OpenAI began in 2015 with a research question: how could increasingly capable artificial intelligence be developed so that its benefits were shared as widely as possible?

For its first years, that largely meant models, research papers and the computing systems required to train them. But as the models became more capable, the organisation increasingly moved beyond research alone. In 2019, it created a for-profit subsidiary to raise the capital required to scale its research and deployment – an early acknowledgement that advances in AI would depend as much on infrastructure and distribution as algorithms.

ChatGPT changed the scale of that challenge.

Released as a research preview in November 2022, a conversational interface turned frontier AI into something millions of people could immediately experiment with themselves. By 2026, OpenAI reported more than 900 million weekly ChatGPT users, while its products had expanded into coding, search, enterprise work, scientific research and tools capable of acting across apps and workflows.

That growth has pulled OpenAI into increasingly unfamiliar territory.

Stargate is building AI compute infrastructure at gigawatt scale in the US and internationally. OpenAI is designing its own accelerator chips with Broadcom. And after the io team founded by Jony Ive joined OpenAI in 2025, hardware and industrial design became part of the company too – with the aim of developing new kinds of AI-native products rather than simply putting more intelligence inside existing interfaces. OpenAI
OpenAI now describes that approach as a full stack: infrastructure and chips supporting frontier models, models powering products and developer platforms, and those products creating new ways for people and organisations to use intelligence.

The research lab did not disappear.

It kept following the implications of its research until the work extended far beyond the lab.