Sam Altman, Open AI’s CEO has laid a roadmap to build a fully autonomous AI researcher by March 2028, capable of doing independent scientific work without human direction.
He revealed this during a livestream press event where Altman and OpenAI’s leadership described this as a concrete, measurable goal, rather than an abstract philosophical claim around artificial general intelligence (AGI).
The company expects to deliver an ‘intern-level research assistant’ by September 2026 which will act as a stepping stone toward the full researcher.
Fully autonomous researcher by 2028 target is for a system that can independently carry out nontrivial research which include formulating problems, designing experiments, interpreting results, and iterating essentially functioning at the level of human researchers.
The Intern-level assistant by 2026 is meant to handle more structured research tasks under supervision, such as literature reviews, hypothesis generation, and basic experiment design.
Read also: OpenAI restructures into For-Profit Entity under new foundation
Scaling ‘test-time compute’ is a key part of the plan which is giving models more computational resources when reasoning, allowing them to think longer per problem.
Altman and the team said algorithmic innovations plus sheer compute scaling are essential to hit their goals.
“Our aspiration is that we can build an infrastructure factory where we can create one gigawatt a week of compute,” he said.
After a string of deals, OpenAI has current commitments for $1.4 trillion in spending on AI infrastructure for roughly 30 gigawatts of new computers in the coming years. Altman noted that he wants more.
He stated that if OpenAI’s research continues to advance alongside sustained consumer demand, he would like OpenAI to build infrastructure that would produce 1 gigawatt of compute a week.
Altman said, “Our aspiration is that we can build an infrastructure factory where we can create one gigawatt a week of compute.
“We would like to get that cost down significantly to like $20 billion over that five-year lifecycle of that equipment.”
