Prologue

Next week, I will be standing in a town of around seventeen thousand people wedged between a Chevron refinery and the southern runways of LAX, meeting with some of the most important companies in the American hard-tech renaissance. One builds nuclear reactors on a production line, the way Ford built cars. One builds the autonomous weapons that broke the primes’ fifty-year monopoly on the Pentagon, and is now worth more than most of them. One is an accelerator that mints the next cohort, a week in a warehouse that has already produced companies backed by the biggest venture firms in America. This essay is about the system that produced them, and how Britain could build a similar, if not better system but is instead preparing, with great ceremony as per, to fumble it.

The town is El Segundo, Los Angeles County. The founders who moved there call it The Gundo. It covers five square miles, two of which are refineries. In the early days, they burned the week’s packing pallets in bonfires on Friday nights, lived in each other’s factories, and texted each other at two in the morning to check who was still on the floor. The operating system of the place is a group chat: do you have this tool? do you have a vendor? do you have a freight forwarder? do you know someone at this agency? Walk down Main Street and you will bump into somebody who has put hardware into space. The companies compete for the same engineers, watch each other’s shipment numbers in the Wall Street Journal, and describe the resulting paranoia as the best thing that ever happened to them. If Silicon Valley is America’s safe haven for software firms, El Segundo is the equivalent for hardware. It is the beating heart of America’s reindustrialisation revolution.

El Segundo, the Florence of Hard-Tech

A traditional Gundo Bonfire, June 2024

None of this was an accident. One of the founders ran his site selection like Billy Beane, every column in his spreadsheet measuring proximity: more and better hardware talent than anywhere else in America, warehouse space priced for a pre-seed company, technicians available from the primes and the refinery, universities close enough for the science hires, LA money close enough to walk into. In true Moneyball-style, he was scoring what the market wasn’t pricing: distance.

So was the town itself, decades before he arrived. El Segundo grew on the rotting bones of dead Cold War aerospace. Solid rocket boosters for the space shuttle were tested inside one of the buildings these startups now occupy, the warehouses were designed for manufacturing because manufacturing is what they were for, and the workforce never fully dispersed when the contracts did. When the new generation showed up, everything they needed was already on their doorstep. People look at the Gundo and see a portfolio of companies. Look again. The product is density.

Now look at Britain. We are currently pointing more public money at the physical economy than at any time in living memory: a £298 billion Defence Investment Plan, a clean power buildout remaking the grid, AI Growth Zones chasing compute, munitions factories reopening, artillery production back onshore for the first time in a generation. Sovereignty is the word every one of these programmes now reaches for, sovereign capability, sovereign compute, sovereign supply chains, and all of it means the same thing: hard things, built here, by firms we control.

Make no mistake: sovereignty is not a niche enthusiasm of the tech-bro commentariat. When Focaldata ran a conjoint experiment last year, testing which policies actually move votes rather than merely poll well, the sovereignty cluster came out on top. Limiting foreign ownership of critical digital infrastructure, setting independent UK technology rules, requiring British data to stay onshore and giving preference to UK firms in government contracts all outranked the supposed heavyweights: the NHS, immigration, the economy, crime. Voters could not always name the concept, but the instinct beneath it, that Britain should own the things it depends on, was among the most powerful we measured.

If you want a picture of what that instinct is reacting to, consider the news that broke yesterday as I was writing this. The Royal Navy’s new K3 Scout drone boats, built by the British firm Kraken and operated by the Royal Marines since March, were found to contain camera components that were quietly transmitting “heartbeat” signals to an IP address in China. Kraken had bought the cameras from a third-party supplier who assured them the parts were secure. Nobody caught it until a routine cyber assessment did. The MoD insists no operational data was compromised, and it may well be right. But the deeper point survives either way: a country that does not build its own critical components does not get to know what they are doing. Sovereignty is more than a slogan.

Unfortunately, we are about to make one specific, predictable, very expensive mistake with these record levels of public investment. The mistake of fairness. Every region will get its own cluster, every mayor will get their own press release, and the money will be spread across the map like Lurpak until no single place has enough of it to matter. Britain has institutions dedicated to this regenerative mistake and it essentially is the one thing the Civil Service somehow reliably delivers on time and on budget.

The thesis of this essay is a fairly simple one. The land exists; the money exists; the ambition exists. What’s missing is putting them all in one place.

Part I: The magic is the cluster

£160 million concentrated in one square mile beats £160 million spread across a region. The concentrated version produces a cluster and the distributed version produces a grant programme. These are different objects with different physics. Most British regional policy assumes the opposite, treating economic development as if it were gravy that you can pour evenly across a map. Fifty years of enterprise zones, growth deals and regional funds say otherwise. And so does every hour of footage from the miracle that is El Segundo.

Why? Because a cluster is not a collection of firms. It is a set of compounding effects that only switch on above a certain density. Here are five of them.

The first is shared capital equipment. No pre-seed hardware company can justify a five-axis machine shop, an EMC chamber or a proper test rig on its own balance sheet. Twenty firms within walking distance can, and in the Gundo they effectively do, through an informal borrowing economy that works as a distributed factory. It is the group chat again, every “do you have this tool” a capital expense one startup avoided because a neighbour had already made it. Spread the same twenty firms across a region and each one buys, rents or goes without. The cluster’s equipment bill is a fraction of the diaspora’s, and the difference goes into engineering.

The second is labour-market liquidity. In a dense cluster, engineers change firms without changing homes, which means the talent pool deepens with every hire rather than fragmenting with every departure. When a startup dies, and startups die, its team is absorbed by the companies next door within weeks instead of leaving hardware, or leaving the country. Hiring risk collapses for everyone at once, because joining any one firm becomes a bet on the neighbourhood rather than on a single cap table. The Gundo did not conjure this workforce from nothing: the primes and the refinery had spent decades training technicians within a few miles of each other, and the startups moved into the middle of that labour pool the way you’d build a pub next to a brewery.

The third is supplier and first-customer proximity, which turns iteration loops from procurement cycles into afternoons. When the machinist is down the road, a design change costs hours; when the first customer is down the road, a demo costs a van. Proximity cuts both ways, though. One Gundo company whose entire pitch is making parts fast spent eight months unable to deliver because it could not source specialty alloy ingot in America at any price. Another watched a demonstration in front of senior military officers collapse because a supplier three tiers down had swapped in a propeller with a fatigue flaw. The best hard-tech cluster in the Western world is only as strong as its materials layer, something Britain’s regional investment policy almost always ignores.

The fourth is knowledge spillover, which economists have known about since the days of Alfred Marshall (and later Krugman, Porter et al) who wrote that in an industrial district, the secrets of the trade are “in the air.” Tacit knowledge doesn’t travel in Notion URLs between cities, but in overheard conversations down the pub, in walking through a neighbour’s factory at midnight, in the accumulated scar tissue of people who have already made your next mistake. The founders of El Segundo put it better than any literature could in a podcast from last year: there are people here who know how to see around the technical corners. You cannot email that. You can only stand near it.

The literature around this models knowledge spillover as a weighted sum of what everyone around you knows and leaves the weight open. Different studies fill it in with technological similarity, supply-chain links, shared labour, or distance (Frontier Economics, 2022; after Griliches, 1979). The El Segundo case studies tell us that for the tacit knowledge that actually matters here to effectively spread, nothing matters more than distance. So write the weight as a sharp decay in how far apart two firms sit, and you get:

S = Σⱼ e^(−λdⱼ) Rⱼ

where S is the total useful knowledge firm j absorbs, Rⱼ is how much each nearby firm knows, dⱼ is how far away it sits, and λ is how fast knowledge dies with distance.

In plain English: the knowledge you absorb is the sum of what everyone around you knows, discounted by how far away they sit, and the discount is absolutely ruthless. Spillover doesn’t gradually fade as firms spread out, it just collapses. You can even watch λ get measured in the wild. When researchers tracked patent citations across the UK, the odds of one firm building on another’s work roughly halved between the 0-to-25-mile band and the 25-to-50-mile band, then flattened to nothing. And patents are the easy case, codified, written down, portable. The tacit stuff, the see-around-the-corners knowledge, decays much faster.

The fifth makes the other four compound: the ambition flywheel. When the founder down the street is shipping hundreds of drones a week and the one around the corner is assembling nuclear reactors, your own definition of a productive month recalibrates whether you like it or not. On the same podcast, one of the Gundo founders said: your brothers are building for your country faster than you feel like you are. That pressure’s most measurable symptom is migration. Ask these founders what share of their teams relocated specifically to join the neighbourhood and the answers run from a third to two thirds. Talent moves for density. Nobody truly special moves for a levelling-up strategy.

What ties these all together is non-linearity. Double the spacing between the same twenty firms and you don’t get half a cluster; you get twenty isolated companies and a nice map for the annual gov.uk report. Dispersion doesn’t just dilute the returns to clustering, it forfeits them entirely. Which is why “a bit for everyone” is the one allocation strategy guaranteed to produce nothing, and the one Britain is regrettably reaching for.

Part II: Britain already has every ingredient (in separate boxes)

Britain has already manufactured every component needed for a successful cluster. Parliament has legislated them. The Treasury has funded them. And MHCLG publishes progress reports on them twice a year.

Britain operates 22 Industrial Strategy Zones: 10 Investment Zones, 9 Freeports, and 3 places running both. The Freeport instrument set is actually somewhat decent once you lay it out in one place. £25 million in seed capital per Freeport to remediate land and connect sites. Tax reliefs running to September 2031 in England and 2034 in Scotland and Wales: enhanced structures and buildings allowances, 100% first-year relief on plant and machinery, employer National Insurance relief on every new hire, full business rates relief. Customs sites offering duty deferral and duty inversion, now an optional bolt-on rather than a requirement. And a 25-year retained business rates settlement, which lets the combined authority keep the rates growth its zone generates and borrow against it today. English Freeports alone are already projected to retain more than £880 million on this mechanism. Investment Zones run a parallel offer, up to £160 million over ten years per zone, splittable however the local area likes between tax relief and flexible spend, so a mayor can point it at whatever the cluster actually needs.

Industrial Strategy Zones Action Plan, 2025

Does this work? Sure, for a certain kind of company. The government’s programme report from last year tallies £6.4 billion in private investment landed on Freeport sites, 89% of it FDI, 2,671 hectares of derelict land remediated, £1 billion in private capital unlocked for site preparation alone. If you read the case studies, you’ll notice a pattern begins to emerge. A Korean conglomerate’s wind turbine plant, a Japanese cable factory, a German chemicals expansion. The zones have become excellent at one thing: landing the branch factory of a large incumbent multinational. Nowhere in the programme is anything built for the other species of company, the pre-seed founder with a warehouse and a group chat, the species the entire American hard-tech renaissance is made of.

Well, almost nowhere. Buried in the programme report, one zone is quietly running the proof of concept for this entire thesis. In Plymouth and South Devon Freeport, a marine-autonomy SME leased a facility in the South Yard tax site and, two years later, delivered the Royal Navy’s first unmanned submarine. A few doors down, a prime converted an underused building into an armoured vehicle production line, sourcing half the supply chain from the surrounding region. That is the cluster mechanism in miniature - a shared site, resident customer, a startup and its supply chain condensing around an anchor. It works. It is also subscale, one accidental experiment out of twenty-two zones, and parked on a peninsula at the far end of the South West.

Plymouth and South Devon Freeport

Meanwhile the coordination layer is being softly attempted elsewhere. In June 2026, four Yorkshire mayors launched a defence cluster built on some fairly serious assets: the new artillery factory in Sheffield, Forgemasters, the AMRC, and a £50 million South Yorkshire Defence Growth Deal. What did this big launch actually contain? Early access to MoD contracts, secured by memorandum. And that was it. No tax sites, no shared infrastructure, no equity, and the lot spread across four mayoralties, a hundred miles end to end. For a single number that captures the institutional weight Britain currently assigns to cluster-building, the 2025 Freeports Programme Report supplies it: the average English Freeport delivery team is 6.5 full-time staff. Six and a half people per zone, to do the job that took El Segundo sixty years, a refinery and the entire Cold War aerospace complex. Great, we’ll catch them in no time chaps.

Now, the money. The Industrial Strategy Zones Action Plan commits to a mayoral recyclable growth fund, live from the 2026/27 settlement, devolving Financial Transactions funding to mayors in the North and Midlands. This matters because Financial Transactions capital cannot be granted away. By its accounting rules it must be deployed as loans or equity, and the repayments recycle back into the fund. So, in short, a mayor now holds a permanent, returns-generating investment vehicle, restricted by design to taking stakes rather than writing blank cheques, sitting inside the integrated settlement. It is the closest thing Britain has ever devolved to a regional sovereign wealth fund. But nobody has pointed one at hard tech.

There is some supporting artillery for all this stuff too. A Strategic Sites Accelerator backed by over £600 million, selecting its first wave of sites within the year. The National Wealth Fund, £27.8 billion, with a standing published commitment to work with every Industrial Strategy Zone on access to finance. UK Export Finance with £80 billion in capacity and regional Export Finance Managers being deployed into exactly these clusters. The British Business Bank’s new Cluster Champions programme plus £100 million through its Nations and Regions Investment Funds. Mayoral Development Orders arriving through the devolution framework, letting a mayor grant planning consent across a zone rather than plot by plot. A NESO connections accelerator that jumps high-value demand projects up the grid queue. And, sitting in section 3.2 of the Action Plan, a stated aim to develop regulatory sandboxes within clusters, enabling experimentation and market entry for novel technologies. It is this tucked-away commitment that could let a British zone, under the right conditions, beat El Segundo at its own game. More on this later.

The government has also already invented a sector-specific zone class. When it decided data centres were strategically critical, it created AI Growth Zones, a bespoke designation with streamlined planning and enhanced resource access. So the principle that a strategic industry warrants its own zone type is conceded and published. It is also, like almost every other thing the government tries to do, failing. Fifteen months in, five zones from Oxfordshire to Lanarkshire have produced one newly approved data centre between them, with most claimed investment predating designation. A sector-specific zone works only if you build one, fill it, and plumb it.

So the shelf is stocked. A twenty-two-zone chassis with a decade of tax relief and a borrowable 25-year revenue stream, a working proof of concept in Plymouth, mayors holding recyclable equity capital from next year, and planning, grid, sandbox, export-finance and wealth-fund powers stacked behind them, with a £298 billion Defence Investment Plan, a national grid buildout and a compute race blowing at the back of all of it. Every instrument exists, and that is exactly the trap. A shelf full of instruments looks like a strategy, which is why nobody in the system feels the absence of one.

Part III: How To Make a Hard-Tech Cluster

Three things separate one of Britain’s press-release clusters from a Gundo. None of them appear in any programme the country currently runs. And all of them are cheap!

Ingredient 1: Regulatory and materials plumbing. This is the agglomeration subsidy in physical form, which El Segundo inherited free from dead aerospace. There’s quite a few layers to this and we’re not even halfway through this essay so feel free to caffeine up now.

The first layer is where Britain can do something genuinely new, because it is the layer where even El Segundo failed: the best hard-tech cluster in the Western world cannot test where it builds. The drone companies that define the place sit under the LAX approach and conduct their flight testing somewhere else; one moved an entire team to another state for six months purely to find airspace deregulated enough to work in. The nuclear company got its first regulatory approval in the Philippines, because turning on a reactor was easier at the edge of the world than anywhere in America, and is only now switching one on in Utah. The mining company drills its test holes a three-hour drive past the middle of nowhere, Nevada. The weather company has been banned outright by two states. Add it up and the single biggest constraint on American hard tech is not capital or talent. It is regulatory geography.

Britain has already written the fix for this into policy without realising it. As I highlighted a few paragraphs ago, the Industrial Strategy Zones Action Plan commits in black and white to developing regulatory sandboxes within clusters, to enable experimentation and market entry for novel technologies. Take that commitment literally and put the regulators inside the perimeter: segregated trial airspace over and adjacent to the zone with CAA trial authorities on site, an ONR pre-licensing pathway for advanced reactor work, the Environment Agency permitting run from within the zone, standing range and drilling access on adjacent land. The firm building an autonomous system in the morning flies it in the afternoon; the firm building a reactor demonstrates to its regulator across the road. A British zone designed around permission to test would hold an advantage El Segundo cannot buy at any price.

The second layer is defence. A hardware SME that wins its first defence contract today begins a facility security clearance process that routinely runs twelve to eighteen months, during which it cannot hold classified material, cannot host the customer properly, and burns runway waiting for paperwork. So pre-clear the zone. Sponsor the tax sites to a facility clearance baseline as public infrastructure, with vetted-access buildings, accredited IT and cleared site management already in place, so that an incoming firm inherits a clearance posture the way a tenant inherits a fire certificate. Put a resident MOD and DE&S cell inside the perimeter as a standing first customer, doing for the whole zone what the resident buyer did for Plymouth’s South Yard. And run an export-control fast lane from the same office. A licensing question answered across a desk in a week is an entirely different business environment from one that enters the Whitehall correspondence cycle and emerges a quarter or two later.

Then build the shared plant, once, for everyone: the machine shop, the EMC chamber, the environmental test rigs, the secure comms room, with a classified-rated section for the firms that need it. The Gundo’s informal tool-borrowing economy is a distributed factory no member could afford alone five years ago. Britain’s version cannot be entirely informal due to the nature of our regulatory environment, because some of the equipment a hard-tech cluster shares needs accreditation. But it can be built, and the £25 million Freeport seed capital plus a slice of Investment Zone flexible spend likely gets you over the line. This is what those funds should be for. They are currently spent on spine roads.

And lastly, the materials layer. The Gundo’s fastest parts company lost eight months to an ingot it could not source, and a military demo died three supplier tiers down over a propeller. A cluster is only as strong as the metal in its stores, so take the one Freeport instrument everyone treats as a logistics perk and make it strategic: bonded stockholding of specialty alloys and long-lead components inside the customs site, held under duty suspension, drawn down by zone firms at hours’ notice. Everywhere else in the programme the customs site saves importers a bit of working capital. Here it would become the cluster’s strategic reserve.

Ingredient 2: The equity mechanism. The mayoral recyclable growth fund, capitalised with devolved Financial Transactions money from 2026/27, deployed through a combined-authority investment vehicle with one job: firms receive enhanced relocation and fit-out support, and the vehicle receives equity or warrants in exchange. Keep the structures boring and defensible. Preferred instruments or warrants rather than ordinary shares, minority caps so no one accuses the mayor of nationalising a drone company, ratchets tied to headcount-in-zone so the public upside scales with the thing the public actually wants, and every pound of returns ringfenced for recycling into the next cohort inside the same postcode.

Recall that Financial Transactions capital must be lent or invested rather than granted; this would simply be the first time the mechanism is pointed at one sector. And it wouldn’t stand alone, with the British Business Bank’s Nations and Regions funds sitting adjacent for co-investment, the National Wealth Fund covering the infrastructure-scale tickets, and UKEF guarantees carrying the export-backed working capital once the zone’s firms start shipping.

Ingredient 3: Relocation support. The key ingredient British schemes always omit, because paying people to move somewhere specific looks like favouritism, and favouritism is the main thing British regional policy is designed to avoid. News flash: clusters are made of movers. Nobody has ever built one out of the people who happened to be standing there.

So pay for the moves, loudly and proudly. Moving-cost grants for firms and for named key engineers. Lease-break support for the firm trapped in a science park it has outgrown. And a housing answer, because the Gundo’s founders solved theirs by living in each other’s factories and British building control will probably take a dim view of that. Use the Mayoral Development Order power to consent workforce housing directly adjacent to the tax sites, so the two-in-the-morning culture has somewhere to sleep beyond a mattress on the factory floor. Then a soft-landing service for the thing that actually kills relocations, which is never the founder’s conviction and always the spouse’s job and the children’s school places.

Below critical mass, a cluster is an industrial estate with a logo. And the absence of relocation subsidies is why Britain has twenty-two designations and zero Gundos.

Three ingredients: plumbing, equity, movers. No primary legislation, no new quango, no spending review fight. Every component runs through an instrument that already exists, from the seed capital to the sandbox commitment to the Financial Transactions rules to the development orders. None of it is hard. You cannot decree the culture, but you can build the conditions that let one condense, and every condition is purchasable today. Now, we just need to find somewhere to put them all…

Part IV: Pick One

If nothing changes, Britain will end this Parliament with a handful of zones funded at quarter-strength. They’ll have their own launch event, a photo op, a logo, a 6.5 man delivery team, and none of the compounding effects switched on. It will happen the way it always happens: a bid from every region, a ministerial instinct that saying yes to several is easier than saying yes to one, and a strategy document announcing a “network of clusters” as though the word network were an achievement rather than a confession. But if we go all-in on one of the four shortlisted candidates below, which are not ranked in any particular order, we may end up with something worth putting on a map.

But before we get into them, it is worth pointing out there are contenders in the North that I gave long consideration. Teesside, Barrow, Sheffield, Preston, not least because they would be a far easier sell for a Burnham government to get behind. But I keep arriving at the same conclusion: at this stage in Britain’s levelling-up cycle, being near London is not just a nice-to-have, it is the one advantage we cannot manufacture. El Segundo could sit five thousand miles from New York because it is part of the Los Angeles metro and had the Bay Area next door. America runs on a dozen genuinely global cities, and a hard-tech cluster can grow beside almost any of them. Britain has one. On Oxford Economics’ Global Cities Index, London ranks second in the world, behind New York and ahead of Paris, and first on the entire planet for human capital, the category that measures universities, skilled labour and corporate headquarters, which is to say the exact inputs a hard-tech cluster consumes. Manchester comes 93rd, wedged between Orlando and Tampa. On human capital specifically it ranks 91st, scoring 54.2 against London’s perfect 100. This is not a knock on Manchester, which is a fine city on an upward trajectory. It is simply the arithmetic of a country with a single superstar city. Cluster beside it, or accept a structural handicap no tax relief can close. Ask me again in ten years and Manchester may have earned a different answer!

The Candidates

Oxford (Culham) is the closest thing Britain has to a Gundo company’s natural habitat. Less than an hour to central London, Culham is the country’s first AI Growth Zone: designation already conferred, planning waived under the AI Opportunities Action Plan, and a grid connection scaling from 100 to 500 megawatts on land that is ready now. Around it, a reactor cluster is already condensing, Tokamak Energy and First Light Fusion building demonstrators on the campus, fed by one of the two best physics and engineering faculties in the country a few miles up the road, and the reactor-next-to-data-centre model now taking hold in the US. Reactors, compute and power on one campus is a combination no other candidate can touch. The weakness is a political one. Would a Burnham government, elected on reindustrialising the North, really make its flagship reindustrialisation cluster the one place in Britain associated with the elite it so often villainises?

Culham, Oxfordshire

Cambridge is the one candidate from this list where the money is already present, with its startups raising around $420 million of venture capital in the first half of 2025 alone, second only to London. Cambridge’s specialism is also very helpful to the hard-tech scene: semiconductors and compute, the Arm lineage, graphene photonics, the edge-inference and optical-interconnect firms. The city, also less than an hour away from London, runs the research computing service behind the Sunrise fusion supercomputer, which is compute depth nowhere else can claim, and the University generates many times the national patent rate. Eight hundred years of it have made Cambridge, on paper, the best founder-generation engine in the country. Sadly, due to the local greenbelt protectionism plaguing the region, its spinouts find themselves priced out by land costs second only to central London and starved of the lab space and grid a hardware company needs to grow into. A Gundo needs cheap sheds, and Cambridge has fewer of them, dearer, than anywhere on this list.

Cambridge Life Sciences Cluster

Bristol and Filton has the deepest concentration of aerospace engineering in the country outside Lancashire: Airbus and Rolls-Royce at Filton, a defence prime's cyber and AI arm in the city centre, MOD Abbey Wood with its twelve thousand procurement staff a bus ride away, strong university presence, and a robotics and semiconductor scene the other candidates lack. On the same human-capital measure where Manchester sits 91st, Bristol ranks 63rd, ahead of both Milan and Detroit, the city that gave the world the assembly line and the manufacturing cluster itself. The brilliant team at Fractile, building custom silicon to break the AI inference bottleneck, chose Bristol for their new hardware facility because Isambard-AI, the UK's most powerful supercomputer, is already there. They are exactly the founder-species firm a zone built to land a factory in Hull or Redcar never produces. Abbey Wood is the closest thing Britain has to a resident customer that is not a dockyard; the DE&S cell could be staffed by people who already live there. It is also ninety minutes from London. The weakness is that Bristol holds no zone designation, so the chassis would have to be conferred from scratch rather than repurposed from something already in place.

Filton Airfield

Milton Keynes is the corridor’s own answer to the factory problem. Around 30 mins from London, it is the largest city between Oxford and Cambridge by population and output, running at 128% of the national GVA per worker, and it has spent a decade casting itself as the place where the other two cities’ ideas get tested and deployed. The autonomy scene is the most mature in Britain: the world’s largest fleet of autonomous delivery robots, driverless shuttles on public roads, a citywide standalone 5G network, and a Smart City posture that already waves live deployment through the kind of regulatory gate the rest of the country spends pages trying to build. The land is cheap by corridor standards, the grid and road geometry were designed from scratch, and a serious government engineering presence already sits on the ground. East West Rail puts Oxford forty-five minutes away from this year. The weakness is the opposite of Cambridge’s. Milton Keynes builds superbly and invents little; it has no university of the front rank, so the founder-generation engine that fills a cluster runs weak. Perhaps someone could build an NMITE-style engineering university or Imperial could set up a hard-tech satellite campus…

Milton Keynes

Part V: What the zone looks like

If you work for a mayor or a department, this is the part to screenshot.

• Three contiguous tax sites under a single combined authority. If two sites cannot see each other, redraw the boundary.

• A regulatory sandbox inside the perimeter, per the Action Plan commitment: segregated trial airspace with CAA trial authorities on site, an ONR pre-licensing pathway for advanced reactor work, Environment Agency permitting run from a desk in the zone, and standing range and drilling access on adjacent land.

• The defence-anchor module: facility security clearance held at zone level, with pre-cleared buildings, accredited IT and vetted site management inherited by incoming firms on signature of lease; a resident MOD/DE&S cell as standing first customer; an export-control fast lane run from the same office.

• Shared plant, built once from Freeport seed capital and Investment Zone flexible spend: machine shop, EMC chamber, environmental test rigs, secure comms, with a classified-rated section for the firms that need it.

• Bonded stockholding of specialty alloys and long-lead components in the customs site, under duty suspension, drawn down by zone firms on demand.

• Grid connection secured at designation through the NESO connections accelerator, sized for the zone’s build-out rather than its first tenant.

• A combined-authority investment vehicle drawing on the mayoral recyclable growth fund, taking warrants or preferred equity against relocation and fit-out support. Minority caps, headcount ratchets, returns ringfenced within the zone.

• Relocation support as standing policy: moving grants for firms and named engineers, lease-break support, and a soft-landing service covering spouse employment and school places.

• A Mayoral Development Order consenting workforce housing adjacent to the tax sites.

• Strategic Sites Accelerator and National Wealth Fund applications packaged at designation.

• A British Business Bank Cluster Champion and a UKEF Export Finance Manager on the delivery team.

• One board, publishing its minutes, running a delivery team an order of magnitude above the English Freeport average of 6.5.

Conclusion

The window to make something like this a reality is very short.

The mayoral recyclable growth funds land in the 2026/27 settlement, which is to say they are landing now, and the first deployments will set the template for every one after. The Strategic Sites Accelerator picks its first wave within the year, and a hard-tech zone that misses it joins a queue behind every gigafactory and lab campus in the country. And the demand is not a lake but three rivers running at once: a £298 billion Defence Investment Plan committed monthly, programme by programme; a grid buildout letting contracts by the gigawatt; a compute race buying power and hardware faster than either can be made. The suppliers to all three are choosing where to expand as the contracts arrive. Every month of dispersed spending is density that never compounds, and unlike money, density cannot be reallocated later.

So, a prediction, in the same spirit as the ones I made about data centres last month, which have so far had the poor taste to keep coming true. If no single zone is chosen this funding cycle, Britain’s next generation of hard-tech founders who would have filled one will not be waiting patiently in a science park outside Warrington. They will be in El Segundo with everyone else. The talent is not loyal to the map; it is loyal to the density, and the density is currently five thousand miles away.

Lessons from Gundo to follow.

Thank you for reading. Feel free to get in touch with any questions james@amphio.com

Prologue: Hill County, TX

On the morning of May 12, 2026, five county commissioners took their seats in a courtroom in Hillsboro, Texas, fifty-five miles south of Fort Worth. The room was full. By three votes to two, Hill County, with a population of thirty-seven thousand, passed a one-year moratorium (suspension) on new data center construction in its unincorporated areas. The first county in the history of Texas to do so.

Pictured: Hill County Commissioners voting to pass a temporary moratorium on data centers, May 12 2026.

This isn’t California or New York, where any attempt to build something is treated as a crime against humanity. Texas is the most pro-development state in America and the center of gravity of the AI sector’s buildout, with an estimated 400 data center projects operating or in development. Developers went there because the land is cheap and the state leaves builders alone. The Hill County vote, however, put as many as eight planned projects in jeopardy overnight. A Dallas developer sued almost immediately. Within a month the county backed down. The moratorium was rescinded and replaced with a mandatory Major Industrial Development Checklist and review framework. In other words, mountains of red tape of the kind you’d expect to encounter in Greenwich or Jackson Hole.

Next door in Hood County, commissioners tried twice to pass a similar pause and failed twice, after a state senator warned them they were acting beyond their legal authority. Developers sued the county anyway, twice. The county clerk called her county’s failure to act the biggest disappointment of the whole episode. “Hood County should have been the tip of the spear,” she said.

In Angelina County, in the pine country of east Texas, residents packed the commissioners court to oppose a facility proposed for the site of an old paper mill outside Lufkin. The county judge told them he shared their concerns and could do almost nothing with them. “The Texas legislators have consciously limited what we can do,” he said, “and they’ve done it on purpose.”

Then the state itself began to move. On May 18, the Texas Agriculture Commissioner, an elected Republican, called for a statewide moratorium on hyperscale data centers. On June 10, Governor Greg Abbott ordered the Public Utility Commission and ERCOT to draw up rules requiring data centers to generate their own power and recycle their own water, at their own expense. Three weeks later, at a campaign stop in East Texas, he went further: “We must prohibit them from building AI data centers in rural Texas neighborhoods.” He wants their tax break eliminated too. This is the same governor who stood up in November to celebrate a $40 billion Google investment and call Texas the epicenter of AI development. Eight months from champion to prohibitionist. That is how fast this thing is moving. Governor Abbott has no doubt been reading the same polling as me.

Everything I’ve read about the AI race is set in the same few zip codes, the labs of the Bay Area and the trading floors of New York. Its physical body however belongs to rural America, and the body’s future arrives first in places like Hillsboro, in county commissioners courts that open with a prayer, and in rooms where a nine-figure project can die on a Tuesday morning in front of five unknown elected officials.

The AI sector’s buildout runs on three inputs. The first is land, and the industry has a plan for land: site acquisition in low-regulation zones. The second is power, and the industry has a plan for power: behind-the-meter generation and billions in committed grid spend. The third is the consent of the people who live beside these projects. Virtually nobody has a plan for consent beyond regional PR firms who charge $3k a month for some F2F with local politicians more interested in what time Chili’s closes.

Part I of this essay documents the collapse in public opinion of data centers, the fastest ever recorded against any category of infrastructure. Parts II through V describe the machine that produced it and the election that is about to weaponize it. Parts VI and VII explain why the position is far more winnable than it looks, and exactly how it gets won. Part VIII is a checklist for anyone looking to not have their pipeline crushed between now and November.

Part I: The Collapse

By the market research industry’s standards, public opinion in America does not usually move quickly. Views on nuclear power took a decade and a meltdown to shift ten points. Same-sex marriage, the textbook case of rapid attitude change, moved roughly twenty points across ten years. Pollsters build entire careers on swings of five.

Last September, Heatmap Pro asked Americans for the first time how they would feel about a data center built near their home. The country split down the middle: 43% in support, 42% opposed, 15% unsure. A coin flip question where most respondents could barely describe what one was. In short, America had not yet formed an opinion.

This February, the same tracker had opposition at 51%. An increase of 9 points.

By June, the question was settled. In the latest wave, fielded by Embold Research across 4,118 registered voters in all fifty states, at least seven in ten Americans opposed a data center near their home. An outright majority of the entire country, 55%, described themselves as strongly opposed. Support had fallen to 21%. Heatmap’s own summary was that the public had “swung 49 points against data centers in just nine months.”

Nine months. Charles Franklin has run the Marquette Law School poll for fifteen years and says he has never tracked anything develop faster. There is no equivalent in the modern polling record for an infrastructure category. Wind farms took twenty years to become contested. Data centers went from unknown to the bottom of the table in under two.

Since 2001, Gallup has asked Americans whether they would oppose a nuclear power plant near their home. Across twenty-five years of asking, through Fukushima, opposition never once exceeded 63% . In March 2026, Gallup put the same question about data centers into the survey for the first time. Opposition came back at 71%. In the same survey, a nuclear plant drew 53%.

To clarify: Americans would now rather live beside a nuclear power plant than a data center.

For the developers reading this looking for geographic refuge in the crosstabs - you will not find one. No region of the United States polls below 69% opposition. Among Americans aged 18 to 34, opposition runs at 80%: the generation that actually uses these models every day is the most hostile to their physical footprint. And then there is Virginia, the industry’s home turf, host to the largest concentration of data centers on earth and two decades of lived experience with them. Support for new construction there fell from 69% in 2023 to 35% in an April 2026 Washington Post poll. The sector’s oldest host community has turned against it by nearly two to one. Where familiarity was deepest, the collapse was steepest.

To summarise, social license is an input to every data center on every pipeline in the West, and that input has been repriced by 49 points in nine months. If any other core input had moved like that, it would be the only item on the sector’s agenda.

This is on almost no one’s agenda.

A prediction, the first of several in this essay: national opposition to local data center construction does not fall below 65% on any credible tracker before the November midterms. Nothing that would push the number down currently exists. Everything pushing it up is accelerating, and the next four parts are about what, exactly, is doing the pushing.

Part II: There Is No Cavalry

Every industry that gets into political trouble has a mental model of who will save it. Oil has Texas. Coal had West Virginia. Crypto bought itself a Senate caucus. For data centers the assumed cavalry was pretty specific: rural counties would welcome the tax base, and red states would defend capital investment on principle. These were reasonable assumptions to make, given they had held for every category of American infrastructure for fifty years.

Here is what the polling shows instead. In Heatmap’s June wave, 78% of Harris voters oppose a data center near their home. So do 63% of Trump voters. The fifteen-point gap looks meaningful until you notice that both numbers are landslides. There is no version of American politics in which 63% of a party’s voters oppose something and its elected officials defend it for long.

The rural numbers are worse. Rural Americans oppose local data centers 73% to 19, a net of minus 54, which makes the countryside more hostile than the suburbs and more hostile than the cities. Now compare that against the industry’s siting map. The entire buildout strategy of the last three years has been to move into exactly these places, on the assumption that they would compete for the investment, when actually they’ve turned out to be its most committed opponents.

The sector needs to stop for a moment and register how strange this is in a country that no longer agrees on anything. Polarization carries one consolation for anyone under attack: if one tribe decides to hate you, the other tribe will adopt you out of spite. That reflex has rescued oil companies, gun manufacturers and pipelines. It is the closest thing American public life has to a safety net, and data centers fell through it. The physical substrate of the AI economy has achieved the rarest outcome available in this country, genuine cross-partisan consensus, and the consensus is against them. Bravo.

So what have we learnt in the last nine months?

1. No audience belongs to a data center developer by default

2. No district is red enough to guarantee a welcome

3. No county is poor enough that the tax revenue argument wins on its own

Later in this essay I will show that winning these people over is more achievable than the numbers in Part I suggest. But the sector needs to be honest with itself. The cavalry is not coming, because the cavalry has joined the other side. They got there first.

Which raises the question: fifty years of pro-development instinct in rural America did not dissolve on its own in nine months. Someone dissolved it. Part III is about who.

Part III: The Machine

The industry’s comfort word for its opponents is NIMBY. It implies something spontaneous, local and disorganized, a scattering of retirees with yard signs who will tire themselves out after the second hearing. Spend any time digging into the last nine months and you realise you are up against an incredibly well-funded, organised, layered machine.

Start at the bottom. Data Center Watch, a research project that tracks opposition activity across the country, counted 396 active local opposition groups at the end of 2025. By the end of March 2026 it counted 833, spread across 49 states. The number of organized groups fighting data centers more than doubled in a single quarter. Petition volumes in those three months nearly matched the entirety of the previous half year. And the groups are no longer just reactive in nature. Data Center Watch found opposition mobilizing before any paperwork had been filed, organizing against projects that existed only as gossip. Consider what that means operationally for your pipeline. The opposition now has sensors in the ground ahead of the industry’s own announcements. It is a distributed early-warning network covering the whole country, staffed by volunteers, running at zero marginal cost, and the sector has built nothing equivalent. You can watch it operate in real time: the trackers at datacenterwatch.org and the live opposition map at brockovichdatacenter.com update continuously, like weather radar for a hurricane the industry thinks are scattered showers.

One layer up sits the professionals. The December letter I will come to in a moment was organized by Food & Water Watch and signed by, among others, the Sierra Club, the NRDC, Greenpeace and Friends of the Earth. These are organizations with full-time litigators and standing relationships with every national newsroom, and they have spent decades running this playbook against pipelines and coal plants. They have also started publishing doctrine. In December 2025 the AI Now Institute released a “North Star Data Center Policy Toolkit,” a manual for using local and state policy to stop data center expansion. A month later MediaJustice published its own organizing guide, The People Say No. Read those documents and then ask what the equivalent artifact is on the development side. There isn’t one. For the opposition, defeating your project is the job. For most developers, community engagement is a line item, usually the smallest one on the sheet.

Another layer up, the machine touches the legislature. On December 8, 2025, more than 230 organizations co-signed a letter demanding a national moratorium on data center construction. One hundred and seven days later, a national moratorium bill appeared in Congress, the AI Data Center Moratorium Act, S.4214 in the Senate and H.R.9442 in the House, its provisions tracking the letter’s demands closely. Below the federal level, more than 300 data center bills landed in state legislatures in the first six weeks of 2026. Statewide moratorium proposals are moving in 14 states. Maine passed both chambers before a gubernatorial veto killed it, and the override failed narrowly. New York’s legislature has passed a one-year pause on large permits. At the local level, 54 moratoriums have already been enacted, with nine more pending. Follow the full arc and it runs from a Facebook group to a filed bill inside months. That is a textbook conversion funnel, and someone built it.

Which brings us to the money. According to a May 2026 report by the Bitcoin Policy Institute and testimony delivered to the House Ways and Means Committee in February, more than $2 billion has flowed through foreign-tied charitable vehicles into the American advocacy infrastructure now aimed at data centers. The Wyss Foundation, controlled by a Swiss national, has moved roughly $700 million into US advocacy through its action fund. The Oak Foundation, also foreign-controlled, has granted over $750 million to more than 150 American groups. The American Energy Institute traces $39 million in foreign-linked funding to twelve organizations actively opposing data center development. A separate network documented by the New York Times, and now the subject of four House committee inquiries, has produced anti-US-AI-infrastructure content for five years. Chinese state media outlets run openly attributed campaigns against American data centers while Beijing cuts the power bills of its own AI operators. And in a follow-up published two weeks ago, the same researchers attributed $23.6 billion in stalled or blocked American projects specifically to what they call a China-aligned ground game.

Now take a step back and realise what this machine actually is. The opposition is running a modern political campaign. It sets the narrative in a community before the developer has scheduled a town hall. It pairs distributed local organization with national message discipline. It escalates from petition to ordinance to statute along a tested path. Against this, the industry is running a permitting playbook from the 90s: acquire quietly, file, present the PR firm’s benefits package at the statutory hearing, absorb the anger, appeal, repeat.

A second prediction. Active opposition groups exceed 1,200 by election day. The count doubled last quarter and two full quarters remain, so 1,200 actually assumes the machine slows down.

The obvious question is what all this has already cost the sector. Onto the invoice…

Part IV: The Ledger, and the Texas Tell

Opinion is one thing. Money is another.

In the first quarter of 2026, according to Data Center Watch’s tracking, at least 75 projects worth roughly $130 billion were blocked or delayed by local opposition. Three months. The figure is the largest ever recorded in a single quarter, and it roughly equals the total for all of 2025. Heatmap Pro’s separate count, using a stricter definition, found 20 projects cancelled outright in the same window, double the previous quarter, erasing more than $41 billion in planned investment and at least 3.5 gigawatts of electricity demand. I will let you do the maths on where the annual figure will land if the first quarter turns out to be the slow one.

We established in Part III that opposition groups are forming on rumor, before any filing exists. The financial consequence is that political risk has moved upstream of acquisition. A developer can now inherit an organized, named, fundraising opposition group as a condition of the site, before owning the site. In Livingston County, Michigan, a farming area that went 61% for Trump, hundreds of residents packed public meetings against a proposed 1,000-acre, billion-dollar Meta facility until the company withdrew. The county was chosen precisely because it was considered friendly to business.

Which brings us to the Lone Star State. On paper Texas is the perfect siting jurisdiction and the industry has treated it accordingly: an estimated 400 projects operating or in development, the largest concentration in the country. The reasoning is obvious. Texas counties have almost no zoning power, so developers have increasingly routed projects into rural unincorporated land where, as one Hill County commissioner put it, the regulatory “sweet spot” sits. But the same absence of process that makes these places easy to build in makes them impossible to read. No zoning means no hearings until late, no hearings means no early signal, and no early signal means the first time a developer meets the community is often the day the community has already organized. The prologue showed what that bomb looks like when it detonates. Hill County’s moratorium was the first in state history, Hood County has been sued twice over its attempts, and in Angelina County the anger is now aimed at a judge who agrees with the residents but holds no legal power to act.

So what do Texans actually think? Focaldata has been polling the state’s registered voters on exactly this, and two findings matter more than all the others.

The first is what I call the proximity gap. Asked about building new data centers in Texas, the state is close to evenly divided, net minus 3, and Trump voters are actually in favor, net plus 8. Asked about a data center near your community, the same electorate moves to net minus 12, and Trump voters flip to minus 10. That is an eighteen-point swing among the buildout’s supposed base, produced by nothing except moving the project from the abstract to the local. Harris voters go from minus 14 to minus 18. Unaffiliated voters fall from minus 13 to minus 28.

Shot:

Chaser:

The second key finding is the moratorium appetite. We asked Texans how they felt about other counties pausing data center construction, the Hill County move. Net support: plus 46. Among Trump voters it is plus 57, higher than among Harris voters at plus 47.

In other words, in the reddest large state in America, which hosts the greatest concentration of projects on earth, the voters most supportive of freezing construction are the ones the industry has been counting on as its base.

So statewide sentiment says the market is fine but local sentiment says every individual site is the devil. A developer reading state-level polling, or worse, reading nothing, would naturally see a green light that does not exist at any actual location. The sums detailed above is what driving through that phantom green light has cost the sector so far, at a moment when the only people acting on these public setiment numbers are county commissioners. The far more expensive moment arrives when professional politicians start reading them, and as Part V will show, that moment is here.

Part V: The Uncontested Election

Everything so far has been a development story, told in permits and polling. This part is about the moment it stops being one, because there is a fixed date on the calendar, the first Tuesday of November, when data centers stops being an investment strategy and starts being an instrument for acquiring public office.

Start with what kind of issue this actually is, because the industry keeps misfiling it. Data centers poll like a cost-of-living issue, and they poll that way because voters experience them as one. When Gallup asked opponents why they object, the answers were water, energy, pollution and bills. Heatmap’s polling finds more than half of Americans hold data centers largely responsible for rising electricity prices. Marquette’s national survey finds 71% say the costs outweigh the benefits. Whatever the engineering truth of any individual project, the voter has already categorized the asset class: it is the thing making life more expensive and the reason they can’t get their kid a new toy for Christmas. Make no mistake: cost of living is the master issue of this electoral cycle, the frame through which every campaign in the country is being built.

Local officials have started losing their jobs over data centers. Festus, Missouri recalled half its city council over a six-billion-dollar proposal. Warrenton, Virginia voted out its entire council after it approved an Amazon facility. A New Jersey official told Newsweek: “Do I think that the data center vote cost me the election? Yes I do.”

Every politician in America can read an obituary. The lesson being absorbed at the county level right now is that there are careers to be lost by approving these projects and careers to be made by stopping them.

Statewide campaigns are building platforms on this issue as I write this essay. In Wisconsin, all seven prominent Democratic candidates for governor back data center regulation or an outright moratorium. In Pennsylvania, the sitting Democratic governor demands that data centers bring their own power while his Republican challenger tries to outflank him with a full development pause. A Republican candidate for governor of Florida has pledged to ban them in all 67 counties. Illinois has paused its tax incentives. A Republican senator from Missouri has filed a bring-your-own-power bill in Washington while a socialist senator from Vermont campaigns for a national moratorium. Notice what is missing from this list. A party line. The issue is being claimed simultaneously from the left as corporate greed and from the right as ratepayer protection. Both parties’ candidates are converging on the same position because both parties’ pollsters are reading the same crosstabs.

Next prediction. I believe data centers run as a top issue in the November midterms, packaged inside cost of living, and I believe this cycle produces the first national politician who makes the issue theirs. When that happens, a mechanism activates that the industry has never faced before, so it is worth spelling out slowly.

A national figure claims the issue. Every local zoning fight becomes evidence for a national argument, which means every contested site arrives with a press corps attached. The cost of a botched hearing stops being a nine-month delay and becomes a news cycle with your company’s name in the chyron. And that footage does not expire. It plays again at the next hearing, in the next county, presented by the next opposition group as proof of what your company does to communities like their’s. Local failures currently die locally. After the looming nationalization of this issue, every failure is permanent, portable and in the training data of every future fight.

Also note the asymmetry in who benefits from this. For the opposition, nationalization is pure oxygen. National attention fills the petition lists, fills the fundraising pages and recruits the next 400 local groups at zero cost. The machine from Part III is literally built to metabolize attention. For a developer, the same attention flows straight into the discount rate. Lenders reprice, insurers reprice, counterparties add contingencies, and projects that pencilled at the old level of political risk stop pencilling. An opponent that gets stronger from the fight while you get weaker from it is the definition of a losing position.

Our Texas data shows how far down the ballot the sorting has already reached. We asked Texans which Senate candidate would best handle data center development, and they had answers: 35% picked the Democrat, 20 and 17% split between the two Republicans, and the preferences sorted cleanly by 2024 vote. The question did not confuse them, because the issue has already entered their candidate-choice architecture, in the state with more projects than any other. And for anyone in the industry hoping state capitals will simply preempt the troublesome counties, the voters have a view on that too: asked who should control data center development, 44% of Texans side with local counties against 37% for state authority. The Austin-will-fix-it escape hatch is underwater with the very electorate that would have to forgive its use.

Between now and November, every competitive campaign in a state with a pipeline will discover what the candidates above have discovered. By election day, the anti side of this issue will be established in essentially every contested geography across the country. Community sentiment, once set, costs an order of magnitude more to reverse than it would have cost to shape, because after November the developer is no longer filling a blank; the developer is arguing with a position the community voted for. This is one big game of who gets there first.

This whole essay exists to deliver one message: the firms with no basic community-building infrastructure in place are going to get absolutely steamrolled when this flips in November.

The one thing this part has not told you is why any of it is survivable. That is Part VI.

Part VI: The Tiger is Paper.

Congratulations for making it through the first five parts of dread. Now for good news: the opposition is far softer than its headline numbers, and the softness sits in specific, measurable places.

Yes, the 71% opposition is real. But it is also, in large part, a coalition of the convinced and the merely defaulted, people who absorbed one scary fact about water or bills from the only side talking to them. This can be fixed.

Start with the finding I consider the most important in our Texas data. We asked Texans directly what should happen with data centers and the grid, and gave them the full menu. Only 17% want new data centers paused until ERCOT (Electric Reliability Council of Texas) is strengthened. Only 10% say build them even if it strains the grid. And 47%, the outright plurality of the state, say data centers should be built if they generate their own renewable energy or pay to upgrade the grid. Translation: in the most contested data center state in America, the prohibitionists are a sixth of the electorate, and the largest bloc by far is holding out terms.

The policy questions double down on this. Requiring data centers to contribute to grid upgrade costs polls at net plus 59. Requiring renewable energy, plus 51. Restricting foreign ownership near military installations, plus 52. A dedicated state regulatory framework, plus 45. On the other end of the spectrum, tax incentives to attract data center investment sit at net minus 4, and at minus 30 among unaffiliated voters. Texans are largely pro-rules and anti-subsidy. They will accept the buildout but they refuse to pay for the privilege. Break the grid question out by 2024 vote and the picture sharpens further: among Trump voters, making developers pay for grid upgrades polls at plus 70. The single most Republican demand in our entire survey is that the industry cover its own costs.

The diagnosis writes itself. The electorate aren’t refusing, they’re negotiating. And it’s a negotiation to which the industry has simply not shown up.

Now cosider the proximity gap from Part IV, because it tells you the second diagnosis. The same Trump voter who is plus 8 on the Texas buildout is minus 10 when the project lands near his home. Ideology does not invert with distance. Unaddressed local risk does.

Solution: address that risk with clear answers.

The audience willing to listen to those answers is still absolutely massive. After nine months of collapse, roughly a third of Texans are uncommitted on a data center near their community, answering neither or don’t know. The fight for the middle has barely been joined.

What would the answers need to address? The opposition’s own voice is the best guide. When Gallup asked opponents in an open-ended follow-up why they object, they said water, 18%. Energy, 18%. Pollution, including noise, 16%. Higher bills, 15%. Every single item is answerable with facility design and binding commitments. Closed-loop cooling answers the water number. Behind-the-meter generation answers the energy number. Enforceable bill protection answers the cost number. These are information failures and design failures, and both kinds are correctable.

The residual core of opponents for whom no concession works, the people who believe none of this should be built at all, exists in every local fight I have worked. It is loud. It is also small, and the correct strategy is containment rather than conversion. Meanwhile the raw material for a pro coalition is already lying on the ground: 45% of Texans name job creation as a benefit of data centers and 35% name tax revenue for local schools and communities. Supporters exist. They are silent because nobody has ever organized them, and silence reads as absence in a hearing room.

Do persuasion facts actually move people, or is that just consultant’s cope? The public evidence says they move. In Heatmap’s September 2025 message testing, the claim that data centers create high-paying construction and operations jobs polled at plus 26 net convincing. The catch is that the opposition’s messages currently hit harder, with the bills argument at plus 46 and the water argument at plus 34. Notice what that asymmetry is made of. Both sides possess persuasive facts. One side has been delivering its facts to the low-information majority for nine months while the other sent a press release to datacenterdynamics.com. That gap comes from sequencing and repetition, and sequencing can be changed.

This is testable against my own record: across the sites I have worked on, contested projects that ran aggressive methodologies such as sentiment mapping, segment-level messaging and concession design before the decisive hearing have ended in acceptance more than 80% of the time. The methodologies behind that number is the subject of the next chapter.

One more finding. The single cheapest variable in the whole fight is brand awareness. Texans’ net comfort with Google owning a local data center is plus 33. Microsoft, plus 25. Amazon Web Services, plus 20. Meta scrapes by at plus 1, and xAI sits at minus 12. TikTok polls at minus 41, and a company backed by the Chinese or Saudi government at minus 52. The trust discount starts before a word is spoken, and it varies by 85 points depending on whose logo is on the fence. Note also what the foreign-ownership numbers hand to American developers: the most potent security argument in the survey, restrictions on foreign-controlled facilities near military sites at plus 52 overall and plus 63 among Trump voters, is an argument only the domestic industry can make. And almost nobody makes it.

So the tiger is paper. A sixth of the electorate wants prohibition, a third is unclaimed, and the plurality is waiting for an offer that matches terms they have already stated. But paper tigers still win wars against opponents who never show up to the field of play, and every month of absence hardens soft opposition into the firm kind. The concrete is setting, and it sets permanently around election day. The question is whether anyone on the building side owns instruments precise enough to find the soft places while they are still soft. Those instruments exist. I use them for a living, and Part VII is an honest tour of how they work.

Part VII: The Counter-Machine

None of this section is exotic. Any figures in this section are representative examples, included to show the shape of the output rather than findings from a live study.

Begin with the problem that defeats ordinary polling. A respectable national survey of 20,000 people sounds enormous until you ask it about one rural county, at which point you discover it contains perhaps thirty respondents there, a sample from which you can conclude nothing. Since the buildout’s fate is decided precisely in such counties, conventional polling enjoyed by the industry is structurally blind exactly where the war is. The workaround is a technique called MRP, multilevel regression with post-stratification. Instead of treating each county as its own tiny poll, you use the entire national sample to model how opinion varies with the characteristics of people and places: age, tenure, education, housing type, local industry, distance from existing infrastructure, dozens more. That model is then run against the census, which tells you exactly how many people of each type live in every precinct in the country. The output is a statistically disciplined estimate of opinion for every small geography in a state, including ones where the raw poll spoke to five people. British readers will remember this as the technique that correctly called dozens of individual constituencies off a single national poll in the 2017 election while conventional polling missed the result, and it descends from the same family of methods the Vote Leave data operation used to find persuadable voters the traditional campaigns were ignoring. The payoff for a developer is blunt. A sentiment map of an entire state, resolved to the precinct, before a single acre is optioned. Community risk priced into the acquisition alongside interconnection risk and water rights, instead of discovered at the first hearing.

The maps tell you where. The next instrument tells you who. The single most common strategic error I see is segmenting an audience by demographics, as if age and income were what determined a person’s view of a server hall. They are weak predictors. What matters is attitude, and attitudinal segments recur so reliably from state to state that I can sketch the cast in advance. There are jobs-first pragmatists, usually the only reliably favorable bloc, who lack nothing except a reason to speak up. There are benefit-conditional families who will trade acceptance for schools and roads. There are anxious ratepayers, the pivot of nearly every fight, whose entire relationship to the project runs through their utility bill and their well, and who read as opponents but are actually unanswered questions. There are place-protective sceptics who distrust outside developers on principle and respond to nothing except demonstrated permanence. There is a values-driven core opposed to the technology itself, for whom the honest strategy is containment rather than conversion, because no concession addresses an objection to existence. And there are the disengaged, who become whatever the first persuasive voice makes of them. Run one message at all of these audiences at once, which is what a standard community-relations campaign does for many of the developers I speak to, and you overspend on the immovable while underselling the movable. The segments need different offers and different facts.

Which offers? That question also has an instrument. Conjoint analysis shows respondents pairs of hypothetical projects with randomized attributes, water arrangements, power sourcing, bill protections, hiring commitments, fund structures, and forces a choice, thousands of times across a sample. The statistics recover the independent weight each commitment carries in the decision. The practical translation is that the community benefit agreement stops being a gut-feeling negotiation and becomes a priced weapon of mass destruction: which package flips which precinct, and whether a hiring guarantee buys more acceptance per dollar than a community fund. Our Texas polling already shows the headline terms in that state’s market, own power and grid payment, sitting in plain sight at plus 51 and plus 59. Conjoint is how you find the precinct-level price beneath the headline.

A related instrument, MaxDiff, does for messages what conjoint does for concessions. Respondents repeatedly pick the most and least convincing from small sets of factual claims, and the output is a ranked hierarchy of arguments for each segment. The finding that matters is that the same set of true facts reorders completely by audience. The jobs message that tops the ranking for pragmatists can land mid-table for ratepayers, whose ranking is led by bills and water. Open the town hall with jobs in front of a room of ratepayers and you have spent your strongest card on people it was never going to move, which is roughly the story of every failed benefits presentation I have reviewed.

Then there is the question of who delivers the message, and it may be worth more than everything above depending on the site. Nobody believes the developer. In every local persuasion study I have seen, the company spokesperson and the paid advertisement sit at the bottom of the trust table, and the neighbour, the local business owner and the credibly converted sceptic sit at the top. Our owner-comfort data in Part VI showed the discount begins with the name on the gate itself, before anyone has spoken. The operational conclusion is that the campaign cannot be fronted by the company. It is fronted by validators, recruited early, armed with the facts that MaxDiff says their segment finds convincing, and the company’s job is to make those validators impossible to embarrass, by building the project that matches what it promised.

Assembled, the instruments form a loop rather than a report. Baseline poll, map, segmentation, concession and message testing, campaign through validators, then re-poll to prove movement. Alongside the loop runs continuous legislative tracking, so a moratorium bill is visible while it is still a draft. Across the sites I have worked on, that loop, run before the decisive hearing rather than after the first disaster, is the machinery behind the +80% contested-to-accepted record I cited in Part VI.

Now the caveat, which doubles as the strategy in a roundabout way. The returns on all this are site-specific. A perfectly run campaign moves sentiment around that site, for that developer. It does not detectably lift the category, and Part I shows the category trend swamping everything else. This is why the sector’s instinct to wait for a trade-association fix, some national image campaign that rehabilitates data centers in general, misreads the terrain. There is no category-level fix, because trust in this market does not aggregate. It is earned deal by deal, radius by radius, and it has to be built that way, which incidentally means early movers capture something durable: a portfolio of communities that trust them specifically, an asset no competitor and no national mood can take away.

Ten years ago this exact toolkit existed for one purpose, winning elections. It points at infrastructure siting today because the two problems turned out to be the same problem: a large, poorly understood population that is softer than its headline numbers, and a narrow window to reach the right person with the right fact before someone else arrives with a worse one. The opposition understood the equivalence last year and imported the same machinery. The industry is still treating what should be a political campaign as a permitting exercise. Part VIII is about closing that gap.

Part VIII: The Window

If the first seven parts are right, then a firm with a data center pipeline in the US needs a specific set of community-building infrastructure in place before November. The above research is not a PR product but the absolute minimum standard of care for operating a politically exposed asset class in an election year.

Run the tape forward twelve months and watch two firms, same state, same quarter. Firm A does what the industry has always done. It acquires quietly, files, and walks into the statutory hearing with a generic benefits package and a company spokesperson. The room it walks into was organized three months before the filing, by people who have run this play a dozen times, in front of officials who have read Part V’s obituaries even if they have never read a poll. The footage from that hearing follows the firm to its next county. Firm B knew the precinct’s persuadable share before it optioned the land. It designed the facility to the stated terms of that electorate, generating its own power, closing its water loop, protecting resident bills in writing, because its testing said those commitments carry the decision. The person at the microphone on its behalf lives four miles from the site. Same asset class, same political weather. One of these firms compounds a reputation, one of them compounds a discount rate, and the gap between them was built in the eighteen months before either hearing was scheduled.

Then zoom all the way out, because the stakes of what we are talking about here run past any single portfolio. Every serious analysis of the AI race, every scaling forecast, every national-security memo, rests on an unexamined assumption: the compute gets built. Chips have national-security-grade attention and an act of Congress. Capital has more commitment than it can deploy. Power is being solved with a seriousness that would have seemed fantastical three years ago. Consent, the fourth wall of the structure, has no owner anywhere in the system, no line in any strategy document I have seen, and it is the wall currently failing. The strategic competitor does not share the problem. China does not hold planning hearings, and the documented record in Part III suggests it has noticed that we do. If the Western buildout dies, it will not die in a fab or a bond market. It will die in county commissioners’ chambers, one Tuesday night at a time, in rooms that hold forty people, and the entire edifice reasoned on top of compute will quietly go with it.

So end where this began, in Hillsboro. The Hill County moratorium was rescinded, this time, because one developer sued and one small unprepared county who could not afford the fight. Nothing about the underlying room changed. The voters who produced that vote are the same voters our polling finds at plus 46 for the next county’s pause, and the machine that organized them is adding groups at a pace that doubles each quarter, and the election that converts all of it into mandate is now inside four months. The next hundred Hill Counties arrive with campaign action plans attached. After November, the developer no longer introduces itself to a community; it answers to a position the community has already voted for.

To conclude, the opposition is organized, funded and early, and it is also softer than its numbers, holding terms the industry could meet tomorrow. The middle is large, reachable and unclaimed, and it stays unclaimed only until the first side arrives. This is a race to the persuadable middle of a few thousand American counties, and at time of writing the builders are losing it by default, in most places without having entered.

Land has a plan. Power has a plan. The window for consent closes in November. Good luck!

Thank you for reading. Feel free to get in touch with any questions james@amphio.com