Data Centres – Perhaps capturing more than data? (Part 2)

By Alicia Lucas

Continued from Part 1

Changing World – Looking after themselves – USA rural and urban local governments (increasingly but not always) listening to communities

In the USA, Big Tech data centre builders such as Microsoft, Amazon and Google as well as private investors are facing increasingly costly resistance to their data centres. It is reported that over “140 local groups had managed to block or delay more than $60 billion worth of investment in US data centre projects in little over a year, according to the non-partisan research firm Data Center [sic] Watch.”

The large and hyperscale data centres are the problematic ones. There is no one definition of hyperscale nor “large” for that matter. Data centre size is often indicated using megawatts (MW) or gigawatts (GW). These present size in terms of a centre’s energy use or more specifically how much power is used at peak load. A MW is 1 million watts and a GW is 1 billion watts. Big Tech companies are usually the ones involved in large or hyperscale data centre development. There are a lot of medium to small data centres around. Government or companies may house servers and data in their own data centre. Some companies may build data centres for others to rent. Data centres may only be kilowatts in size.

More recently investors are choosing relatively flattish, wide-open rural farmland for data centre development while the focus has previously been on more built up areas. Communities and even individuals are fighting back in both cases. A farmer in Pennsylvania refused a USD 15 million offer to buy his property while state politicians are moving to protect farmland. This includes a Republican who is leading the push for banning data centre development on all Pennsylvanian farmland. Boulder County, Nevada, Commissioners have adopted a 6 month moratorium on data centres and the Boulder City planning commission, cancelled the proposed high density 167 MW data centre proposal on 89 acres of city-owned land described as “desert land” after community protest. Residents in the city of Monterey Park, California, used a ballot initiative and voted to permanently ban data centres. An Australian company was behind the unpopular plans to build a data centre of reportedly about 4 US football field sizes there. It would have been close to thousands of homes.

It is difficult to find out how many data centres have become available for use over the last few years. Ed Zitron, a British journalist, whose May 2026 article “Where are all the data centers [sic]?” describes his attempts to answer this question. The journalist searched in vain for hyperscale data centres in different parts of the world where construction had commenced in 2023 or 2024 and were now fully up and running amongst other relevant information. This time period corresponds roughly with the availability of large models labelled “AI”; ChatGPT was publicly released on 30th November 2022. Some data centre projects had been broken into stages as size made it impractical to build large centres all at once and he did note that some stages had been completed but not all.

Zitron found there are plenty of announcements about hyperscale data centres but not whether they had come online. He notes hyperscalers such as Microsoft, Amazon or Oracle, do not report on how many they have built and their size. He considered how long it may take to build a data centre (this is also a difficult question to answer but 6 small, 450 kilowatts to 60 MW, data centres indicated the construction and fitting stage takes about 18 months. A simple 1 MW data centre took about 11 months, not counting finding the land, obtaining permits, et cetera) and identified reasons why none had been fully completed including planning permit issues, continually planning to make a project  bigger or that data centres are difficult to build. Zitron reports that no 1 GW or bigger data centres have been built despite stated intentions to do so. This provides some guidance as to what big is although not what is really needed for use by a model.

Examples he gave helped show problems encountered when trying to identify how many data centres were working. For instance, information about multi-stage developments could be vague and lead to misinterpretation of the true situation. The Amazon Rainier data centre in Indiana, USA, is an example that illustrates this. A journalist reported the “(allegedly)” 2.2 GW Rainier data centre was operational but later in the article explained only 7 out of 30 buildings were working with a further two in construction. On the same day a press release by Amazon reported the centre was fully operational but did not mention the actual size of the data centre.

Another example showed how the relationships between Big Tech companies could potentially lead to project management reporting confusion. OpenAI announced it would open a giant 230 MW data centre in Norway in 2025 and less than a year later OpenAI backtracked and said it would rent the required amount (it is not necessarily the whole data centre) from Nscale, a UK company. Then Microsoft said it would take over the rent from OpenAI with OpenAI saying it would rent from Microsoft. Hopefully this doesn’t lead to a “double-up in reporting” type situation. Recent reports indicate Nscale got permission in May 2026 to build a 230 MW data centre in Norway. The original OpenAI proposal was to be its first data centre in Europe.

Further to this last example, Tom K. in a Tech Times article published on 19th May 2026 wrote OpenAI is under “sustained financial pressure” and would no longer be building data centres as intended as part of the USA’s Stargate project. Oracle is now being sued by its investors due to losses from this AI construction proposal. Since this article was published, a new investment was found for a 1 GW data centre development in Saline, Michigan, in which both OpenAI and Oracle are involved although this does not seem to replace the foregone data centres.

Stargate was announced by President Trump at the White House in January 2025 as a $500 billion project to build new OpenAI infrastructure. OpenAI, Oracle, Japanese investment company SoftBank and MGX a tech investment arm of the United Arab Emirates government planned to reach a target of 10 GW from data centres built in the USA and other parts of the world. Later in 2025, the UK became involved due to the Norway project. Six data centres were proposed in the US. Only a stage (350 MW) of the data centre in Abilene, Texas, was working and Microsoft took over construction of further stages (600 MW) from OpenAI and Oracle in March 2026. The Stargate proposal is still continuing but in a largely different form.

OpenAI, it appears, will now mainly rent from Amazon but also other companies including Google Cloud and Microsoft. The Tom K. article reports that OpenAI still haven’t made a profit and projected inference cost, that is, the cost involved in providing a result from a publicly available model based on the texts entered by people, is expected to increase greatly. Companies like Amazon and Microsoft have other businesses so data centres can be used to serve other purposes unlike OpenAI whose focus is only providing “AI” labelled products and this may explain why the former are able to continue with most of the development.

A data centre proposal that raises the question of data centre use by the military and is being strongly challenged by local community is known as the Stratos Project Area in Box Elder County, Utah. It is a joint venture between a Canadian millionaire USA TV celebrity investor and a new company. This development, if completed, may create one of the largest data centres in the world and would eventually be 7.5 to 9 GW centre within three areas totalling 40 000 acres (approximately 16 190 ha) – an area recognised as greater than 30 000 USA football fields and requiring at least near double the current power use of the state of Utah.

A gas power plant would be built along side the data centre so it would not need to draw on external power but may it also lessen the ability of the communtiy to understand and control the centre – if it is an “AI-pality”? The celebrity raises its importance from a national security point of view and it gained state approval on the 20/4/2026 from an agency known as the Military Installation Development Agency. The land is a mixture of public and private. The community only became aware of the proposal shortly after this permission was granted and knew little about it before the local county Republican commissioners approved it about two weeks later on the 4/5/2026. The county has been described as conservative.

Residents are now strongly objecting to the proposal and are appealing against the rejection, by the county, of proposals for a referendum about cancelling the project. One concern of, at least, some in the community is the possible relationship the project has with the military. A suggestion is that military involvement may simply be about ticking a box to show President Trump and his administration his executive order to have hyperscale data centres is being addressed. Another suggestion is the developers are using “security needs” for marketing purposes. A bit out there thought is perhaps they feel there is a need to digitise and store the layout of land elsewhere in the USA and any hazards that may trip up, for example, Elon Musk’s robots or those facial recognition robodogs now being rolled out to protect footy fans in Mexico. It should be recognised it is the humans in charge of “AI” that may be a problem rather than the far-fetched idea of “AI” running amok.

Changing World – Australia plods down the same well-worn pathway

Hypothetical data centre developer (no relationship to Microsoft or any other USA company) states: ‘With pathways narrowed in Europe and the USA, where to lock-in some fresh money? Australia, isn’t that the country that effectively gives away its own gas so Japan can on-sell to other countries and make a profit? Also, a handy way to meet the President’s Executive Order promoting “USA AI dominance” – just say that last bit very, very quietly. It doesn’t matter if the government knows but just don’t let the community hear’.

Microsoft said it would again build data centres in Australia, this time as part of a $25 billion investment. Last year the company cancelled a $5 million investment as they did in a number of countries. The company has also announced investment in other south-east Asia countries such as India, Malaysia, Japan and Thailand. Japan is part of the Stargate project. Microsoft may be pulling back in Indonesia. Amazon, Google and others may also be expanding their data centre reach into these countries. For some reason (specified as Australia can be come a leader in the coming technological change) Microsoft has been allowed to develop data centres in Australia since 2014 and by 2023 had 29 sites (Energy Magazine, 2026).

These newer data centres will go some way to protecting data sovereignty as they will use the newer “sovereign”-type Microsoft designs. These are the ones the Europeans appear to have questioned although noting it is not quite the same situation as Australia has an agreement about data transfer for selected reasons with the USA.

USA companies may be being subsidised by the USA government as Microsoft’s investment may be, at least, partly underwritten by USA export financing if this is being done as part of the President’s Executive Order promoting “USA AI dominance” through exporting of an “AI Technology stack”. The package brings together the range of elements needed to run “AI”-labelled models and that must, as much as possible, be sourced from the USA and range from hardware such as chips, data centre storage, data pipelines and labelling systems (where images are labelled, for example, “label picture of banksia ‘banksia’,” so they can be used in a model. Uber’s idea was drivers could label images as piece rate work while waiting for the next passenger), to the “AI”-labelled models, themselves, cybersecurity and apps such as those for education, health and transportation. The apps indicate they do not just want to supply models and their infrastructure but also undertake digital operation of typical government functions.

Australia may have the second highest number of data centres in the world. The USA has the most. Australia wants more. It is not clear how many data centres Australia needs and what consequences there will be if the “AI”-labelled models and/or data centre bubbles burst. Could “saving face” lock us into accepting other technologies that may not be wanted such as another countries humanoid or doglike robots? This sort of scenario may be aggravated by the government strategy of including “AI” throughout the economy.

As a strategy broad deployment of “AI”-labelled models sounds democratic. An economy can be simply described as the production and consumption of goods and services. So that labelled “AI” should be available to all organisations or individuals producing or consuming goods and services. This can cover near everything in Australia including education, shops and incarceration as well as consumers. Everyone or organisation has the opportunity to experience and, perhaps, build on new technology. In practice, this assumption of “democratic advantage” may be largely unfounded or even, if not, may cause harms as social media is now recognised as having done.

The USA’s “One laptop per child” program was based on this “democratic advantage” model. People may recognise the idea that children are “digital natives” behind this quote:

“How many of us have questions about a computer and ask someone who is older to fix it? None of us! You find the youngest person in the room and they’ll have it fixed in a second. These kids get it, and we need to give them the tools that they’ll need to succeed.” Mayor of Birmingham, USA, Larry Langford as recorded in Morgan G. Ames, Mark Warschauer, and Sheila R. Cotten (preprint draft) The ‘Costly Lesson’ of one laptop per child Birmingham’.

There may have been good intentions for this project but it was shown not to have worked.

Other interpretations about children and old people and technology are worth considering, for instance, could this blanket exposure really lead to children both unquestioningly accepting information technology as integral to their lives as well as giving them a false sense of their ability to understand and control that technology? It might also have hastened widespread, unconsidered, adoption of digital products as adults took this attitude as a challenge and came to grips with information technology. Now, as retail jobs are replaced by partial automation self-service, on-line ordering, maybe even “checkout” trolleys and we are given non-threatening glimpses of robots in the news, including at least one serving in a shop overseas, young people seem to go along with it.

Perhaps, young people are also misunderstanding the older folk still using the staffed checkouts aren’t technologically illiterate but rather are attempting to save jobs for retail workers. The retail industry is the second largest employer in South Australia and Tasmania, third largest employer in Queensland and Victoria, fourth largest employer in New South Wales and Western Australia and fifth largest employer in Northern Territory according to the Labour Market Dashboard in January 2026, Jobs and Skills Australia. That’s a lot of people.

It is not apparent that the widespread availability of information technology and early exposure of children to using computers and information technology including “coding” has resulted in a commanding information technology sector in Australia. This may be the same with “AI”-labelled models.

Industry seems to continue on its usual path as well. Companies are subject to minimal “AI” regulation and implement it without consideration of its ramifications to society. Big companies often appear to have jumped right in without trying to understand what “AI” is and possible problems. For example, a recent article explained the CEO of the Commonwealth Bank, the former “People’s Bank”, has been aggressively pursuing “AI” only “treading more carefully”following the earlier “AI” failure that lead to rehiring customer service staff when problems made an “AI” roll back necessary. A number of these companies seem to be using the USA’s “AI”-labelled models and could it be that they are effectively becoming part of the USA “AI”-labelled model companies? How much sovereignty do they really retain? Companies could have taken the approach of German industry where “AI” is targeted to specific problems.

If companies develop machine learning models they may gain a more realistic understanding of the advantages and limitations of using “AI” and, particularly, true costs and whether “AI” was leading to profits. Those using “AI”-labelled models for communicating with the public should take notice of David Astle’s article, also in The Age, where a straightforward task with TicketMaster was made time-consuming and onerous. These models don’t work on “this is the answer to this question” approach. They can give different answers even if a similar question is asked.

It makes one think that, perhaps, if the “People’s Bank” still existed, Australia would have taken a  more informed, sensible and sovereign approach to what experts have argued is an over-hyped product with baked-in problems (aka “hallucinations”) and, in Australia’s case, perhaps taking more than just our data.