Every time we switch on Netflix we are contributing to the enormous consumption of water, land and electricity needed to power artificial intelligence.
Whether we know it or not, the streamer is using AI to tailor our viewing.
Likewise, a simple Google search integrates an AI summary even when we don’t ask for it, again using energy unnecessarily.
Measuring the impact of that single search is impossible, but for the first time a report spells out the cost of AI across the world on our land, water, and electricity. And it estimates how that cost will grow in the next four years.
The UN’s Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints says that by 2030, water used by AI is expected to equal the needs of 1.3 billion people; that the electricity consumed is to be projected to be nearly three times the annual electricity use of Pakistan, Bangladesh and Nigeria with their combined populations of 650 million people; and that the land footprint could be more than 14,000 square kilometres, nearly as much as the whole Auckland region covering land and sea.
RNZ’s climate correspondent Kate Newton tells The Detail the report aims to make people aware of “how enmeshed this energy use now is into everything that we do”.
“But the other thing that they point out is that there are options that make it easier for us to not use AI, if we want to; there are options that these businesses could build in, you could have opt-outs.
“At the moment if you log into Microsoft products, Copilot is there, it’s always on. They’re pointing out that these things don’t need to be on as a default, we don’t need summaries necessarily at the top of the Google search.”
New Zealand is not mentioned in the damning report from the UN University Institute for Water, Environment and Health, but Newton says there’s a lot to learn from it and despite their environmental footprint, energy-guzzling data centres could be good for this country.
“The most pertinent thing about New Zealand is that it doesn’t really figure in this report, and I’ve talked to AI experts here based in computer science, they’re saying that a lot of these issues are theoretical for New Zealand because we don’t have a lot of AI infrastructure,” Newton says.
The most ambitious development yet, Datagrid’s planned $3.4 billion data storage facility in Southland, has fast-track approval. Once built it will supply AI technology and cloud services to customers around the world and link up to a new data cable to Australia.
Auckland University senior lecturer in computer science Ulrich Speidel says the UN report highlights the need to examine developments like Datagrid’s.
He says the near-immediate consequence for the growing environmental hunger of AI is the impact on the national power budget. For a start, prepare for higher power bills, he warns.
He says the datacentre has approval to consume 280 megawatts, roughly the same as Wellington or Christchurch.
“Currently, excess generation from the South Island goes to the North Island via the Cook Strait cables, and this means that these 280 MW simply won’t be available to send north,” he says.
“This means that we need to offset this with an equivalent amount of new generation – which could be on either island, but it needs to be built and will require upfront investment that I guess isn’t going to be coming from the data centre operator but from the rest of New Zealand.
“The consumer will pay,” he says.
Speidel says the report raises questions about whether data centres are going to benefit New Zealand.
“How much need for hefty AI models does a country with a population of six million really have?
“The number of jobs generated onshore will be fairly small on an ongoing basis, and the AI will largely work for overseas parties. There could perhaps be something like an AI tax that people pay who want to train their models on our power, but quite what that would look like I’m unable to say.”
On the question of how much energy used in one simple question to ChatGPT, Speidel says some questions are more power hungry than others but “at the lower end, 300 queries an hour consume enough power to heat a room in winter”.
“So if you’re using AI a couple of times a day that probably doesn’t amount to much but if your job is to write a large number of bespoke-looking letters an hour, you might find it’s significant.”
Newton says experts argue there are some benefits for New Zealand having its own data centres.
“Many, many countries are being left behind in terms of sovereignty over this infrastructure. China and the US are by far in the majority of this technology and also the models too,” she says.
“So, you start to not only not have the infrastructure within your country with all the controls that you might be able to exert over how they use resource – if technology is built in New Zealand we can control the power sources, we can say you can come here but you have to build your own methods of generation.”
The UN report’s researchers also point out that if a country like New Zealand develops its own models and technology then it retains control over data storage and has models that are built specifically for New Zealand’s purposes.
“That’s important for a few different reasons. We’re starting to see, even within AI summaries, cultural assumptions that are being made and answers that are being given that are based very much within a single culture, say you look at the United States,” Newton says.
It will also mean that New Zealand data is kept and protected here.
“There are experts in Māori data sovereignty that talk about the dangers of not having our precious data, things like whakapapa, that are actually housed within a New Zealand system rather than being in servers in another country.”
Check out how to listen to and follow The Detail here.
You can also stay up-to-date by liking us on Facebook or following us on Twitter.
