The Green Party is calling for a one-year moratorium on the consenting of new “hyperscale AI data centres”, citing the environmental impacts of AI.

The press release (which can be read here) calls for a “one-year moratorium on new AI data centres”, with a particular focus on “hyperscale AI data centres”. The document explicitly references the (already consented) Datagrid data centre, which will be built near Invercargill and will use approximately 7% of New Zealand’s electricity generation (second only to the Tiwai Point Aluminium Smelter).

I’m not going to try to argue against the policy. In fact, I actually agree with aspects of it. After all, AI does indeed come with a host of environmental impacts.

However, like all good press releases, the document is gloriously vague. The term “AI data centre” appears six times in the 418-word statement, yet the term “AI data centre” is never defined, which leads me to the following question: what is an AI data centre? Is it a computing facility that exclusively works with AI models? Or is it a facility that, like virtually all other large computing facilities, works with a mix of AI and non-AI computing?

If that sounds strange, remember this: the same company that runs Gemini runs Google Search, YouTube, and Gmail. Welcome to modern computing.

Fundamentally, a data centre is a group of servers. I could install a server rack in my bedroom, train AI models on it, and call it an “AI data centre”.

Definitions matter because laws and regulations do not apply to political slogans. They apply to real things with real impacts. If Parliament were to pass a law targeting “AI data centres”, it would have to qualify what an “AI data centre” actually is. Is the criterion ownership? Workload? Hardware? Electricity consumption? GPU count? Percentage of computing power allocated to AI? A Google facility that spends 95% of its time serving YouTube videos but occasionally trains Gemini models becomes an "AI data centre", while a university supercomputer performing climate modelling with thousands of GPUs does not. The distinction is political, not technical.

If the real problem is electricity use, then regulate electricity use. If the problem is water use, regulate water use. If the concern is the strain on the national grid, assess projects on the demand that they place on the grid. Those are objective criteria that apply equally to any large computing facility, whether it runs language models, weather simulations, protein-folding research, or video streaming. Regulating what software happens to execute on the servers is a remarkably poor way of addressing environmental impacts.

AI is the political and social villain of the moment, so "AI data centre" has become a convenient label. But a data centre is ultimately just a building full of computers. The environmental impact comes from how much electricity and water it consumes, not whether the processors spend Tuesday training ChatGPT 5.5 or serving Gmail. If we believe hyperscale computing needs greater scrutiny, then let's scrutinise hyperscale computing. We should write policy that survives the next technology trend, rather than policy that chases the current one.