The daily cost of ChatGPT, OpenAI's chatbot, is estimated at around 700,000 dollars, according to an estimate from last year. Other analyses put the figure as high as one million a day. What is certain is that maintaining and running products of this kind is extremely expensive, and the bills can only increase as services become more widespread. OpenAI, and other companies in the sector, would therefore be trapped in what amounts to a vicious circle: reaching new users in order to grow turnover, while knowing that each new user can only drive their costs higher.

A technological breakthrough is needed to make AI more sustainable

It is partly for this reason that Microsoft, an investor and commercial partner of OpenAI, is working on a new type of chip that should make these systems more efficient and less energy-hungry, and therefore more sustainable. Sam Altman himself, co-founder and chief executive of OpenAI, recently stated that the only way to resolve the issue is a "technological breakthrough" capable of making the development of AI cheaper. Take GPT-4, the large language model underpinning ChatGPT — the one that enables it to understand and generate text and images: beyond the enormous upfront cost of so-called "training", each request and interaction with GPT entails a far-from-negligible consumption of energy. As of today, however, no breakthrough of this kind appears to be on the horizon.

The sustainable alternatives available on the market

So, while Microsoft invests in chips, Altman attempts to navigate around the issue and dreams of ultra-cheap energy sources. The entrepreneur has in fact long invested in Helion, an experimental startup working on nuclear fusion whose aim is to create "artificial suns" — infinite and extraordinarily powerful sources of clean energy. Nuclear fusion has by now become the Holy Grail of renewable energy, but thanks to pressure from Altman, Microsoft has signed an energy purchase agreement with Helion starting from 2028. In the meantime, however, there are sustainable alternatives available today that could be pursued.

Sunjya Schweig, MD @drschweig

𝕏

The amount of #energy that #AI uses is huge. Crash course with #climate and global warming.

Arijit Sengupta: “The world is actually headed for a really bad energy crisis because of AI unless we fix a few things."

futurism.com/sam-altman-ene…

💬 1 20 gen 2024 · Apri su X →

+775%

The problem is set to grow considerably larger as AI becomes more widespread in the coming years. According to research from the University of Washington, handling hundreds of millions of requests to ChatGPT would consume around one gigawatt-hour, "or the equivalent of the energy consumed by 33,000 American homes." According to Arijit Sengupta, chief executive of an AI sector company, we are currently at roughly 1 per cent of the adoption of such systems that we will see in the coming years — a rosy outlook for the sector, but one that conceals an enormous environmental risk and the possibility of "a rather serious energy crisis."

Artificial intelligence is only one part of the energy consumption and CO2 emissions of the entire digital sector, whose environmental impact is set to grow exponentially in the coming years. Between now and 2040, a recently published study in the journal Nature argues, CO2 emissions linked to the digital sector are set to grow by 775 per cent, rising from 1.6 per cent of the total recorded in 2017 to 14 per cent. To put these figures in perspective, if the digital sector were a nation, it would rank fifth among the world's largest emitters of carbon dioxide, accounting for 3.8 per cent of total emissions. That is a value four times greater than that of an industrial power such as France.