Intro
The prime catalyst for AI data centres growth is Trump’s fetish for such centres.
The global IT capacity of data centres under construction represented 23 gigawatts (GW) by September 2025, 75% of this capacity is in the U.S.
For 2028, the energy impact is estimated to be around 6.7% to 12% of electricity U.S. demand dedicated to data centres.
Seven companies, dominate U.S. data centres landscape, Alphabet, Amazon, Apple, Meta, Tesla, Microsoft and Nvidia.
The exponential race to build U.S. data centres is
- driven by a subsidized private debt bubble;
- undermined by the credibility and/or quality of responses from AI; and
- mostly accommodated by fossil fuels
Much of content produced is slop. The word “slop” is a recognized word in the Merriam-Webster English dictionary for “digital content of low quality that is produced usually in quantity by means of artificial intelligence.”
The bubble
Globally, 23.1 GW of capacity is under construction at 831 sites, the Americas accounting for 17.1 GW over 311 locations. For the Asia Pacific region, it’s 3.2 GW over 283 sites and Europe Middle East and Africa, 2.8 GW across 258 sites.
Private equity and subsidized debt financing are the pillars for continuous growth of U.S. data centres.
In the U.S., nearly 100% of AI investments in data centres comes from subsidized private equity and debt financing. These centres have yet to prove they are profitable. This model can only work if there is infinite growth. The vast amounts of debt incurred for data centres has created an inflated bubble which would burst if the growth stopped.
Large global banks and Wall Street are concerned about the speculative nature of these investments.
As well, since chips and servers evolve, there is built in planned obsolescence.
In 2025, in Pennsylvania alone, the Trump administration and the state have announced $100 billion in investments to accomodate data centres. This includes $1 billion for the restart of Three Mile Island nuclear facilities, left idle for 3 decades, to supply Microsoft and other data centres in the state. Another $20 billion was committed from Amazon, Google, Westinghouse, Meta and others.
Slop: Poor quality AI answers
Only half of AI pattern produced packaged answers are credible. For the rest, AI invents wrong answers and/or uses dated content when it has no idea of the right answers.
Israel’s Lavender and Gospel projects used predefined autonomous data matching sets and sensor information from all sorts of systems, eg cell phone data and social media, to target “potential” Hamas combatants. These algorithms are prone to getting it wrong, but Israel’s military forces applied it for the machines to learn to make decisions, nothing to do with human intelligence and integrity. This intelligence was responsible for Gaza genocide 72,551 deaths and 39,000 children being orphaned, losing, and/or separated from, one or both parents.
The backbone for the pattern matching is the product of slave-like workers reviewing thousands of data sets in poor countries. For this work, the average pay is $2/hour.
Social economic impacts
Microsoft CEO Satya Nadella argues that the AI data centres will contribute to major productivity gains which will deliver economic growth Hence, data centre sector will expand indefinitely.
But 95% of surveyed businesses using AI claim AI makes no difference on returns. AI benefits have primarily been applied to sales and marketing pilots.
Most of the jobs that come with new data centres pertain to the construction of the humongous sites. But once constructed, there are not many jobs, apart from maintenance and operation.
Also, typically, a data centre consumes 1.14 million litres of water/day for cooling purposes, but large data centres can consume up to 19 million litres of water/day. Collectively, data centres in the U.S. will consume 13.8 million litres of water by 2028, equivalent to the needs of 360,000 households
Two-thirds of the U.S. data centres to-date are in water-stressed states, among them Arizona, Texas, Nevada, and the Colorado River Basin.
Decision-making on U.S. data centre projects usually bypass local governments, leaving it to regional and national governments. The hasty pace for approvals by the companies concerned entails risks, environmental degradation coupled with a lack of time for defining new regulations.
If the local utility energy must expand its infrastructure quickly, the result is either higher utility rates or fossil fuel alternatives.
The the centres are often planned for rural communities, because of the amount of land required for these centres. This often results in a total rupture of the way of life of inhabitants of such regions. Taking precedence is Trump’s AI fetish belief that state legislatures cannot sideline AI projects.
For example, Digital 1/PAX-1 is one rural data centre data centre planned for Pennsylvania, which will cover nearly 300 hectares – the size of 80 American football fields.
This project involves a $15 billion investment that would supposedly create 450 jobs and generate $65 billion in tax revenues.
A grassroots Pennsylvania organization, Protect PT, struggles to rally opposition to these investments.
Meta’s $7 billion data centre in Michigan, Saline Township, near Ann Arbour, was rejected by local authorities. This was taken to court which reversed the Township rejection.
The small town of Archibald, Pennsylvania, population 5,400, is the location envisioned for 5 data centres requiring 1.2 million square metres (13.4 million square feet).
Similarly, Project Sail, comprises a planned 900 MW data centre of 336 hectares (830 acres) centre in Georgia, equivalent to 600 football fields, in Newnan, 56 km (35 miles) just south of Atlanta.
Another Georgia project, the Coweta County Council endorsed rezoning for Project Peach a large-scale data center being developed by Dallas-based CyrusOne in the town of Palmetto, about 24 km (15 miles) northeast of the Project Sail site.
For both of these Georgia projects, the Trump administration exercised pressure to fast-track the projects, even offering public-private partnerships.
Citizen resistance in other countries include the Chile, MOSACAT (Community Movement for Water and Land)) engaged in a legal battle since 2019 to stop a proposed construction of a hyperscale Google data centre in Cerillos, outside Santiago. After years of protests and organizing spearheaded by MOSACAT’s Tania Rodriguéz, in February 2024, a local court halted the project, demanding that Google revise its plans so that they take into account the intense water consumption and Chile’s significant droughts.
In Ireland, deregulation and fossil fuel development to accommodate these centres are making a comeback.
Not surprisingly, Microsoft CEO Satya Nadella is somewhat concerned about a public backlash.
Clean energy
At first, the big players favoured clean energy options.
On a world scale, large data centres accounted for half of 2025 power purchase agreement (PPA) capacity.
In the Americas, large data centres represented 72% of clean energy procurement in 2025. This often requires increasing the power capacity of existing infrastructure for generation and grid. Many a time, the consequences are an increase in electricity rates and delays for closing coal plants for all in the area concerned.
For the Asia-Pacific region, data centres accounted for one-third of PPA capacity, while in the European, Middle East, African nations, the PPA power procured was one-eighth of capacity.
The exponential burst of U.S. construction projects for quick AI data centres has engendered a race to grab as much energy they can get to come on board quickly.
Installing renewable energy projects cannot keep up with the exceptional speed of data centre projects mushrooming across the U.S.
Since data centres operate 24/7, it is a challenge to have renewables meet the tasks, without cutting capacity for businesses and residential clients.
For Project Sail, Georgia Power announced a $16 billion expansion of 10 GW to its power capacity by 2030, two-thirds of its current capacity. The Trump administration will provide $26 billion loan to Southern Company, the parent firm of Georgia Power.
Georgia Power has tripled its decade long projection for power demand from 12 GW in 2025 to 36.5 GW. Data centres are the source of 95% of increased demand.
Yet right now, data centres on average consume a minor share of a utility’s power, but it may reach 50% in 10 years.
Some data centres rely on their own centre-specific production of renewable energy
Google has invested in clean energy, more than private and public entities elsewhere in the world, to render some of its data centres self-sufficient. Google has signed PPAs for wind and solar, co-locating renewables to connect with its data centres and spent $3 billion on rehabilitating aging hydroelectric dams in Pennsylvania.
Google will bring 2 GW of clean energy online in 2027-28. This preempts utilities raising electricity rates to invest in infrastructure to accommodate data centres.
Google restarted a nuclear plant in Iowa under the umbrella of a 25-year PPA.
The Switch data centre south of the Las Vegas strip operates 100% with renewables, incredible for a centre that covers one square kilometre (mile). It has a 1 GW solar field and is building more solar fields.
Nevada’s NV Energy requires that data centres fund their own energy needs but has no parameters for renewables.
Fossil fuels
The hic is such that regardless of well-intentioned environmental considerations, the sweltering momentum to build data centres ASAP, has got data centre giants focused on any source of energy to put energy supplies online quickly.
This is where fossil fuels come in. Over half of the U.S. extraordinary demand will be accommodated by fossil fuels. By 2035, these data centres will consume 106 GW of power, which is greater than the current installed capacity of hydro and nuclear in the U.S.
If present trends continue, U.S. emissions from generating so much electricity sourced with fossil fuels, that by 2028, the emissions would be equivalent to 10 million vehicles.
One of Trump’s favourite solutions includes the prolonging of the lifecycle of coal plants.
Though Google began with data centres with energy stemming from renewables, the company has mellowed their clean energy goals.
Google’s explanation on new emphasis on fossil fuels is that it is no longer “maintaining operational carbon neutrality”.
Google contracted for 2 methane-fired thermal generating plants. — one in Illinois and one in Nebraska — that would use carbon capture technology – a greenwashing solution.
Google has plans for an Armstrong Texas data centre campus which would be powered by a 933 MW methane thermal generating plant that will emit 4.5 million tonnes of CO2/year. That’s more than the 4 million tonnes emitted by San Francisco per year.
Meta, Amazon and Microsoft have plans for methane thermal power plants for their AI data centres.
Microsoft has entered an agreement with Chevron for a 2.5 GW gas power plant in Texas and another gas facility is planned for a West Virginia data centre.
Musk’s Colossus data centre in Memphis Tennessee is supplied by dozens of methane portable generators because Musk doesn’t have the time to wait for the local utility to meet his needs. Musk is seeking additional debt financing for Colossus 2 nearby, with energy furnished from a former gas plant.
ExxonMobil CEO Darren Woods assures that we need not worry about emissions because 90% of emissions can be captured and abated.
NextEra, serving a dozen states, concluded it cannot meet its net-zero goals for 2045.
NextEra explained that a carbon abated gas power plant at one data centre site in Southeast U.S. will be connected to Exxon’s CO2 pipeline network that encompasses Texas, Louisiana, and Mississippi. Exxon would get a credit for this technology of $85/ton for any carbon captured.
Never mind that CCS NEVER meets goals for emission reductions, costs, and timelines.
Nevada had a goal of 50% clean energy capacity by 2030. But the state cannot meet the power demands of planned data centres, triple the power consumed by Las Vegas, without turning towards fossil fuels.
For J.P. Morgan Asset Management strategist, Stephanie Aliaga, a few million tonnes of emissions would not stand in the way of AI’s major contributions to economic growth.
With the upcoming intensification of energy demand from data centres, North Carolina plans new gas plants and prolonging the “best before expiry” date of coal plants.
Orders for U.S. new gas plants are backlogged.
Mark Carney eyeing AI opportunities for gas exports
Mark Carney’s current Canadian government views America’s mounting needs for AI data centres as an opportunity and thus invited the U.S. to import more Canadian gas. This is consistent with Carney’s “national interest” initiatives, exempt from existing legislation, favouring abandonment of Canada’s climate objectives, all with an emphasis on oil and gas, at the expense of action on climate change.
Also consistent, the Carney administration repealed the Clean Energy Regulation to open the door to the major Alberta gas company, Capital Power, for its intentions to build and supply a large AI centre in the province. This is in addition to the Alberta government exempting a 7.5 GW data centre from an environmental evaluation.
With a similar perspective, the Canadian province of Saskatchewan welcomed data centre with a gas facility energy source just south of Regina.
The takeaway
Increased emissions, higher electricity rates, threatened water supplies, job losses to automation, overload of misinformation and/or the rupture of the small-town life are but some of the challenges posed by AI data centre proposals.
The U.S. accounts for 75% of global data centre capacity. U.S. leadership stems largely from subsidized private debt for financing. The outstanding issue may be when, not if, the bubble will burst.
The majority of businesses are not significantly benefiting from AI in terms of productivity.
There is much AI slop,
AI is beneficial when there is credible training for machine learning.
But there aren’t any constraints on improper training, emissions and environmental degradation at-large.
The speed of development is primarily self-serving antisocial and/or out-of-control parameters which create albatrosses. Accordingly, there are many AI systems capable of lying or misinforming while being whopping new sources of fossil fuel emissions.
Can governments regulate this?
