When generative artificial intelligence came on the scene in late 2022, the technology seemed like magic. A user could type a few words, and out came an automatically generated text or image that, while not perfect, was often good enough. It promised to quickly make a range of jobs—from writers to paralegals—more efficient, if not obsolete.

Almost two years later, generative AI looks less like a fairy godmother than a voracious beast, feasting on investment, energy, and intellectual property while producing little of value in return.

The companies behind this technology will have to drastically curb their ambitions and get to grips with the enormous costs and large carbon footprint – or otherwise prove that their technology can deliver on the promises they have made to transform the economy.

First, there are the fiscal costs. In recent months, investors have become increasingly concerned about how quickly spending on AI infrastructure has increased. Dario Amodei, CEO of Anthropic, which has received investments from Amazon and Google, told Time that the cost of training generative AI models has risen from tens of millions of dollars in recent years to $1 billion this year, and expects it to rise to $10 billion in the near future.

Goldman Sachs released a research report at the end of June saying that companies would spend a total of $1 trillion in the coming years on investments such as data centers, computer chips and electricity.

Jim Covello, head of global equity research at Goldman Sachs, said in the report that there is no evidence so far that the technology companies developing generative AI – or the companies using it – have seen a noticeable increase in revenue despite the spending. The only companies making a profit are those providing the infrastructure, he noted.

Covello also does not expect these costs to fall any time soon, as the number of suppliers – such as chip maker Nvidia – is so small that they have the market power to keep their prices high.

It would be one thing if it was just throwing investors’ money down the drain. (That’s not a good thing for those investors, though, or for anyone with a retirement plan who falls for the hype.)

But it quickly becomes apparent that generative AI has an almost unquenchable thirst for water and energy.

According to the International Energy Agency, data centers are a major driver of global energy consumption. In 2022, data centers consumed 460 terawatt hours of electricity. Given increasing demand, they are expected to consume 1,000 terawatt hours in 2026 – roughly equivalent to Japan’s annual energy needs.

These electronics run hot and need water to cool down. According to one study, AI computers could use 6.6 billion cubic meters of water by 2027, equivalent to the needs of six Danish states. The researchers had previously estimated that the generative AI tool ChatGPT uses half a liter of water for every 10 to 50 incoming requests – the equivalent of repeatedly pouring water bottles onto the floor while working.

AI “is one of the most energy-, water-, and mineral-intensive infrastructures on a planetary scale that we as a species have ever built,” AI researcher Kate Crawford recently said on The Globe and Mail’s “Machines Like Us” podcast. “It’s truly astronomical.”

All of this, of course, is happening as the earth continues to warm and countries work to reduce their greenhouse gas emissions – not find innovative new ways to increase them. The higher demand for air conditioning during increasingly hot summers will put greater strain on power grids, and water will become a scarcer resource as droughts become more frequent and widespread. Given the enormous environmental problems facing humanity, a chatbot tool doesn’t seem to be high on the priority list.

There may be targeted applications of the technology that can be made more efficient and use fewer resources, but still provide benefits. For example, generative AI holds promise as it relieves busy professionals like doctors and lawyers of extensive paperwork, or helps programmers speed up their work.

But as a general-purpose tool, the magic of generative AI seems to be little more than an enormously expensive parlor trick.