Everything About AI
NextFin News - The artificial intelligence buildout that has drawn nearly a trillion dollars of corporate spending into a single year is running into a wall that no amount of capital can buy around: local communities do not want data centres in their neighbourhoods, and their politicians are starting to listen. What began as scattered protests over noise, water and electricity bills has hardened into a political force that is canceling projects, rewriting state rules and threatening the economics of the AI infrastructure boom itself.
The scale of the collision is now measurable. At least $260 billion of data centre investment was canceled this year after sustained local opposition, according to research data published in October, with roughly half of that — about $130 billion — wiped out in the three months ending September 30. Around $1 trillion of planned investment now faces some form of organized local opposition, and about half of the projects that have met backlash this year have been canceled. In the second quarter alone, 45 projects worth $68 billion were blocked or delayed, according to research group Data Center Watch.
The central tension of the AI boom is no longer whether the technology works. It is whether the physical and political infrastructure required to run it can be built at the speed the models demand — and whether the same executives who spent years warning that AI could displace millions of jobs can persuade anyone that the factories housing those models belong in their communities.
The Numbers Behind the Buildout
The spending commitments are historically large. J.P. Morgan Global Research revised its estimate of AI-related capital expenditure by the five American hyperscalers — Amazon, Alphabet, Meta, Oracle and Microsoft — upward to $900 billion for 2026, from $736 billion earlier in the year, and projects global AI capex reaching $5.5 trillion by 2030. "The surge in AI capex-related expenditures has been the defining feature of our economic times," the bank's chief U.S. economist said. Other trackers put the 2026 figure for the five hyperscalers in the $660 billion to $690 billion range, still nearly double 2025 levels.
That money is buying a physical footprint measured in electricity. The International Energy Agency models U.S. data centre electricity consumption at 182.61 terawatt-hours in 2024, rising to well over 400 TWh by 2030 in its base case — more than tripling from 2021, the last full year before ChatGPT. The Electric Power Research Institute's 2026 scenarios are steeper: data centres could draw between 383 TWh and 793 TWh by 2030, taking their share of total U.S. electricity demand from 4-5% today to 9-17%, with some states exceeding 40%.
These are not abstract grid figures. They translate into substations, transmission lines, transformers and water for cooling — all of which need permits, and all of which need neighbours who will tolerate them.
The Political Backlash Is Bipartisan — and Local
Public opinion has turned with unusual speed. A Gallup poll published in May found that 71% of Americans opposed having a data centre in their area, including 48% who opposed it strongly. For context, opposition to a nuclear power station in the same poll stood at 53%, including 34% strongly opposed. A poll published in August found that 75% of Americans would oppose a data centre being built near where they live — a sharp shift from roughly a year earlier, when the country was about evenly split.
The opposition is not confined to one party or one region. Democratic Senator Bernie Sanders has called for a moratorium on new construction.
"People are looking around worried about how quickly the technology is moving, how little control they have over it, how will data centres impact their community in terms of electric rates and water utilisation."
But the pressure is bipartisan. In New Jersey and Virginia, newly elected Democratic governors pushed through legislation compelling new energy tariffs for data centres and disclosure of their energy and water use. In Texas, the most pro-business data centre state, Republican Senate candidate Ken Paxton has campaigned against data centre tax breaks and proposed criminal liability for operators whose AI chatbots "undermine children's safety."
Thirty state legislatures have now implemented rules governing data centre location and resource consumption, according to Data Center Watch. In July, 142 AI data centre protests were staged simultaneously across 42 states in a single weekend. The research group counts 843 opposition groups across every U.S. state except Hawaii.
The UK is not immune. Britain has the third-highest number of data centres in the world, one place ahead of China, and opposition has surfaced from Buckinghamshire and Berkshire to Brick Lane in east London, where plans for a facility likely to serve the City's high-frequency traders have met resistance from residents who would rather see social housing.
The Optics Problem: Jobs Warnings Meet Job-Light Facilities
There is an awkward asymmetry at the heart of the backlash. The same industry leaders who have spent years warning that AI will transform the labour market are now asking communities to accept large industrial facilities that employ very few people. A functioning data centre tends to require just a few dozen permanent staff, with a few hundred at the very largest new sites — plenty of construction work, little lasting local employment.
That dynamic is hard to sell when the companies building the centres are themselves valued in the trillions and are simultaneously telling investors that AI will reshape white-collar work. Anthropic chief executive Dario Amodei once said AI could replace half of white-collar jobs; more recently he has described it as a job multiplier. OpenAI chief Sam Altman has said this year that AI is unlikely to cause a "jobs apocalypse." Both men have softened their earlier warnings, but the earlier warnings traveled further.
The credibility gap widened at the United Nations in September. Altman told the Security Council:
"We could lose control of the future to AI. The risk is that it moves so fast that people can no longer follow what is happening or intervene when needed."
Amodei said: "If managed poorly, I even believe AI could be a risk to humanity as a whole." Amodei has also acknowledged that AI faces a crisis of trust and must produce a breakthrough — curing cancer is the example he and others invoke — to win the public back.
The Industry's Response: Pledges, Transparency and a Divided Front
Tech leaders know they have a persuasion problem. In March, seven Big Tech chief executives stood with President Trump to sign a non-binding "Ratepayer Protection Pledge," promising to "build, bring or buy" their own energy and ensure costs are not passed on to ordinary households. Several large tech firms have announced they will no longer use non-disclosure agreements with local officials over data centre construction, a practice that had fueled suspicions of backroom deals.
Jensen Huang, chief executive of Nvidia, the chipmaker at the centre of the buildout, has taken a more confrontational line. In August he argued on social media that data centres are the essential agent of U.S. reindustrialisation:
"AI is bringing manufacturing back to America and reindustrializing the nation after decades of offshoring. AI is creating demand that drives investment in our aging power grid and sustainable energy, powered by market forces, not subsidies."
Speaking in a television interview, he added: "The technology industry, the builders, and, have to do a much better job working with the communities to get prepared for this AI infrastructure build-out."
The response has not been uniform. After Huang made his reindustrialisation argument, other Big Tech chief executives did not follow with comparable public defenses — a sign, observers noted, of difficult optics. The lesson from decades of infrastructure politics is that national economic benefits rarely outweigh local costs in the minds of the people asked to bear them.
Why This Is Structural, Not Cyclical
It is tempting to read the cancellations as a cyclical pause — a financing or permitting bottleneck that clears once interest rates fall or templates standardize. That reading is wrong, or at least incomplete. The binding constraint is not capital; it is consent.
Three features make this structural. First, the veto point is local. Data centres need specific sites with specific grid connections, and a single county board or town council can stop one. National pledges signed at the White House do not bind a zoning commission in New Jersey. Second, the grievance is tangible and immediate: higher electricity bills, strained water supplies, noise from cooling fans and transformer hum. These are felt within months, while the promised benefits — cheaper power, reindustrialisation, medical breakthroughs — are diffuse and distant. Third, the opposition has already built the infrastructure of a movement: 843 groups, coordinated protest days, state-level legislative playbooks.
The cyclical element is real but secondary. If financing conditions ease, some deferred projects will resume. But easing credit does not make a substation less visible or a cooling tower quieter. The projects that survive will be the ones that can demonstrate local benefit and absorb higher connection costs — not necessarily the ones with the best models.
The Second-Order Consequence: AI Gets More Expensive, and Slower
The first-order effect of the backlash is canceled projects. The second-order effect is what it does to the economics of AI itself. Every delayed interconnection, every new state tariff, every disclosure requirement adds cost and time to the buildout. That raises the marginal cost of compute, which raises the cost of training and running models, which narrows the set of AI applications that are economically viable.
There is a further transmission channel. If the buildout slows, the revenue assumptions underpinning the chip cycle slow with it. Nvidia's data centre revenue reached $89.0 billion in the second quarter of fiscal 2027, up 117% from a year ago — a number that assumes the hyperscalers keep building. A structurally slower buildout does not just delay capacity; it forces a repricing of the entire supply chain that was sized for uninterrupted growth.
Conversely, the backlash could accelerate the industry's shift toward sites and technologies that are less visible and less resource-intensive: brownfield locations, behind-the-meter generation, more efficient cooling, and models that do more with less compute. Constraint breeds innovation — but it also breeds consolidation, favouring the largest players who can absorb permitting risk.
The Counter-Thesis: The Boom Is Too Economically Entrenched to Stop
The strongest case against the backlash thesis is simple: the economic incentives are too large, and the buildout too far advanced, for local opposition to do more than nibble at the edges. Amazon, Google, Meta and Microsoft have already spent more than $1 trillion on AI infrastructure since 2023, with hundreds of billions more expected this year. National governments view AI capacity as a strategic asset in competition with China, which accounts for a large share of global data centre electricity growth alongside the United States. States that have benefited from data centre investment — Texas, Virginia, Iowa — have powerful reasons to keep welcoming it.
There is force to this view. Moratoriums are often temporary by design, buying officials time to write rules rather than to ban development outright. Once standardized tariffs and disclosure regimes are in place, the political temperature may fall and projects may resume. Policy analysts note that several large tech firms have already dropped NDAs and that executive-branch policy is pushing developers to cover their own energy costs — precisely the kind of accommodation that defuses local anger.
But this counter-thesis underestimates the asymmetry of the fight. A developer must win every permit; opponents need only win one. And the grievances are compounding, not fading: as data centres' share of local electricity demand climbs toward double digits in some states, the bill impact becomes harder to ignore.
What to Watch
The falsifying signal for the structural-bottleneck view is specific: if states that enacted moratoriums begin lifting them after standardized tariff and disclosure regimes take effect, and if hyperscaler capital expenditure guidance for 2027 holds above $800 billion while grid interconnection queues clear on schedule, the thesis weakens materially. The confirming signal is equally concrete: if 2027 guidance is cut, or if more than $300 billion of additional projects are canceled in the first half of 2027, the bottleneck is structural and binding.
Short term, watch the November midterm elections: control of both houses of Congress is at stake, and data centre policy has become a local litmus test in swing districts. Medium term, watch state tariff rulings in New Jersey and Virginia — they will become templates. Long term, watch the share of U.S. electricity consumed by data centres: the IEA's base case has it more than tripling by 2030, and EPRI's high scenario has it approaching a fifth of national demand by 2035. At those levels, every new project is a political decision, not just a commercial one.
The AI buildout was always going to be a test of engineering. It is turning into a test of consent — and the industry that warned loudest about the risks of moving too fast is discovering that the public may hold it to its own warnings.
Explore more exclusive insights at nextfin.ai.
