would likely emerge from the convergence of multiple models rather than any single architecture. By 2018, OpenAI Five, an enhanced version of the system, defeated professional human Dota 2 teams in exhibition matches watched by millions online. The research team also published the first version of the Generative Pre-trained Transformer — GPT-1 — in 2018, a language model trained on the BooksCorpus dataset of approximately 7,000 unpublished books. GPT-1 was not itself a commercial product; it was a research paper demonstrating that unsupervised pre-training on large text corpora could produce language representations transferable to downstream tasks. But it planted the seed for every commercial product that would follow. When that proposal was declined, and as Tesla's own AI efforts around autonomous driving created potential conflicts of interest, Musk resigned from the OpenAI board in February 2018. He would later claim in legal filings that he departed because he disagreed with the decision to pursue the capped-profit restructuring, and that he had been promised a different governance outcome. OpenAI disputes this characterization. The acrimony between Musk and OpenAI — particularly Altman — would become one of the defining interpersonal dramas of the AI industry. The decision was controversial internally and externally, with critics arguing it fundamentally compromised the organization's founding mission. The tension between these two positions has never fully resolved and remains the central fault line in OpenAI's institutional identity.