Sam Altman-backed fusion startup Helion in talks with OpenAI

In a world dominated by the race for increasingly powerful language models, the true barrier to the development of artificial intelligence is shifting from the software layer to physical foundations. OpenAI, the giant behind the success of ChatGPT, perfectly understands that the unprecedented hunger for electricity generated by data centers could become a "bottleneck" for future generations of models, such as GPT-5 or ambitious AGI projects. The solution is expected to be a technology that has remained in the realm of physicists' dreams for decades: controlled thermonuclear fusion.

As reported by Axios, OpenAI is in advanced talks with the California startup Helion regarding the purchase of massive amounts of clean energy. This is no ordinary market transaction; both companies are linked by the figure of Sam Altman, who is not only the CEO of OpenAI but also one of the main investors and proponents of the technology developed by Helion. The negotiated contract assumes that OpenAI could take over 12.5% of the startup's total energy production, which, on the scale of the coming decade, means numbers that are difficult to ignore even for the largest players in the energy market.

Scaling power: From gigawatts to market dominance

The terms of the agreement, although still in the early stages of negotiation, outline an ambitious delivery schedule intended to support OpenAI's computing infrastructure. According to available information, the startup would commit to providing 5 gigawatts of power by 2030. However, this is just the beginning, as plans for 2035 envision a jump to 50 gigawatts. To realize the scale of this undertaking, it is worth noting that 50 gigawatts is a capacity capable of powering tens of millions of households or the largest GPU clusters in the world.

OpenAI's strategy seems to be a mirror image of the actions of their main technological partner. In 2023, Microsoft signed a similar agreement with Helion, becoming the startup's first commercial customer with a plan to purchase energy starting in 2028. If the talks with OpenAI conclude successfully, Helion will de facto become the energy backbone for the most powerful alliance in the AI industry. For Sam Altman, this closes the cycle: creating intelligence requires energy, and nuclear fusion is intended to be its cleanest and most efficient source, eliminating dependence on fossil fuels or unstable renewable sources.

Fusion as the only path for large AI models

Why is OpenAI deciding on such risky technological bets? Traditional methods of obtaining energy, including photovoltaics and wind, struggle with the problem of intermittency, which requires expensive energy storage for data centers operating 24/7. On the other hand, classic nuclear power based on fission raises regulatory and social controversies. Helion promises something different — magneto-inertial fusion, which is intended to be safer, generate less waste, and be easier to scale in smaller modules.

Investing in Helion is a move forward for OpenAI. Training models with parameters counted in the trillions requires not only thousands of NVIDIA H100 systems but also stable grid voltage, which current infrastructure may not provide given the exponential growth in demand. Acquiring 12.5% of Helion's production secures the company against energy price fluctuations and potential supply constraints that could affect commercial data centers in the coming years. This is a purely pragmatic move, masked by a visionary narrative about saving the planet.

Conflict of interest or synergy of vision?

The relationship between OpenAI, Helion, and Sam Altman naturally raises questions about corporate governance and potential conflicts of interest. Altman, being the face of both ventures, creates a closed ecosystem where capital flows between entities controlled by the same group of influence. However, from a technological perspective, this symbiosis is logical. Helion needs guaranteed off-take agreements to finance the construction of subsequent reactors, and OpenAI needs energy that no one else can provide in such a "dense" format.

• Business model: OpenAI becomes not only a consumer but also a key strategic partner for the deep-tech sector.
• Time horizon: The first energy deliveries in 2028-2030 coincide with the predicted peak power demand for next-generation AI systems.
• Operational scale: The goal of 50 gigawatts by 2035 positions the OpenAI-Helion duo as a giant in the global energy market.

It is worth noting the technical challenges facing Helion. Although the company has made significant progress in building prototypes, commercially viable nuclear fusion has still not been achieved on a large scale. The agreement with OpenAI is a declaration of faith that a breakthrough will occur within the next four years. If Helion delivers on its promises, OpenAI will gain a cost advantage that no competitor relying on a traditional energy mix will be able to offset.

The determination with which Sam Altman is building the infrastructure around OpenAI suggests that the race for AI is, in reality, a race for physical resources. Control over 12.5% of energy production from innovative fusion reactors may prove to be a more important asset than the machine learning algorithms themselves. In a world where code is becoming a common commodity, the physical power to run it remains the ultimate determinant of power. OpenAI does not want to be just a software company — it aims to be a pioneer of a new industrial era in which AI and nuclear fusion form an inseparable whole.