A surprise issue that has united some Republican and Democrat candidates during the US midterms election, closing on 4 November, is opposition to the scale and impact of the new generation of hyperscale data centres being proposed or being built around the country. Nearly 5500 data centres already in the US make up around 45% of the world’s facilities, according to Statista and Cloudscene, and a further statistic suggests that 54% of hyperscale capacity resides in the US. It is unclear as yet whether the negative talk is going anywhere. Polling in the US, Canada, the UK, and Europe – has found growing disquiet about data centres being assembled near residential communities. If words turn into action, e.g., a more cautious permitting policy, this sentiment may turn out to be one of a number of significant setbacks to the adoption of AI, which depends on the spread of these massive and intrusive facilities.
The most cited local objections to data centres that apply everywhere include the burden on inadequate electricity grids, potentially higher utility bills, the need for alternative power and battery energy solutions, pressure on water supply, environmental pollution fears, impact on climate change, and scepticism about the economic benefits, for example, the number of long-term jobs after the construction phase.
It may not be fanciful to suggest a deeper antipathy is also beginning to manifest itself. Data centres represent the only tangible expression of the unprecedented boom in artificial intelligence (AI) investment. As such, they make stoke a commonly articulated fear that a little-understood technology is taking over our lives, controlled by a few, unregulated global business behemoths.
A deeper antipathy is also beginning to manifest itself.
The numbers are extraordinary. Probably not an exaggeration, one analysis estimates that the 10 largest public companies operating in the AI space—spanning AI chips, cloud infrastructure, and AI-driven software and services—have a combined market capitalisation of at least $25 trillion, bigger than the gross domestic product (GDP) of every country except the US, and larger than the entire Chinese economy. At the top of the tree are the usual suspects: Nvidia and South Korean company TSMC rule the chipmakers; Microsoft, Amazon and Google own the data centres; and OpenAI, Anthropic and Meta provide the research and consumer applications.
A home for more data centres will certainly be harder to find if doubts surfacing in the financial sector about the valuations of AI-led companies begin to materialise into a collapse of confidence, thereby putting massive infrastructure investment plans in jeopardy. The imminent gigantic Anthropic IPO, which is targeting $1.5-2 trillion, could prove a test of the market’s appetite for AI investment. The company reported an $80.6 billion loss in the first half of 2026; it is exposed to $321.4 billion in long-term data centre and cloud infrastructure leases. There have been data security issues, and some argue its revenue model is vulnerable.
Blackstone, the private equity investment giant, has warned about a possible downgrading by the main credit rating agencies. In addition, Anthropic CEO Dario Amodei recently warned about the dangers of AI if the pace of development did not slow down, and a lead researcher in the company suggested that there was a more than 10% chance AI ‘could kill all humans’ in the next decade.
Alarmed by the dominance of the sector by a few, almost exclusively US tech companies (egged on by the current US president), countries such as the UK, Canada, Australia, and Norway have established national AI strategies, and the EU is assembling its own comprehensive set of policies, regulatory frameworks, and financial initiatives. Governments are basically trying to limit their dependency on other countries for AI development, operations, and oversight to protect their data, their national security, and their economic prospects. However, a recent paper published in July from Stanford University's Human-Centred Artificial Intelligence, casts doubt on how totally homegrown ‘AI sovereignty’ independence can exist without some involvement of the big tech operators.
Governments are basically trying to limit their dependency.
What this may all mean for progressing applications in the science disciplines such as geophysics, geology and the earth sciences in general is obviously a matter of speculation. There is obviously no stopping AI research and implementation; the genie is out of the bottle and proving very beneficial in improving efficiency and supporting the development of new solutions at an unprecedented pace. In other words, AI is too big to fail.
We should not get too complacent. There is a scary scenario to be found in a doom-laden article entitled ‘When it all comes crashing down: The aftermath of the AI boom’ authored by science author Jeremy Hsu in last December’s Bulletin of the Atomic Scientists. It speculates on the financial, energy, environmental and societal implications, mainly but not exclusively from a US perspective. Hsu does not specifically mention the impact of a crash on the geoscience community, but it is not hard to speculate on AI tools becoming less accessible, disruption in the labour market focused on digital solutions, crises for companies marketing AI-based solutions, etc.
Hsu concludes: ‘However it ends, the AI bubble’s most enduring legacy may be the global disruptions from any financial crisis that follows—and the societal costs already incurred from betting so heavily on energy-hungry data centres and AI chips that may suddenly become stranded assets … But the alternative is the prolonging of an AI bubble that is increasingly unsustainable in both the financial and environmental senses, with the winners mainly being some of the wealthiest companies on the planet and their investors’.
Views expressed in Crosstalk are solely those of the author, Andrew McBarnet, EAGE Editor Emeritus. He can be contacted at andrew@andrewmcbarnet.com.
