A Different Kind of Water Problem

There are two entirely separate water stories in this section, and conflating them is the most common analytical error in the sector. One is scarcity — whether there is enough water, where, and who competes for it. That belongs to the Strain thread. This piece is about the other one: compliance. Not whether the water exists, but whether the pipes carrying it meet the law.

The distinction matters because the drivers are completely different. Scarcity is driven by hydrology, climate and competing demand — uncertain, contested, hard to time. Compliance is driven by a rule with a date on it. When a regulator sets a standard and a deadline, it converts an aging, deferred-maintenance problem into a schedule of non-discretionary capital spending. That is a far more predictable thing to underwrite, and it is happening now at enormous scale.

Four Mandates Landing on One Set of Utilities

What makes this moment unusual is that several major obligations are converging on the same operators at the same time. Each alone would be a significant capital programme; together they form a wall.

The lead mandate is the most concrete. Full replacement runs roughly $12,500 per line, and the rule gives systems a ten-year window running to late 2037 — a defined, dated, nationwide programme. PFAS is similar in character: fixed limits on specific compounds, a compliance date, and treatment plants that must be built to hit it. Regulators have adjusted details of both rules, and litigation continues, but the core obligations have held.

The Gap Between the Mandate and the Money

Federal support is real but nowhere near sufficient. The infrastructure law dedicated roughly $3 billion a year for five years to lead service line replacement — about $15 billion in total — with the final-year allotment landing near $2.9 billion after Congress trimmed it. Set that against a lead bill of $45–60 billion and the arithmetic is stark. Broader federal water funding faces the same squeeze: a rescission in late 2025 and a reauthorisation decision in 2026 make forward federal support genuinely uncertain.

Whatever federal money does not cover falls to two places: ratepayers, through water bills, and municipal balance sheets, through borrowing. That is the crux of the whole story. The spending will happen — it is legally required — but the question of who funds it is unresolved, and it is being answered one rate case at a time.

Affordability, and the Consolidation It Drives

Here is the strain inside this otherwise-orderly story. The US water sector is extraordinarily fragmented — tens of thousands of community systems, many serving small towns with a few thousand customers and no meaningful engineering staff. A large utility can absorb a PFAS treatment plant across a big customer base. A small system facing the same mandate must spread it across far fewer bills, and the rate increase can be brutal.

The federal response acknowledges this: a large share of state revolving-fund money is required to go out as grants or principal forgiveness aimed at disadvantaged communities. But subsidy alone does not close the gap, which drives the sector’s defining structural trend: consolidation. Small systems that cannot fund compliance are increasingly absorbed by larger regulated utilities or by private platforms with the balance sheet and technical capacity to comply. Compliance economics are quietly reorganising who owns America’s water.

Reading It Through the Frameworks

Where does policy become the cash flow? More directly than almost anywhere else in this section. A regulated water utility that invests in compliance adds that capital to its rate base and earns an approved return on it for decades — the same regulated-asset-base mechanism as the grid rewire. The difference is that here the spending is compelled rather than merely permitted. The regulator does not just allow the investment; it requires it, and then lets the utility recover it.

Where is the moat? In the compliance capability itself — the treatment technology, the engineering and programme-management capacity, and the balance sheet to fund it. Water is a natural local monopoly to begin with; mandates raise the cost of operating one, which strengthens incumbents and squeezes out sub-scale systems.

What stage and what risk? Largely brownfield replacement of existing assets under regulated recovery — the core end of the risk spectrum. The returns are unspectacular and the timelines are long, which is precisely the point: this is the quiet, bond-like corner of the infrastructure decade, and its main risks are political (affordability pushback, funding reauthorisation) rather than technical or commercial.

The Trillion-Dollar Supercycle

The American electric grid is the largest machine ever built, and most of it is old. Now, at the exact moment the load curve is turning vertical — data centres, electrification, reshored industry — that ageing machine has to be rebuilt. US regulated utilities are on track to spend more than a trillion dollars between 2025 and 2029, the most ambitious investment cycle the industry has run in roughly 140 years.

The spending does two distinct jobs, and it is worth keeping them separate. One is hardening — making the existing grid survive a more hostile climate of wildfires, storms and heat. The other is capacity — getting more power through the network to serve the surging load. Both run through the same mechanism: a regulated utility invests capital, a regulator approves it into the rate base, and the utility earns an allowed return on it for decades. That structure is the whole reason this is such a clean investment story — and it is the primer’s regulated-asset-base model in its purest form.

The Hardening Side: Undergrounding

The first half of the spend is defensive, and it is being driven by a brutal piece of liability math. When a utility’s overhead line sparks a wildfire, the utility can be liable for billions — PG&E’s bankruptcy is the cautionary tale the whole industry now plans around. Against that exposure, burying the lines starts to look cheap, even at extraordinary cost.

The programmes are enormous. PG&E is undergrounding some 10,000 miles of line in fire-threat areas at roughly $1.85 to $6.1 million per mile. Florida Power & Light has a 25-year, $35 billion hardening programme; California’s three big utilities committed nearly $24 billion to wildfire mitigation over 2023–25; Louisiana’s Entergy is hardening 269,000 structures across 11,000 miles. Undergrounding is fast becoming the default standard for new build in exposed areas — not because it is cheap, but because the alternative is catastrophic liability.

Undergrounding is not the only answer — covered conductors, steel poles and automated switching are cheaper partial measures — but the direction is unmistakable: the grid is being made physically tougher, and ratepayers are funding it through their bills.

The Capacity Side: The Relief Valve for the Queue

The second half of the spend is where the real ingenuity lives — and where this piece connects straight back to the section’s central bottleneck. Building brand-new transmission lines takes a decade and runs headlong into the interconnection queue and the permitting wall. So the smart money is increasingly on a different approach: get more power through the wires that already exist.

A family of grid-enhancing technologies (GETs) does exactly that, cheaply and fast:

• Dynamic line rating. Sensors measure real conditions and let a line safely carry more power than its conservative static rating — often 10–30% more, for the price of some hardware and software.
• Reconductoring. Replacing old wires with advanced conductors that carry roughly double the current on the same towers — a fraction of the cost and time of a new line.
• Topology optimisation & power-flow control. Software that reroutes power around congestion, like traffic management for electrons.

These are the fast relief valve for the interconnection queue: they add capacity in months to a few years, on existing rights-of-way, without the decade-long permitting fight. Washington has noticed — the DOE’s SPARK programme put roughly $1.9 billion behind accelerated reconductoring in early 2026, and FERC Order 1920 broadened what transmission benefits utilities can recover.

The Bottleneck Within the Bottleneck: Transformers

None of this happens without transformers — the unglamorous steel-and-copper boxes that step voltage up and down at every junction of the grid. And here the whole modernisation programme runs into a hard physical wall: there aren’t enough of them. Demand from data centres, EV charging and reshored manufacturing has collided with a historically low-margin, slow-to-expand manufacturing base, and the result is a worsening shortage with multi-year lead times.

Nearly $1.8 billion of North American manufacturing expansion has been announced, but analysts still expect the pad-mount transformer deficit to widen, warning that extended lead times and elevated costs risk derailing grid modernisation itself. It is the same pattern seen across this whole section: capital is abundant, but a specific physical input — here, a transformer — is the true constraint. The money is ready; the boxes are not.

Reading It Through the Frameworks

This is the most textbook-clean application of the primer’s models in the entire Build thread, and worth stating plainly.

The catch is affordability. Ratepayers are already uncomfortable with rising bills, and regulators are responding by scrutinising spending and emphasising affordability. That is the ceiling on the story: the return is only as secure as the regulator’s willingness to keep approving the spend and passing it through. Watch the regulatory relationship, not just the capex plan.

The Body on the Table

Offshore wind was supposed to be a pillar of American decarbonisation: vast, steady wind close to the dense coastal cities that consume the most power. On paper, it was one of the strongest cases in the whole energy transition. In practice, it became the sector’s most instructive failure — a cautionary tale about what happens when a capital-intensive industry is scaled too fast, on the wrong contracts, in a hostile policy environment.

This piece is deliberately placed in the Build thread as its counterweight. Not everything gets built. Understanding why a well-funded, technically-proven, strongly-supported industry stalled is worth more than another success story — because the failure modes here (contract structure, supply chain, political risk) recur everywhere else in infrastructure. There were two distinct causes of death, and they arrived in sequence.

Cause of Death I: The Economic Heart Attack (2023–24)

The first collapse was purely financial, and it struck the whole Atlantic industry at once. Developers had signed long, fixed-price offtake contracts in the low-rate years around 2019–2021 — committing to deliver power at a set price years before building anything. Then inflation and interest rates spiked together. Offshore wind is about as capital-intensive and long-lead as infrastructure gets — turbines, foundations, ports, specialised vessels, all paid upfront — so a jump in financing costs and equipment prices hit it harder than almost any other asset. The contracts, fixed in a cheaper world, could no longer be delivered profitably.

The cancellations followed fast. In late 2023 Ørsted scrapped the 2,400 MW Ocean Wind 1 and 2 projects off New Jersey, citing rising rates, inflation and supply-chain delays, and took billions in write-downs. It later withdrew from the Skipjack projects in Maryland. The damage was not confined to America: Ørsted discontinued its 2.6 GW Hornsea 4 project in the UK in 2025 on a “challenging business case,” and RWE cancelled a 2 GW Australian project as commercially unviable. The technology hadn’t failed. The financial structure had.

Cause of Death II: The Political Assault (2025–26)

Just as some developers restructured and pushed on, the second blow landed — this time from Washington. In January 2025 a presidential memorandum withdrew all areas of the Outer Continental Shelf from new offshore-wind leasing and paused federal approvals, permits and loans pending review. Over the following year, the administration escalated from blocking new projects to halting ones already under construction.

Stop-work orders hit Empire Wind, then Revolution Wind, and in December 2025 the Interior Department suspended five major projects at once — Vineyard Wind, Revolution Wind, Coastal Virginia Offshore Wind, Sunrise Wind and Empire Wind — citing national-security concerns about turbines interfering with military radar. That rationale was contested: the projects had cleared years of review involving the Coast Guard, Navy and Air Force, and several national-security figures publicly disputed it. The developers sued; courts granted injunctions letting construction continue; and by April 2026 the administration had missed the deadline to appeal, letting the five proceed — for now.

The precise damage is almost beside the point. The deeper lesson is that a fully-permitted, 80%-built project could be halted overnight — that the permit, the thing developers spent nine years securing, turned out to be revocable at political will. For an asset class whose entire premise is long-dated, contracted certainty, that is close to an existential problem.

Why Europe Succeeded Where America Stalled

The contrast with Europe is the analytically useful part, because it isolates what actually went wrong. Europe did not have better wind; it had a better environment for building — assembled patiently over three decades.

The Jones Act is the sharpest example of a self-inflicted wound. This century-old law requires that goods moving between US points travel on US-built, -flagged and -crewed vessels — and virtually no Jones Act-compliant offshore-wind installation vessels exist. Developers were forced into slow, expensive workarounds with feeder barges, adding cost and delay to an already-strained industry. Europe simply used its purpose-built fleet. America made the same job structurally more expensive by law, then acted surprised when it cost more.

The Nuance: Not Dead, Just Bloodied

A fair post-mortem has to note that the patient is not entirely dead. Several projects are delivering power. Vineyard Wind and South Fork are operating; on one December 2025 day, offshore wind supplied nearly 11% of New England’s electricity. The technology works, the resource is real, and the projects that survived the two collapses are producing clean power close to demand exactly as promised.

So the correct reading is not “offshore wind failed” but “the American attempt to scale it too quickly, on fragile contracts, in a reversible policy regime, failed.” That is a subtler and more useful conclusion — because it points to what a durable version would require, rather than writing off the resource.

Reading It Through the Frameworks

Offshore wind is a masterclass in the two risks the primer warns about most, and it is worth being explicit about them.

The revenue model was the first killer. A fixed-price offtake with no indexation converted an ordinary rate shock into an extinction event. The lesson generalises: on any long-lead capital project, the structure of the contract matters as much as its price. Indexed, flexible offtake survives a changing world; a locked price does not.

Policy reversibility was the second. The primer’s principle that “policy is the return” has a dark mirror: when the cash flow depends on a permit, a change of administration can become the dominant risk. Offshore wind priced the engineering and financing risk carefully and the political-durability risk barely at all — and it was the unpriced risk that did the most damage.

The Pivot Nobody Predicted

For two decades, nuclear power in the West was a story of managed decline: plants closing early, projects cancelled, an industry written off as too slow and too expensive for a renewables age. Then the AI build-out ran into the power wall — and in barely eighteen months, every major hyperscaler pivoted to nuclear at once.

The logic is the primer’s framework in action. A training cluster needs power that is firm, around-the-clock, and carbon-free — and nuclear is the only source that delivers all three at scale, running at 95%-plus capacity factors against 25–35% for solar or wind, on roughly fifty acres. Just as important, the hyperscalers brought the one thing the industry always lacked: a credit-worthy buyer willing to sign a twenty-year power-purchase agreement. Collectively they have now committed to something like 10 gigawatts of nuclear capacity across more than a dozen deals.

That last figure is the tension in miniature. The demand is real and enormous. Whether the industry can actually deliver it this decade is the entire question — and the answer depends heavily on which kind of “nuclear” a given deal really means.

Three Very Different Bets Wearing One Label

The headlines lump everything together as “nuclear for AI,” but the deals fall into three tiers with radically different risk and timing. Confusing them is the single most common analytical error in this space.

Microsoft’s deal to restart Three Mile Island Unit 1 — now the Crane Clean Energy Center, an 835 MW reactor closed in 2019 for purely economic reasons — is a restart: proven technology, and its timeline has actually been pulled forward to 2027. Amazon’s Susquehanna arrangement is largely colocation and uprate at a running plant. But Google’s Kairos order, Amazon’s X-energy stake, Meta’s TerraPower and Oklo deals are new-build SMRs — a different universe of risk, delivering in the early-to-mid 2030s.

The Economics — Told Honestly

Nuclear’s cost problem has not gone away; the AI demand has simply made buyers willing to pay it. The numbers are sobering, and they are why this is a bronze-tinted story rather than a green one.

A first-of-a-kind SMR lands around $100–180 per MWh — several times the cost of the existing nuclear it is meant to emulate, and far above renewables. The economics only close with three things stacked together: a carbon-free mandate that rules out cheap gas, a twenty-year PPA that guarantees the revenue, and federal support — production and investment tax credits plus DOE loans. Remove any one and most new-build projects stop penciling. And the cost is a chicken-and-egg problem: SMRs only get cheap after many are built, but few will be built until they are cheap.

The Skeptic’s Case: Timelines Don’t Lie

Against the enthusiasm sits fifty years of the industry missing its own schedules. The cautionary tale is Vogtle Units 3 and 4 in Georgia — the most recent large US reactors, and “proven” AP1000 designs. They still took roughly a decade and ran past $30 billion. If proven technology behaves that way, first-of-a-kind SMRs deserve deep skepticism on both cost and schedule.

Two hard bottlenecks compound the timing problem. HALEU fuel — the higher-enriched uranium many advanced designs require — is barely produced outside Russia, and domestic supply is only now being built. And the nuclear workforce has atrophied through decades of decline, leaving a thin talent pool of licensed engineers and specialised construction crews. Money can be summoned quickly; enriched fuel and trained people cannot.

Signal or Hype? Both — and the Map

The honest verdict is that both readings are correct, for different tiers. The restart-and-uprate story is a real signal: proven assets, twenty-year contracts, power before 2030, and a moat in the finite set of restartable reactors. The new-build SMR story is, for now, mostly hype-adjacent optionality — potentially huge, but unproven on cost and schedule, and unlikely to matter before the 2030s. The investment map has to respect that split.

Why the “+” Matters More Than the Solar

A standalone solar farm has a problem that no amount of cheap panels can fix: it produces a flood of power at midday and nothing after sunset, on a schedule set by the weather rather than by demand. That makes its revenue erratic and its output merchant — and, by the primer’s framework, keeps it at the risky, uncontracted end of the spectrum. It is generation, but it is barely infrastructure.

The “+” — a co-located battery — fixes exactly that. By storing the midday flood and releasing it when the grid needs it, storage converts an erratic, price-taking output into a firm, dispatchable, schedulable product. That single change moves the asset up the risk spectrum: from merchant toward contracted, from volatile toward bankable. The market has already voted. In California, more than 92% of the solar capacity now seeking to connect to the grid includes storage. The battery is no longer an accessory to the solar farm. It is the part that makes the solar farm an infrastructure asset.

Why Now: The Duck Curve Forced It

The urgency comes from solar’s own success. As more solar floods the grid at midday, the midday price collapses — sometimes to zero or negative — while the value concentrates in the evening ramp after the sun sets. This is the “duck curve,” and it has been steepening for years. For a standalone solar farm it is an existential threat: it produces most when its power is worth least.

Storage is the natural hedge. It buys the midday glut for nothing and sells into the evening scarcity, turning the duck curve from a problem into a revenue opportunity — and as battery costs have fallen, that arbitrage has become the default design. With variable renewables set to rise from roughly a third of global generation today toward well over half within a decade, the grid increasingly cannot absorb more raw variability. It needs firm output. The “+” is how renewables supply it.

The Breakthrough: Firm Renewables Now Beat Gas

For years the honest case against renewables was that firming them — making them reliable around the clock — was so expensive that gas remained cheaper for dependable power. As of 2026, that argument has broken. A May 2026 analysis from IRENA put the cost of firm, round-the-clock solar-plus-storage at roughly $54–74 per MWh, against $60–95 for new combined-cycle gas. Firmed renewable power now undercuts newly-built gas on cost — not raw solar versus gas, but the genuinely comparable product: power available when you actually want it.

Co-Location: One Connection, Two Assets

There is a second, quieter advantage to the hybrid, and it ties straight back to the section’s central bottleneck. Building the solar and the storage behind a single interconnection point means one grid connection instead of two — one queue position, one set of network-upgrade costs, one study. In a system where interconnection is the scarce, years-long constraint, sharing a connection is not a minor efficiency. It is a way to get twice the useful asset through the same narrow gate.

It also uses the connection more fully. A solar farm alone leaves its expensive grid link idle for much of the day; add storage and the same wire exports closer to its rated capacity for far more hours. The hybrid is, in effect, a way to squeeze more value out of the scarcest thing in the whole system — the connection itself.

The other flavours of “+”

Storage is the dominant “+,” but not the only one. Hybrid wind-plus-solar pairs two sources with complementary profiles — wind often blows when the sun doesn’t — smoothing output before storage is even added. Agrivoltaics stacks solar over farmland, letting one parcel earn from both crops and electrons, easing the land-use conflicts that slow projects. And longer-duration storage — beyond today’s roughly four-hour lithium batteries — is the next frontier, extending firmness from hours toward days.

Reading It Through the Frameworks

How does it get paid? The whole point of the “+” is to move the revenue model leftward on the risk spectrum — from merchant solar (price-taking, volatile) toward contracted, firm supply that can sign a long offtake. The battery is a risk-transformation device as much as a technical one.

Where is the moat? Not in the panels or the cells, which are global commodities. It is in the interconnected hybrid site (scarce, per the queue) and in the dispatch software — the optimisation that decides when to store and when to sell across three revenue streams. Running a hybrid well is a trading problem, and the firms that master it capture disproportionate value.

Where does policy become the cash flow? Storage tax credits, capacity-market rules that reward firmness, and a growing set of mandates requiring new renewables to include storage all convert directly into project economics.

The Inheritance

Walk onto the site of a recently retired coal plant and the obvious assets — the boilers, the turbines, the smokestack — are mostly worth their weight as scrap. The valuable thing is invisible. It is the point of interconnection: the high-voltage substation and transmission tie that was built to push a gigawatt of power out to the grid, and which can just as easily pull a gigawatt back in.

The hub piece on the interconnection queue established why that matters: a new project on bare land waits four to seven years for a grid connection it may never get. A retired power plant has that connection already built and energised — along with water rights, cooling infrastructure, transmission access, rail, road, and a workforce that knows the site. In a market where the binding constraint is time-to-power, inheriting all of that is not a discount. It is the entire investment thesis.

The Coal-to-X Menu

Once you see the site as a live connection, the question becomes what to plug into it. There is a growing menu — the industry calls it “coal-to-X” — and each pathway reuses the same inherited infrastructure for a different purpose:

The nuclear pathway is the most striking, because the physical and social fit is so close. A retired coal plant already has the turbine hall, the cooling, the transmission, a trained industrial workforce, and a community whose economy was built around baseload power. Dropping a reactor onto that foundation, rather than fighting for a virgin site, is both cheaper and faster.

Real projects are already moving. PacifiCorp selected the retiring Naughton coal site in Wyoming for an advanced sodium-cooled reactor with molten-salt storage; Romania picked a coal site at Doicesti for its first small modular reactor. At least eleven US states have publicly backed the coal-to-nuclear idea.

The gas twist: reusing the pipes, not just the wires

There is a parallel second life on the gas side. An existing gas plant — and the pipeline network feeding it — can increasingly run on renewable natural gas: biomethane captured from landfills, dairy digesters and wastewater. The molecule is nearly identical, so the entire installed base of turbines, pipes and storage can keep operating on a lower-carbon fuel without being rebuilt. It is the same logic as coal-to-X, applied to the gas system: reuse the infrastructure, change what flows through it.

Why the Value Is Spiking Now

Second-life sites have existed for years; what changed is that three forces converged to make the inherited connection suddenly precious.

First, the interconnection queue made the existing connection scarce. Second, soaring capacity prices — PJM’s roughly 800% auction jump — made firm, connected capacity extraordinarily valuable. Third, policy: federal programmes specifically reward reusing these sites, with brownfield grants, loan guarantees, and clean-energy tax credits that carry bonus “energy community” adders precisely for former fossil-fuel sites. The site that was a stranded liability three years ago is now a subsidised head start.

Reading It Through the Frameworks

Where is the moat? It is the most durable kind there is: you cannot manufacture new interconnected sites. The supply is fixed — it is exactly the set of power plants that were built decades ago — while demand for connection points rises every quarter. Whoever controls a portfolio of retiring, connected sites owns an appreciating, non-replicable asset.

Where does policy become the cash flow? Directly. The “energy community” tax-credit bonus, brownfield remediation grants and federal loan guarantees are not background — they can swing a coal-to-X project from marginal to compelling, and they exist specifically to steer capital onto these sites.

The Heat Problem

The previous pieces were about getting power into the data centre. This one is about the law of physics that follows immediately: nearly every watt that goes in comes back out as heat, and it has to go somewhere. For most of computing history that was a trivial afterthought — a few fans, some cold air. AI broke that assumption in a single hardware generation.

The driver is rack density. A traditional server rack draws 5 to 15 kilowatts, and air cooling handles it comfortably. But an NVIDIA Blackwell GB200 rack draws around 140 kW — roughly ten times as much heat, in the same physical footprint. Air simply cannot carry that much energy away fast enough; at those densities, chips throttle their own clock speeds to avoid cooking, and the expensive GPUs you paid for stop delivering the performance you bought. Next-generation designs push toward 900 kW per rack and single chips near 2,000 watts. The industry’s own thermal standards body now recommends liquid cooling above 20 kW per rack — a line virtually every serious AI deployment crossed in 2025.

The Technologies — and How They Compare

The shift underway is from moving air to moving liquid, because liquid carries heat up to a thousand times more effectively. There are three broad approaches, in ascending order of density and complexity:

• Direct-to-chip (cold plate). A metal plate sits directly on the GPU, and coolant flows through it, carrying heat away at the source. It retrofits into existing racks with minimal disruption, which is why it dominates today — more than half the market.
• Immersion cooling. Entire servers are submerged in a non-conductive dielectric fluid. It handles the highest densities and slashes water use, but requires purpose-built tanks and specialised fluids — a bigger commitment.
• Air (the baseline). Still the incumbent for lower-density workloads, increasingly supplemented by rear-door heat exchangers as a transitional step.

The efficiency gap is stark, and it is measured in PUE — power usage effectiveness, the ratio of total facility power to the power actually reaching the computers. A PUE of 1.0 is the theoretical ideal (no overhead); everything above it is energy spent on cooling and losses.

The likely future is not one winner but a dual track: direct-to-chip serving the mainstream because it retrofits easily, immersion taking the ultra-dense and water-constrained deployments. Both are liquid; the air era is ending for anything running AI.

The Market — and the Business Model Underneath It

Cooling has crossed its inflection point. Liquid cooling penetration was around 3% in 2021; by 2026 it is roughly 37% — a more-than-tenfold jump in five years, driven by hardware that leaves no choice.

The Water Tradeoff — Where Build Meets Strain

Here is the catch that turns a growth story into a constraint. The cheapest way to reject heat is evaporative cooling — letting water evaporate to carry heat away — and it is thirsty. A conventional evaporatively-cooled data centre consumes on the order of 2 to 5 million gallons of water per megawatt, per year. Scale that across a gigawatt-class AI campus and the number becomes a genuine claim on a regional water supply.

In the American Southwest — Phoenix, Las Vegas, much of Texas — water rights are finite and increasingly contested, and local permitting reviews now scrutinise a data centre’s water draw as closely as its power draw. Water, in other words, has joined the interconnection queue as a gate that can stop a project before it starts. This is where the cooling choice becomes a siting decision, and where the Build thread runs straight into the Strain thread.

The escape is a genuine three-way tradeoff, with no free option:

• Evaporative cooling: lowest energy, highest water — fine where water is cheap, disqualifying where it isn’t.
• Dry / adiabatic coolers: minimal water, but higher energy use — you trade the water bill for the power bill.
• Immersion: cuts water use 90–98% and improves efficiency — but demands specialised fluid and purpose-built design.

Reading It Through the Frameworks

Where is the moat? Not in the commodity hardware — cold plates and pumps will commoditise. It sits in three places: proprietary thermal IP and the coolant-distribution systems that are hard to replicate; the retrofit lock-in of a cooling loop that, once installed, is expensive to swap; and the service annuity attached to it. Own the recurring relationship, not the box.

Where does policy become the cash flow? In two ways. Water-permitting rules increasingly mandate low-water cooling in scarce regions, effectively legislating demand for immersion and dry cooling. And PFAS regulation cuts the other way — the phase-out of certain fluorochemical fluids used in two-phase immersion creates a real transition risk for that specific technology, and a cost advantage for single-phase and fluid-free approaches.

What the Queue Actually Is

Before any power plant, wind farm, battery or data centre can plug into the grid, the grid operator has to study what connecting it would do — whether the surrounding wires can carry the extra load, and what upgrades would be needed to keep the system stable. That study, and the wait to receive it, is the interconnection queue. Nothing connects until it clears.

The study is not a formality. It determines the single most important number in a project’s budget: the cost of the network upgrades required to accommodate it — which can range from trivial to project-killing, and which the developer usually has to pay. Until the study is done, that cost is unknown, financing can’t close, and the project can’t proceed. The queue, in other words, is where a project’s economics are decided by someone other than its owner — and where most projects quietly die.

This piece is a hub: several other posts in Bifrost Systems — the AI demand shock, the colocation bypass, grid modernisation, second-life assets — all run into this same wall. Rather than re-explain it each time, this is the reference. If you understand the queue, you understand why so much of the infrastructure decade moves in slow motion.

Why It Broke

For decades, the queue worked on a “first-come, first-served” basis: projects were studied one at a time, in the order they applied. That was fine when a handful of large power plants joined the grid each year. It fell apart when tens of thousands of smaller solar, wind and storage projects — cheap to propose, easy to file speculatively — flooded in at once.

Two design flaws turned a backlog into gridlock. First, cascading restudies: because each project’s upgrade costs depended on everything ahead of it, whenever a higher-placed project withdrew, everything behind it had to be studied again, resetting the clock. Second, speculative squatting: with little cost to hold a place in line, developers filed far more projects than they intended to build, clogging the queue with applications that were never real. The result is a queue full of “zombie” projects — occupying position, triggering restudies, and never intending to connect.

The attrition is the part outsiders miss. The headline queue figure — 2,200 gigawatts, larger than everything currently plugged in — wildly overstates what will actually get built, because most of it never does.

The Fix — and Its Limits

In July 2023, FERC issued Order No. 2023, described by the commission as its largest interconnection reform in two decades. Its central move was to replace “first-come, first-served” with “first-ready, first-served.” Instead of studying projects one by one in filing order, grid operators now study them in clusters, all at once, and prioritise the ones that can demonstrate genuine commercial readiness — site control, financial deposits, real intent to build. The reform also imposes firm study deadlines, with financial penalties on grid operators that miss them.

The logic is sound: cluster studies stop the cascading restudies, and readiness requirements price the zombies out. But the rollout has been slow and uneven. FERC had to order PJM — the largest US grid operator, serving 65 million people — to redo its compliance plan in 2025 for not meeting the rule. Regions are now layering on their own expedited “fast lanes” for shovel-ready projects: PJM’s expedited track (accepted mid-2026) targets a roughly ten-month path to a signed agreement, and MISO and SPP have their own accelerated study processes. Even so, the backlog has kept growing, swollen by a wave of new solar-plus-storage applications.

Reading It Through the Frameworks

This is the canonical test of the primer’s central question, so it’s worth working through carefully — the logic recurs across the whole section.

That reframing explains several other pieces in this section at once. It is why colocation exists — the bypass is an attempt to avoid the queue entirely. It is why a retired coal or gas site is valuable beyond its hardware — the interconnection rights come with it. And it is why the relief technologies (reconductoring, grid-enhancing tech) are an investable theme in their own right: anything that moves more power through existing, already-connected wires sidesteps the queue by definition.

The Investment Map

If the scarce thing is an interconnected position, the map follows directly — and it splits cleanly into owning the scarcity versus owning its relief.

The Bypass Logic

The previous piece ended on a bottleneck: a grid connection now takes four to seven years, while the data centre that needs it can be built in one. When the wait is that lopsided, the rational move is not to wait — it is to build the data centre where the power already is, and skip the grid entirely.

That is colocation. Instead of connecting a data centre to the grid and drawing power across it, you site the facility directly beside a generator — a nuclear plant, a gas plant, a solar-plus-storage complex — and wire it straight in. In the purest form, the load sits “behind the meter”: it never touches the public grid at all, so it never joins the interconnection queue, never waits on new transmission, and can energise in a year or two rather than half a decade. For a hyperscaler racing rivals for compute, that time saving is the entire game.

The appeal is obvious, and so is the problem hiding inside it. A plant that used to sell its output to the grid, serving everyone, now sells it to a single data centre behind a fence. The electrons haven’t changed — but who they serve has, and that is where a straightforward engineering shortcut turns into one of the most contested questions in US power policy.

The Case That Wrote the Rules: Talen, Amazon & Susquehanna

No single deal has shaped this debate more than Amazon’s arrangement with Talen Energy at the Susquehanna nuclear plant in Pennsylvania. Its two-year journey — from quiet acquisition to national test case — is effectively the story of how colocation went from clever workaround to regulated activity.

The Fight Underneath — and Who Pays

Colocation has quietly split the power industry in two. On one side are the independent power producers — Constellation, Talen and their peers — who own generation and see colocation as the fastest way to sell it at a premium to credit-worthy hyperscalers. On the other are the regulated utilities — Exelon, FirstEnergy, PPL, AEP — who own the grid and argue that behind-the-meter colocation lets data centres free-ride on a network they still depend on for reliability.

The utilities’ strongest argument is physical, not commercial. When a nuclear plant that used to serve the grid is redirected behind the meter to a single data centre, that capacity effectively leaves the shared system. Everyone else’s supply just got tighter — and in a market already short of power, tighter supply means higher prices for every other customer. The data centre gets its electrons; the ratepayer gets the bill.

This is what turns a Build-thread topic into a Strain-thread problem in the same breath. The speed that makes colocation attractive is inseparable from the cost-shift that makes it politically radioactive — and the politics is now the primary risk to the economics.

Reading It Through the Frameworks

How does it get paid? A colocation deal is a bilateral contract between a generator and a single, usually investment-grade, buyer — closer to a long-dated corporate PPA than to merchant power. That contracted, take-or-pay shape is genuinely attractive. The catch is that its regulatory standing, not its counterparty, is the swing variable: the Talen saga showed that a signed deal can be unwound by a regulator who decides the structure shortchanges the grid.

What stage is it at? Colocating at an existing plant is brownfield on the generation side — the reactor already runs — but greenfield on the load side, and the combination sits in value-add territory. The economics lean heavily on the speed premium, which erodes the moment the regulatory path lengthens.

The Investment Map — and the Tail Risk

If the scarce asset is generation sited next to demand, the winners are the owners of that generation — and the losers are whoever gets stuck holding a structure the regulator later rejects.