- OpenAI draw up a plan for a researcher of IA self-employed, with a training version in 2026 and initial intern-type functions before then.
- The organization becomes a public benefit corporation; the OpenAI Foundation retains 26% and manages $25.000 billion.
- Technical strategy based on algorithmic innovation and more 'test time compute'; models now reason for up to five hours.
- A commitment of 30 GW of capacity and $1,4 trillion in infrastructure, with ethical and regulatory debate on the European horizon.
OpenAI has presented a roadmap that aims to bring the Artificial Intelligence going beyond simple assistance and bringing it closer to the realm of autonomous research. During a live broadcast, Sam Altman announced that the company wants to have an assistant with intern-level skills in the short term and, later, a system capable of acting as fully autonomous AI researcher in complex projects.
The announcement was accompanied by a momentous change in its structure: the firm is definitively abandoning its original non-profit model to operate as a public benefit corporation. This transition to a public benefit corporation It seeks to maintain the pace of investment and infrastructure required by its scientific plan, without abandoning a branch focused on security and the advancement of research.
A hybrid model to finance the next scientific wave
OpenAI has established a dual-nature system: on the one hand, a for-profit entity to raise capital on a large scale, and on the other, a foundation that preserves its scientific mission and security. The OpenAI Foundation will retain 26% of the organization and will manage a $25.000 billion fund —about €21.500 billion— intended to apply AI to cure diseases and to research initiatives and responsible governance.
In addition, the tech company announced an update to its agreement with Microsoft and outlined its infrastructure plans in a commitment to 30 gigawatts of computing capacityThis is associated with a financial obligation estimated at $1,4 trillion. The goal is to have the necessary energy and computing infrastructure to support future scientific advancements.
What does OpenAI mean by an AI 'researcher'
Jakub Pachocki, the company's chief scientist, described that milestone as a system capable of plan, execute and deliver results of research without direct supervision. This is not a metaphor, he stated, but a concrete goal supported by the trajectory of current deep learning models.
According to Pachocki, the industry could be less than a decade away from achieving superintelligence, understood as the ability to outperform humans in a wide range of critical tasks. To get there, the plan rests on two pillars: sustained algorithmic innovation and a dramatic expansion of what is known as 'test time compute', this is, There of reflection that is given to the model before responding.
In practice, models can already address tasks with reasoning horizons of up to five hours and compete at the level of the best humans in highly demanding tests, such as the International Mathematical Olympiad. OpenAI's goal is to enable them to think more and better, even if that means dedicating enormous computing resources to a single scientific problem.
Computing, algorithms, and dedicated data centers
The company assumes that the qualitative leap will not come solely from larger models, but from giving them more time and resources to deliberate. Under this logic, it might be reasonable to allocate the capacity of complete data centers to a single scientific question, provided that the return in knowledge justifies it.
- More 'test time compute' to increase the depth of reasoning.
- Continuous algorithmic improvements to squeeze the same capacity with greater efficiency.
- Energy and infrastructure hardware at scale (30 GW committed).
This approach, combined with industrial alliances and a structure that facilitates capital raising, seeks to accelerate discoveries in areas such as physics, medicine, and engineering, where intensive computational exploration can open up unprecedented avenues.
Calendar, potential uses and safeguards
OpenAI plans a practice version of the assistant for 2026 SeptemberThe company is testing the system on real-world problems and plans to have it fully operational two years later. The company emphasizes that this progress will be accompanied by ethical controls, audits, and transparency in its development.
The leap to an autonomous 'researcher' raises profound questions: who holds the intellectual property What about the findings generated by AI? How is the reasoning of opaque systems audited? What happens if their capacity to generate knowledge exceeds our understanding? These questions are already being debated in Europe, where frameworks such as the future AI Act seek to guarantee security, traceability, and accountability.
Altman framed this entire movement as part of an accelerated research ecosystem, where AI not only assists but also proposes hypotheses, designs experiments, and produces new knowledge. The stated commitment is to maintain a responsible governance as capabilities and infrastructure scale.
The roadmap positions OpenAI in a race to transform AI into an agent of scientific discovery: with a trial version in 2026, a restructuring that unlocks capital raising, 30 GW of computing power, and an emphasis on test-time compute, the company aims for an autonomous system to act as Researcher in 2028, opening up opportunities and forcing a redoubling of regulatory and ethical scrutiny, also in Europe.
Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.