On February 23rd, according to The Indian Express, OpenAI CEO Sam Altman responded to concerns about the high energy consumption of AI affecting the environment at an event hosted by The Indian Express during the India AI Impact Summit 2026, while criticizing Tesla and SpaceX CEO Elon Musk's proposed space data center plan as "absurd".
At the event, host Anant Goenka quoted a previous conversation between Altman and Bill Gates, asking if the statement 'whether the energy consumed by a single ChatGPT query is equivalent to consuming 1.5 iPhone charges' is accurate.
Altman responded, 'It is absolutely impossible to approach that level.'. He complained that many discussions about ChatGPT's energy consumption are unfair, especially those that focus on "how much energy is needed to train a model, compared to how much it costs for humans to perform a reasoning query".
Altman believes that when you ask ChatGPT, how does the energy required to answer this question, given that the model has been trained, compare to the energy required for humans to answer the same question? ”This comparison is more fair. When measured in this way, AI may have caught up with or even surpassed humans in energy efficiency. ”
Training a person also requires a lot of energy. ”Oltman explained, "It takes about 20 years of life and all the food you eat during that time to become intelligent. Not only that, but it also requires the extensive evolution of the 100 billion humans in history who have lived and learned not to be eaten by predators, learned to study science, and other abilities to create who you are today
Meanwhile, Altman also pointed out that some claims about AI water usage on the internet are "completely fabricated," but he also acknowledged that "we have used evaporative cooling technology in data centers in the past," which was indeed a problem.
We no longer do that, but you still see phrases like 'don't use ChatGPT' and 'it consumes 17 gallons of water per query' on the internet. ”Altman emphasized that this is completely groundless, utterly absurd, and has nothing to do with reality.
Altman added that due to the widespread use of AI worldwide, it is reasonable to be concerned about energy consumption, not just for a single query, but for the overall energy consumption. In his view, this means that the world needs to quickly shift towards nuclear, wind, and solar energy.
However, Altman is not optimistic about Elon Musk's proposed "space data center" plan, which aims to solve the energy consumption problem of data centers by utilizing the almost infinite solar energy in space. He said that from the current perspective, the plan is "absurd".
I truly think that the idea of sending data centers into space is absurd in the current environment, "said Sam Altman." Although it will make sense someday, if you roughly calculate the cost of launch and the electricity we can use on Earth, not to mention repairing broken GPUs in space that still frequently break down, unfortunately. We haven't reached that point yet. ”
In fact, the cost of sending 1764 pounds (800 kilograms) into low Earth orbit using SpaceX rockets is $5.6 million, but for rockets planned to launch tens of tons, the price per kilogram may decrease. However, Nvidia's NVL72 GB200 rack level solution weighs between 3000 and 3245 pounds (1360 to 1472 kilograms) depending on the specific configuration, and does not have data center level connectivity, cooling, and power infrastructure. Even with discounts, sending data centers into space is still extremely expensive, so it is currently unclear whether economic benefits can be achieved in the near future.
There will always be a day when space is great for many things, "Altman added. Space data centers will not become an important factor in the large-scale development of this decade
In addition, the cutting-edge process technologies used to manufacture Nvidia B200/B300 and other cutting-edge AI accelerators, advanced CPUs, complex DPUs, and network processors - such as TSMC's N4 (4nm level) - are not radiation resistant versions, which means they cannot cope with the radiation environment in space. However, anti radiation manufacturing technology is still at a mature stage (such as 90 nanometers), so new process technologies must be developed before computing hardware suitable for space emerges.
