Can you explain the economics behind the datagrid proposal in Southland? 3.5 billion to build for a return of 60 million per year in gdp. That seems like a very low return on investment
Good question. Datagrid is a private company, so we've less info about their internal figures than we do with a publicly traded company like Google (e.g. through quarterly earnings reports).
The only source I could find for the ‘$60m/year of GDP figure was a quote in an RNZ article. However, it's widely reported that datagrid expects to generate ‘hundreds of millions of annual revenue’, which is in line with what you'd expect for a 280MW datacentre. In fact, GPT-5.5 Pro estimates annual expected revenue at $400m, which, not coincidentally I think, would result in $60m of GST revenue for the government.
My read is that the $60m figure is actually estimated GST receipts based on a $400m revenue round figure, and the interviewee either misspoke, or ‘GST’ was transcribed in the article as ‘GDP’ in error.
Note all these figures are almost certainly underestimated, especially given the dramatic increase in both chip and token costs I talk about above.
Is it possible that all that revenue gets booked to an overseas entity and they end up paying very limited tax on that revenue, in the same way Google, etc currently do?
The tax situation becomes very complicated very quickly, but the general answer is no, structuring through an offshore entity element can ensure that company tax on profits is paid in a different jurisdiction, but GST still needs to be paid on services rendered to NZ customers. Remember that a significant proportion of revenue is electricity costs, and GST is paid on these as well, and power companies are very keen to expand capacity at the moment to meet demand, which will help strengthen the grid and overall tax revenues.
No, they'll sell to whoever they please. Local datacentres advantage local consumers due to lower latency (setting aside the strategic advantages I discuss above), but Datagrid will look to maximize the utilisation of the datacentre chips as much as possible. Tokens sold to overseas consumers will be recorded as exports and subject to IRD's transfer pricing rules. Not being a tax accountant, I have not looked into the specifics of how these will apply.
Keep in mind the IP and capital for this stuff is all coming from offshore, so we shouldn't expect to profit from this to the same extent as we would if it were all homespun, though many Kiwis will benefit from access to the infrastructure itself.
As far as the grid is concerned, datacenters (particularly those dedicated to inference rather than training workloads) have the ability to scale up and down extremely quickly, so although operators obviously want to maximize chip utilisation in general, can quickly scale down operations at times of peak loading with minimal impact (something no other user of even remotely similar scale can do) reducing the need for gas peaker plants for example in the case where there's little wind during the evening peak. Further, expanding overall supply to meet this new demand also allows expansion of other forms of generation, particularly geothermal, reducing reliance on coal, and on hydro which, while generally excellent, can have constraints in times of drought. Without demand growth, we're mostly stuck with what we've got.
Can you explain the economics behind the datagrid proposal in Southland? 3.5 billion to build for a return of 60 million per year in gdp. That seems like a very low return on investment
Good question. Datagrid is a private company, so we've less info about their internal figures than we do with a publicly traded company like Google (e.g. through quarterly earnings reports).
The only source I could find for the ‘$60m/year of GDP figure was a quote in an RNZ article. However, it's widely reported that datagrid expects to generate ‘hundreds of millions of annual revenue’, which is in line with what you'd expect for a 280MW datacentre. In fact, GPT-5.5 Pro estimates annual expected revenue at $400m, which, not coincidentally I think, would result in $60m of GST revenue for the government.
My read is that the $60m figure is actually estimated GST receipts based on a $400m revenue round figure, and the interviewee either misspoke, or ‘GST’ was transcribed in the article as ‘GDP’ in error.
Note all these figures are almost certainly underestimated, especially given the dramatic increase in both chip and token costs I talk about above.
Is it possible that all that revenue gets booked to an overseas entity and they end up paying very limited tax on that revenue, in the same way Google, etc currently do?
The tax situation becomes very complicated very quickly, but the general answer is no, structuring through an offshore entity element can ensure that company tax on profits is paid in a different jurisdiction, but GST still needs to be paid on services rendered to NZ customers. Remember that a significant proportion of revenue is electricity costs, and GST is paid on these as well, and power companies are very keen to expand capacity at the moment to meet demand, which will help strengthen the grid and overall tax revenues.
So the services are definitely being rendered to nz customers only?
Can you elaborate on how this strengthens the grid?
No, they'll sell to whoever they please. Local datacentres advantage local consumers due to lower latency (setting aside the strategic advantages I discuss above), but Datagrid will look to maximize the utilisation of the datacentre chips as much as possible. Tokens sold to overseas consumers will be recorded as exports and subject to IRD's transfer pricing rules. Not being a tax accountant, I have not looked into the specifics of how these will apply.
Keep in mind the IP and capital for this stuff is all coming from offshore, so we shouldn't expect to profit from this to the same extent as we would if it were all homespun, though many Kiwis will benefit from access to the infrastructure itself.
As far as the grid is concerned, datacenters (particularly those dedicated to inference rather than training workloads) have the ability to scale up and down extremely quickly, so although operators obviously want to maximize chip utilisation in general, can quickly scale down operations at times of peak loading with minimal impact (something no other user of even remotely similar scale can do) reducing the need for gas peaker plants for example in the case where there's little wind during the evening peak. Further, expanding overall supply to meet this new demand also allows expansion of other forms of generation, particularly geothermal, reducing reliance on coal, and on hydro which, while generally excellent, can have constraints in times of drought. Without demand growth, we're mostly stuck with what we've got.
Sounds good. What is the mechanism we use to ensure they are the low priority user in dry years and still winter nights?