OpenAI’s long term infrastructure goals could require an unprecedented amount of electricity and hardware, raising serious concerns about environmental impact and global resource strain. As reported by Tom’s Hardware, CEO Sam Altman outlined an internal plan in September 2025 to reach 250 gigawatts of compute capacity by 2033.
Truthdig estimates that this level of power consumption matches the electricity needed to run all of India, a nation of 1.5 billion people. It would also emit twice as much carbon dioxide as ExxonMobil, which the report identifies as the largest non state carbon emitter in the world.
To run 250 gigawatts of compute, OpenAI would need around 60 million Nvidia GB300 GPUs at any given time. That translates to about 30 million new GPUs every year to maintain continuous operation. While some analysts claim high wear and tear from 24 hour usage, the more accurate explanation is rapid generational turnover, which reduces the economic value of older chips. Regardless, the scale of hardware required is staggering.
OpenAI is not alone. Other major companies are planning massive deployments of AI hardware. Elon Musk’s xAI aims to acquire 50 million H100 equivalent GPUs by 2030, which would require about 5 gigawatts of power. The global demand for energy to support these data centers is already affecting national power grids, pushing electricity prices higher and lowering power quality for households. The immense cooling requirements also raise concerns over water supply and local environmental stress.
The report goes beyond data centers and examines the environmental impact of chip manufacturing. Explosive demand for AI processors has triggered the construction of 97 new semiconductor fabrication plants worldwide in the last two years. These facilities consume enormous amounts of power and water and rely on toxic chemicals used in advanced chipmaking.
For example, TSMC’s Fab 25 will require at least 1 gigawatt of power, enough for more than 700,000 homes in Taiwan. It will also need 100,000 metric tons of water every day, equal to the daily usage of nearly 200,000 residents of Taichung. Labor groups in South Korea have also linked chemical exposure at Samsung factories to cases of worker cancers.
From mining rare earth materials to powering data centers to operating chip factories, the entire supply chain behind modern AI demands vast natural resources. Tech companies are spending billions to scale up computing power while governments compete to limit each other’s access to advanced technology.
As the report concludes, while the industry focuses on how much computing is needed to advance artificial intelligence, a more urgent question may be how much artificial intelligence the planet can realistically sustain.
