SmarterArticles

The clock tower is the detail that sticks. In the hills of Guian New Area, in China's south-western province of Guizhou, there is a cluster of buildings done up in a kind of continental European pastiche: red roofs, arcaded facades, a multi-arched bridge, and a tower with a clock on it. When AFP visited in July 2026 the whole confection was emitting a low, permanent hum. It is not a resort or a theme park. It is Huawei's largest data centre, and the hum is the sound of several hundred megawatts of cooling and compute doing whatever it is that compute does.

Outside it, on the road, street vendors were selling lunch to data centre workers beneath a banner exhorting everyone to promote high-quality development. A shopkeeper named Shu Peihua told the news agency what the change had felt like from ground level. It used to be a barren mountain, Shu said, but since the area has developed, transportation has become more convenient, trade has picked up, business opportunities have emerged. Another resident, Li Xixiu, put it more plainly still: the centres had really boosted the economy of this entire area, and the villagers in the neighbourhood now find jobs nearby, where before they had to go to other places.

Hold on to those two statements, because they are true. They are also, in the way that ground-level truths often are, an incomplete account of what has happened to Guizhou. In the same report, researchers at Taiwan's Research Institute for Democracy, Society and Emerging Technology laid a different set of numbers beside them. Guizhou has averaged 7.4 per cent annual GDP growth over the past decade. Its wage growth over the same period was second to last in the country. And the province, having borrowed heavily to build the roads, substations and fibre that make it attractive to a hyperscaler, now carries one of the highest debt burdens in China.

That is the puzzle. A place where the growth arrived and the prosperity did not.

What the Ledger Actually Says

The DSET analysis is worth reading slowly, because it separates two things that boosters routinely fuse. Data centres, the researchers found, drive heavy investment in land and equipment, but their impact on boosting local per capita income remains limited. Investment is not income. Capital formation is not a wage. A billion yuan of servers sitting in a shed in Guian counts towards provincial output in exactly the way that a billion yuan of anything else does, and it counts whether or not a single additional person in Guizhou is better paid as a result.

The debt side is starker. By 2024, according to DSET's figures, Guizhou ranked thirtieth among China's provincial-level jurisdictions on debt-to-revenue and twenty-seventh on debt-to-GDP. There are thirty-one of them. This is not a province that dabbled at the edges of the borrowing economy; it went in at the deep end and stayed. The South China Morning Post has reported that Guizhou's outstanding government debt in 2023 amounted to around 72 per cent of provincial GDP, well above the 60 per cent threshold the central government treats as prudent, after years of hefty infrastructure spending on projects that did not all deliver what local officials hoped.

Andrew Stokols of Singapore Management University offered AFP the general version of the finding. Data centres, he said, do not necessarily create a huge spillover effect on local jobs, and immediate benefits have been elusive, in China and elsewhere.

Elsewhere is the operative word. The Guizhou story reads as if it were about Chinese state planning, provincial competition and the peculiarities of local government financing vehicles, and in part it is. But strip away the Chinese institutional furniture and what remains is a much more general fact about a particular kind of asset: enormously expensive to build, almost costless in labour to run, and structurally disinclined to share.

How a Poor Province Became a Server Farm

Guizhou did not stumble into this. It went looking.

The province is mountainous, landlocked, historically among China's poorest, and for most of the reform era its principal export was people. Karst topography makes farming hard and heavy industry harder. What it has is altitude, a cool and stable climate, geological stability, cheap land and a lot of hydropower. In 2016 Beijing designated Guizhou as the country's first national big data comprehensive pilot zone. Guiyang, the provincial capital, began hosting an annual international big data expo. Apple's Chinese iCloud operation was routed through a facility in Guian built with the state-backed operator Guizhou-Cloud Big Data. Tencent went in. So did Huawei, at scale.

In 2021 the strategy got a name, and in February 2022 it got a budget. Eastern Data, Western Computing, or 东数西算, is the National Development and Reform Commission's programme to build eight national computing hubs and ten national data centre clusters, siting the compute-heavy, latency-tolerant workloads of China's digital economy in the energy-rich, land-rich, underpopulated west while the east keeps the customers. The hubs sit in Beijing-Tianjin-Hebei, the Yangtze River Delta, the Greater Bay Area, the Chengdu-Chongqing corridor, Inner Mongolia, Ningxia, Gansu and Guizhou. The NDRC's own projection was that the hubs and clusters would drive roughly 400 billion yuan of investment a year.

The logic is not stupid. AI training is brutally energy-hungry and largely indifferent to a few dozen milliseconds of latency. China has ordered that data centres draw 80 per cent of their power from renewable sources by the end of the decade, and the west is where the wind and the sun and the water are. Simeng Deng of Rystad Energy told AFP that the facilities help absorb the surplus of renewable power generation, which is a real service: western China curtails a great deal of clean electricity it cannot move east fast enough. China is on course to nearly double its data centre capacity within five years.

There is a wrinkle in the clean-power story that deserves stating. Guizhou is a hydropower province, but it is also a coal province. Its grid leans hard on thermal generation, and leans harder when the reservoirs are low. Between 2020 and 2024 the clean share of Guizhou's generation actually fell by five percentage points, one of the steepest declines in China, as fossil output grew faster than total generation. The 80 per cent renewable target is a target, not a description. And the curtailed wind and solar that western data centres are meant to soak up is curtailed partly because transmission out of the west is inadequate, which is the same infrastructure gap that made siting compute there attractive in the first place. The policy is elegantly circular: build the load where the power is stranded, because the power is stranded.

For a province like Guizhou, the pitch to Beijing and to the market was straightforward. We have the power. We have the land. We will build the rest. And it did build the rest, with borrowed money, which is the part of the sentence that ended up mattering most.

The Machine That Employs Almost Nobody

The single most useful piece of evidence on what a hyperscale facility does to the place around it was published this month, and it is not about China at all.

In a working paper dated 7 August 2026, Dany Bahar of Brown University and Greg C. Wright of the University of California, Merced set out to test the spillover claim directly. Their opening line is the thesis: a hyperscale data centre can cost more than a billion dollars while employing only a few dozen people. Using a registry of 341 hyperscale facilities in the United States, assembled from a Pacific Northwest National Laboratory atlas, operator announcements, subsidy records and state filings, they asked whether that investment propagates outward through the three classic channels economists expect: a deeper shared labour pool, denser supplier linkages, and firm clustering with knowledge spillovers.

Their method is elegant. Because operators screen sites for power, land and fibre, the places that get a data centre are systematically unlike the places that do not, which wrecks naive before-and-after comparisons. So Bahar and Wright compare the immediate vicinity of a completed facility against the surrounding area at the same site, and separately compare 341 built campuses against 84 hyperscale projects that were publicly announced and never constructed. Satellite imagery does the dating: land clearing and night-time lights mark the moment construction begins.

At the parcel, the effect is enormous and unmistakable. Night-time lights rise by 38 per cent within the first kilometre when construction starts. Vegetation clears. The place is visibly transformed. And then, moving outward, the signal decays to approximately nothing by five kilometres. Announced projects that were never built show no comparable change, which is the control working exactly as intended.

Beyond the fence line, the findings are a sustained deflation. Advertised salaries do not rise; the authors can rule out any increase above 5.5 per cent. New firm registrations and business applications do not rise, with upper bounds of 1.6 and 5.7 per cent. Supplier job postings rise by 5.9 per cent, but the confidence interval runs from a 15 per cent decline to a 32 per cent increase, which is a polite way of saying the data cannot tell. County-level data-processing employment rises 26 per cent and establishments 27 per cent, but that category includes the facility itself, and related industries show no consistent response. Compute-using firms are indeed found near data centres, but 69 per cent of them were already there before the nearest facility opened. Nearby rents may rise a few per cent, imprecisely. Multifamily permitting does not rise. Net migration does not rise. Foot traffic and commercial spending show no robust change. Residential electricity prices, in their US sample, do not rise either.

The summary sentence is one that ought to be pinned above every county planning committee and every provincial development office on earth: at this scale, the sites are transformed, but there is little evidence of a new local cluster.

That is Guizhou's 7.4 per cent GDP growth and its second-from-bottom wage growth, derived independently, on the other side of the Pacific, from satellites and job adverts.

Two Kinds of Job, and the Gap Between Them

The employment arithmetic is not hidden. It is simply presented in a way that encourages people to add the wrong numbers together.

A hyperscale build is a construction event of genuine magnitude. Industry staffing analyses drawing on the Uptime Institute's 2024 Global Data Center Survey put a 100 megawatt campus at roughly 850 construction workers across an eighteen-month build. These are the industry's own numbers, published by recruiters who profit from the boom, which makes their shape more telling rather than less. That is real money moving through a local economy: rented rooms, diesel, lunch, aggregate, portaloos. It is also, by design, temporary. When the last commissioning engineer drives away, a fully built 100 megawatt hyperscale campus typically retains somewhere between one hundred and two hundred permanent on-site staff.

The gap between those two figures is where the political trouble lives. A community is shown the construction number, experiences the construction number, and then is left with the operations number, which is smaller by an order of magnitude and often filled by specialists who commute or relocate rather than by the people who used to work the mountain.

Guizhou has a particular reason to feel that gap keenly. For three decades the province's most reliable export was working-age adults, sent to the factory belts of Guangdong and Zhejiang, leaving behind the phenomenon that Chinese social policy calls left-behind children. Li Xixiu's observation that villagers can now find work nearby is, in that context, an enormous statement. It is also one that depends on which phase of the project you are standing in. Construction employment for a build-out of dozens of facilities can run for years, and while it runs, it looks like a structural change to the local labour market. It is not one.

Ireland offers the cleanest illustration in the world, because the Irish state publishes both sides. The Central Statistics Office found that data centres consumed 22 per cent of all metered electricity in the Republic in 2024, and 23 per cent in 2025. The Department of Enterprise's own assessment, which dates from 2018, put direct employment at roughly 1,800 people, with a further 1,900 a year in related construction, the latter figure supplied by the Construction Industry Federation. Slightly less than a quarter of a national grid, in exchange for a direct workforce that would fit comfortably into a mid-sized secondary school. The jobs number is eight years older than the electricity number, which tells its own story about what gets counted. The Commission for Regulation of Utilities has rewritten connection policy to favour applicants who bring their own dispatchable generation or storage and can offer demand flexibility. Ireland is not hostile to the industry. It simply ran out of grid before it ran out of enthusiasm.

Loudoun County, and Why Guizhou Cannot Copy It

There is one place where the bargain has unambiguously worked for residents, and it is worth understanding precisely why, because the reason does not travel.

Loudoun County, Virginia, hosts the densest concentration of data centres on the planet. Its fiscal 2027 budget anticipates roughly 417 million dollars in real property tax from data centre buildings and about 879 million dollars in personal property tax on the servers and equipment inside them, nearly 1.3 billion dollars in total, or 45 per cent of the county's nearly 2.9 billion dollars in tax revenue, from a county of about 440,000 people. Industry-adjacent analysis by Mangum Economics for the Northern Virginia Technology Council estimates that without that revenue, residential property tax rates would have to rise by 91 per cent, nearly double, costing a typical homeowner some 5,800 dollars a year. The average completed facility employs about 50 people.

Loudoun is not a story about labour spillovers. Bahar and Wright would predict, correctly, that the wage effects there are muted. Loudoun is a story about a fiscal linkage: a local government with the legal power to tax the equipment inside the building, annually, at high value, and to spend the proceeds on its own schools and roads.

Almost nowhere else has arranged things that way. In the United States, at least thirty-five states now offer tax incentives aimed specifically at data centres, with cumulative awards approaching twenty billion dollars by the authors' tally from the Good Jobs First subsidy tracker. Good Jobs First's own analysis of eleven data centre megadeals found an average public cost of about 1.95 million dollars per permanent job, with the largest single per-job subsidy, 6.4 million dollars, awarded by North Carolina to Apple. The organisation's recommendation was that all state and local subsidies combined be capped at 50,000 dollars per permanent job. Set against 1.95 million, that recommendation reads less like policy advice than like an intervention.

Guizhou's structural problem is that it has neither Loudoun's tax handle nor the option of declining the deal. Chinese local governments do not levy a meaningful recurring property tax. Their revenue historically came from land sales and from off-balance-sheet borrowing through local government financing vehicles, which build the roads and substations and repay the loans out of the growth the roads and substations are supposed to produce. When a province competes for a hyperscaler by discounting power, discounting land and building the grid connection itself, it has converted the fiscal linkage from an asset into a liability before the first rack is energised. The investment lands. The debt service lands. The wage bill, being tiny, lands somewhere between the two and barely registers.

An Enclave With Better Cooling

Development economists have a name for this shape, and it is much older than the cloud.

Albert Hirschman argued in 1958 that the developmental value of an industry lies not in its size but in its linkages: backward, to the suppliers it pulls into existence, and forward, to the industries that add value to its output. In a 1977 essay on staple exports he generalised the scheme, adding the fiscal linkage, meaning the public revenue an industry generates, and the consumption linkage, meaning the local demand created by the wages it pays. An enclave economy is what you get when all four are weak. The classic cases are extractive: a capital-intensive mine or oil field employing very few people relative to its contribution to output, importing its equipment, exporting its product, and touching the surrounding economy mainly through a fenced perimeter and a haul road. UNCTAD's work on extractive industries describes exactly this combination, capital-intensive, labour-light, linkage-poor, as the reason resource wealth so often fails to convert into local development.

A hyperscale data centre is an unusually pure specimen. Its backward linkages are global: the GPUs come from a handful of foundries, the transformers and chillers from specialist manufacturers, the network gear from a shortlist. Bahar and Wright's inconclusive supplier estimates are what you would expect from an industry that buys almost nothing locally except concrete, security and landscaping. Its forward linkages are, by construction, non-local: the entire premise of Eastern Data, Western Computing is that the value-added services consuming the compute stay two thousand kilometres east. Its consumption linkage is capped by a payroll of dozens. And its fiscal linkage is the one variable that policy can actually set, which is precisely why competition between jurisdictions tends to bid it towards zero.

This is not an argument that data centres are bad. It is an argument that they are a particular category of thing, and that the category has a well-documented behaviour which the promotional literature systematically ignores. Guizhou did not misunderstand data centres. It understood them as a growth engine, which they are, and hoped they would also be a development engine, which they largely are not.

The Racks That Nobody Rented

The second Guizhou problem is that a good deal of the capital did not even deliver the compute.

In March 2025, MIT Technology Review reported that of the more than five hundred data centre projects announced across China in 2023 and 2024, at least 150 had been completed by the end of 2024, and that local publications were reporting up to 80 per cent of new computing capacity sitting idle. GPU rental prices collapsed accordingly: an eight-GPU Nvidia H100 server that had commanded around 180,000 yuan a month fell to about 75,000.

The DSET researchers Angela Glowacki and Cartus Bo-Xiang You, writing for the Australian Strategic Policy Institute's Strategist in May 2026, mapped the same problem onto the western build specifically. By 2024, 633 hyperscale and large data centres had been built and made operational under Eastern Data, Western Computing, lifting national computing capacity to 268 exaflops. Some western facilities, they wrote, sit empty, with utilisation rates as low as 20 to 30 per cent, a far cry from the original policy goal of more than 60 per cent. Beijing has since restated that all data centres should run at no less than 60 per cent utilisation, and that no new large or super-large facilities should be built in cities where existing ones operate below 50 per cent.

The reasons are mundane and instructive. Remote regions lacked the fibre-optic cables needed to move large volumes of data in real time, forcing operators to spend more on transmission than the model assumed. Renewable curtailment in the western region still exceeds 30 per cent, which undercuts the cheap-clean-power premise. Many facilities were built on the assumption that state-owned enterprises and government agencies would buy the compute, and that demand did not fully arrive. Inter-governmental competition produced speculative overbuilding, and more than a hundred state-backed projects have been scrapped in the past eighteen months against eleven cancellations in the whole of 2023.

Guian, to be fair, is at the better end of this distribution. Local reporting in August 2026 put first-half electricity consumption growth in the new area at 33.7 per cent year on year, on a total of 3.12 billion kilowatt-hours through late June, with big data operations alone consuming 1.93 billion of that, up 52.2 per cent. Guian is busy. But an idle rack and a busy rack impose the same debt service, and a province cannot know which it has bought until several years after it has paid.

Who Actually Pays the Bill

If the benefits are concentrated at the parcel and diffuse to nothing by five kilometres, the costs run in the opposite direction. They start at the parcel and travel.

A 2026 arXiv preprint by Danbo Chen, Zijun Zhou, Yongyang Cai, Jiahong Qin, Ani Katchova and Lei Chen models this directly, coupling language-model analysis of corporate compute plans with energy-system modelling. It projects that electricity consumption by the six largest AI firms will rise from roughly 118 terawatt hours in 2024 to between 239 and 295 terawatt hours by 2030, about one per cent of global power demand, with more than 90 per cent of new capacity landing in North America, Western Europe and Asia-Pacific. Crucially, the burden is not evenly distributed. The authors construct a Power Stress Index and find values above 0.25 in Oregon, Virginia and Ireland, while diversified grids in Texas and Japan absorb the load more comfortably. Their conclusion is that AI infrastructure has become a structural component of power-system dynamics rather than a marginal load, which means it now has to be planned for rather than merely connected.

Water follows the same logic. A preprint by Yuelin Han, Pengfei Li, Adam Wierman and Shaolei Ren, revised in March 2026, estimates that if 2024 water-use intensity persists, US data centres could collectively require between 697 and 1,451 million gallons per day by 2030, comparable to New York City's entire supply. Even assuming aggressive efficiency gains of 10 per cent a year, the range is 227 to 604 million gallons daily. Associated public water infrastructure costs reach roughly ten billion dollars, rising to fifty-eight billion under high-growth scenarios. Their central observation is the one that matters here: these impacts are highly concentrated on communities hosting data centres. The compute is national. The reservoir is not.

Memphis has become the American shorthand for what concentration looks like when it goes wrong. The NAACP, the Southern Environmental Law Center and Earthjustice sued xAI in April 2026 over the operation of 27 unpermitted methane gas turbines in Southaven, Mississippi, effectively a power plant assembled to feed the Colossus 2 facility. This is an airshed where Shelby County in Tennessee and DeSoto County in Mississippi have both received an F grade for ozone from the American Lung Association. The plaintiffs include residents of the Whitehaven and Boxtown neighbourhoods of South Memphis, downwind. The Department of Justice has since intervened on xAI's side. The turbine count, meanwhile, never stopped rising. The plaintiffs went back to court on 6 May 2026 seeking an emergency order to halt operations, and the installations continued regardless: by mid-July, correspondence between xAI's environmental consultant and Mississippi regulators, obtained by Reuters through a public records request, documented 59 unpermitted turbines, at least 57 of them at Southaven, roughly double the number the company had publicly acknowledged. The resolution took the most direct form available: Mississippi's Permit Board had already approved a permanent 1.2 gigawatt plant of 41 turbines on the same ground in March 2026, a month before the suit was filed, and on 31 July xAI agreed a schedule to strip the temporary machines out of the Stanton Road site, beginning in August 2026 and finishing by July 2027. The fight over unpermitted temporary turbines has been settled by making the power station permanent and lawful in the same airshed, breathed by the same people, which is the tell that it was never really about permits. Whatever the litigation concludes, the geography of the dispute is the point: the model is trained everywhere and the generation sits in one postcode, with permission now to stay there.

What Happens When You Ask

There is a final piece of the research picture, and it is a strange one, which is what makes it interesting.

In a preprint first posted in November 2025, Zhifeng Wu, Yuelin Han and Shaolei Ren asked whether large language models could stand in for community consultation on data centre projects. They built a framework that polls AI agents, prompted with local demographic and geographic context, on how they would respond to a proposed facility, and compared the output against real human survey data. The agents identified water usage and utility bills as the dominant concerns and tax revenue as the principal perceived benefit. Responses varied meaningfully depending on which model was used and where the hypothetical project sat. And, notably, the synthetic responses aligned substantially with findings from actual human surveys. The authors propose it as an efficient early-stage instrument for folding neighbourhood perspectives into siting decisions before the plans harden.

You can read that finding two ways, and both are uncomfortable. The optimistic reading is that we now have a cheap way to anticipate what a community will object to, months before the first hearing, at a stage when the design can still change. The bleak reading is that the industry has arrived at a technique for simulating consent, and that the reason such a technique is attractive is that the genuine article is expensive, slow and increasingly likely to say no.

The Guizhou villagers quoted by AFP were not polled, synthetically or otherwise. They were asked a question by a passing reporter and answered it honestly, which is a different exercise from being consulted before a decision. Nothing in the Eastern Data, Western Computing framework required anyone in Guian to be asked whether the mountain should become a campus, and it is worth being clear that this is not solely a feature of the Chinese system. Across the United States, the same decision is routinely taken under non-disclosure agreements and by-right zoning, and communities learn what has been approved after the approval.

Note also what the agents converged on. Water. Bills. Tax revenue. Not wages. Not careers. Not the long-run transformation of the local labour market. Even a synthetic public, prompted to reason about a data centre, does not appear to expect it to be a jobs programme. The expectation gap that Guizhou is living through is largely one that promoters created and that residents, given a moment to think, do not fully share.

What Shu Peihua Is Describing

So return to the shopkeeper on the road outside Huawei's clock tower, because nothing in the preceding sections makes what Shu Peihua said untrue.

The road is real. In a karst province where a mountain village might once have been two hours from a trunk route, a dual carriageway built to carry transformers and chilled water plant is a permanent improvement to the lives of everyone along it. The universities are real; a local vendor told AFP that two had been established since the facilities went in. The customers are real, and so is Li Xixiu's point that people find work nearby now instead of boarding a train to Guangdong. Guizhou has been one of the great exporters of migrant labour in modern China, with all the social cost that implies, and a family that stays together because a parent can get a security job or a canteen job or a fit-out job forty minutes from home has received something that does not show up in a wage-growth ranking.

What the evidence says is narrower and harder. It says that these gains are the consumption linkage of a construction boom plus the ordinary agglomeration of a new-town development, and that they are front-loaded. It says the operational phase which follows will not employ many people, will not raise local salaries measurably, will not spawn a cluster of firms that were not already coming, and will not, on the American evidence, move rents, migration or retail spending much either. It says the electricity, water and land are consumed locally while the value of what they produce is realised somewhere with better weather and higher salaries. And in Guizhou's case, it says the province financed the entry ticket with debt that now ranks second-worst in the country relative to revenue, against assets that may be running at 20 to 30 per cent utilisation.

None of that argues for refusing the data centre. It argues for pricing it honestly, and for noticing which linkage is doing the work. The fiscal one is the only channel a host can reliably control, and it is the one that inter-jurisdictional competition destroys first. There are policy shapes that hold onto it: recurring taxation of the equipment rather than one-off land revenue, as in Loudoun; subsidies capped per permanent job, as Good Jobs First proposes; published utilisation and load data so that a province can tell a productive asset from a monument; large-load tariffs that make the operator, not the household, pay for the substation; and sunset clauses that return the abatement when the promised employment does not materialise.

The alternative is what the resource curse literature has been documenting for sixty years in copper and oil, now rendered in reinforced concrete and immersion cooling. Something enormous arrives. The output figures move. The mountain gets a road, and then the road gets quiet, and the ledger that recorded the growth turns out never to have been the ledger that measured the prosperity.

Shu Peihua is right that it used to be a barren mountain. The question Guizhou has yet to answer, and that a hundred counties from Virginia to Kildare are asking in their own accents, is what a mountain becomes when the thing built on it needs the mountain far more than the mountain needs it.

References and Sources

1. China goes rural with data centres in quest to power AI – AFP, via The Star, 19 August 2026
2. Abundant electricity isn't enough: China's overbuilt AI computing power is underused – Angela Glowacki and Cartus Bo-Xiang You, The Strategist, Australian Strategic Policy Institute, 6 May 2026
3. The Local Economic Impact of Data Centers – Dany Bahar and Greg C. Wright, working paper, 7 August 2026
4. Concentrated siting of AI data centers drives regional power-system stress under rising global compute demand – Danbo Chen, Zijun Zhou, Yongyang Cai, Jiahong Qin, Ani Katchova and Lei Chen, arXiv preprint 2604.06198
5. Small Bottle, Big Pipe: Quantifying and Addressing the Impact of Data Centers on Public Water Systems – Yuelin Han, Pengfei Li, Adam Wierman and Shaolei Ren, arXiv preprint 2603.02705
6. What AI Speaks for Your Community: Polling AI Agents for Public Opinion on Data Center Projects – Zhifeng Wu, Yuelin Han and Shaolei Ren, arXiv preprint 2511.22037
7. China built hundreds of AI data centers to catch the AI boom. Now many stand unused. – MIT Technology Review, 26 March 2025
8. Western regions to gain computing power – China Daily, 17 March 2022
9. Big data key to high-tech development – China Daily, 30 May 2023
10. China's debt-ridden Guizhou faces reckoning after years of splashing out on pricey projects – South China Morning Post
11. AI and Data Centers Surge in Guian as First-Half Electricity Demand Jumps 33 Percent – ChinaTechNews, 20 August 2026
12. Chinese government plans data center capacity reseller network amid overbuild concerns – Data Center Dynamics
13. Data Centres Metered Electricity Consumption 2025, Key Findings – Central Statistics Office, Ireland
14. Government Statement on the Role of Data Centres in Ireland's Enterprise Strategy – Department of Enterprise, Trade and Employment, Ireland, June 2018
15. CRU Publishes its Decision on New Electricity Connection Policy for Data Centres – Commission for Regulation of Utilities, Ireland
16. Loudoun County, Virginia: The Heart of the Data-Center Boom – Judge Glock, City Journal, 26 April 2026
17. The Impact of Data Centers on Virginia's State and Local Economies, 6th Biennial Report – Mangum Economics for the Northern Virginia Technology Council, February 2026
18. Money Lost to the Cloud: How Data Centers Benefit from State and Local Government Subsidies – Good Jobs First
19. A Generalized Linkage Approach to Development, with Special Reference to Staples – Albert O. Hirschman, 1977, reprinted in The Essential Hirschman, Princeton University Press
20. Extractive Industries: Optimizing Value Retention in Host Countries – United Nations Conference on Trade and Development
21. Illegal Pollution from Data Center Power Plants Shouldn't Harm Our Communities. We're Suing xAI. – Earthjustice
22. Pollution from Musk's unpermitted xAI power project hits hardest in Black communities – Disha Raychaudhuri and Valerie Volcovici, Reuters, via The Star, 14 July 2026
23. State sets dates to retire temporary xAI turbines but allows some to go past original deadline – Mississippi Today, 31 July 2026
24. China's north cleans up its power mix as the south lags – Centre for Research on Energy and Clean Air, 19 March 2025
25. How Many Jobs Does a Data Center Create? – Data Center Geeks, compiling the Uptime Institute 2024 Global Data Center Survey

---

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...