One System Now: AI, Electricity, Inflation, and the New Geometry of Geopolitics

01 The System, Assembled

This analysis is a capstone: it takes four strands developed separately in this series — offshore wind federalism, network resilience, the build-capacity policy paradigm, and the AI demand accelerant — and joins them into the system they now actually form. The claim to be examined is that AI, energy, inflation, and geopolitics have become one interconnected system: AI requires capital and electricity; energy shocks affect inflation; inflation affects interest rates; rates determine AI financing costs; and the supply chains that all of this runs on intersect U.S.–China competition. The claim is not a metaphor. Each named link is a real transmission channel with observable quantities at both ends, and the loops close — which is what makes it a system rather than a list of related topics.

State the wires precisely. Wire one: AI capability competition drives capital expenditure by a small number of hyperscale firms, and that expenditure is physically embodied in data centers — buildings full of processors, cooling, and above all electricity demand, arriving at the grid in large, firm, contiguous blocks. Wire two: that demand meets a power system whose build-rate is the binding constraint analyzed in the build-capacity pivot — interconnection queues, transformer lead times, transmission scarcity — so AI growth converts directly into competition for grid capacity against every other load and every other clean-build ambition. Wire three: energy shocks — supply disruptions, fuel price spikes, the geopolitics of hydrocarbons that the Wikipedia reference summary for Energy security describes when it defines energy security as the association between national security and the availability of natural resources for energy consumption, noting that access to cheaper energy has become essential to modern economies while the uneven distribution of supplies among countries creates significant vulnerabilities (source: Wikipedia summary — Energy security) — feed into consumer and producer prices, which is the classic inflation channel of every energy episode since the 1970s. Wire four: inflation shapes central-bank policy, policy shapes interest rates, and interest rates set the discount rate under which AI capex — long-duration, revenue-back-loaded, almost entirely financed in capital markets — is evaluated. Wire five: the chips, batteries, transformers, turbines, and mineral inputs for every physical node of the system sit in supply chains whose chokepoints are objects of great-power competition, so the system's throughput is hostage to the trade policy of the two powers that are also the system's principal competitors.

Why assemble the wires now, rather than five years ago or five years hence: the system exists only when all the loops are simultaneously load-bearing, and the load test is now. AI capex is at a scale that moves the grid planning of entire interconnections; electricity demand growth is at a scale that forces the build-capacity policy paradigm; energy-price volatility is at a scale that monetary policy must treat as an inflation input rather than noise; and supply-chain competition is at a scale where chokepoints — advanced lithography, grid equipment, critical minerals — are explicitly governed as strategic terrain. Remove any one wire five years ago and the loops opened; the components could be analyzed separately, and mostly were. Today they cannot.

The core systems insight to carry through everything below: each loop in this system is individually stabilizing or destabilizing in a known way, but the system's behavior is dominated by how the loops couple — stabilizing loops can neutralize each other, and destabilizing loops can reinforce each other through channels no single-domain analyst watches. The rest of this analysis is, in order: the loop inventory, the coupling analysis, the historical analogy and its limits, the second- and third-order structure, the scenarios at system level, and the indicators that reveal which system behavior is materializing.

02 The Loop Inventory: Five Loops, Three Stabilizing, Two Destabilizing

A feedback loop is destabilizing when it amplifies deviations and stabilizing when it corrects them. The system contains at least five major loops, and the inventory is the analysis's backbone.

Loop one — the demand-finance loop, destabilizing. Rising AI capex raises electricity demand; rising demand tightens power markets and raises delivered electricity prices; higher energy prices feed inflation; inflation raises policy rates; higher rates raise the cost of the capital that AI capex requires; and here the loop's arm forks. In the benign fork, higher financing costs suppress capex, which relieves demand, which stabilizes the system — a stabilizing correction, and the one equity markets implicitly bet on. In the malign fork, capability competition makes capex inelastic to financing cost — firms believe the returns to being first are so large that they build through high rates, absorbing the cost as competitive insurance — and the loop then amplifies: capex persists, demand pressure persists, energy pressure persists, and monetary tightening tightens against a demand source it cannot reach, because it is not interest-sensitive. The loop's behavior therefore hinges on an empirical parameter: the interest elasticity of AI capex. It is unmeasured at any useful horizon, and it is the single most important unknown in the system.

Loop two — the build-capacity loop, stabilizing, with delay. Demand growth justifies and finances infrastructure: rate cases, offtake contracts, transmission projects, and manufacturing scale all expand under demand certainty, and the expanded capacity eventually relieves the tightness that caused it. The catch is the delay: the loop's correction arrives years after its provocation — transmission on decade timescales, generation on multi-year ones, manufacturing capacity between — so the loop oscillates rather than settles: today's scarcity drives overbuilding that becomes the glut that collapses investment, which becomes the next scarcity. This is the classic hog cycle of agricultural economics, transposed to electrons, and the summary's observation that uneven distribution of energy supplies creates vulnerabilities (source: Wikipedia summary — Energy security) is the geopolitical residue the oscillation leaves in procurement patterns.

Loop three — the substitution loop, stabilizing. Energy price pressure accelerates the substitution of domestic renewable generation for imported fuel: solar, wind, storage, and firm clean capacity are the structural hedge against exactly the hydrocarbon shocks that drive the inflation channel. The loop is stabilizing because the substitution reduces both the trade-vulnerability input and the price-volatility input to inflation. Its weakness is the one this series developed in the federalism and build-capacity analyses: substitution runs through the permitting and interconnection pipeline, so the loop's stabilizing force is throttled by the slowest institutional throughput in the system. The stabilizing loop exists, but its bandwidth is limited by administrative law.

Loop four — the competition loop, destabilizing. Great-power competition converts supply chains into strategic terrain: export controls, tariff walls, domestic-content mandates, and mineral stockpiling fragment what were efficiency-seeking global supply chains, raising the cost and reducing the elasticity of every physical input — chips for AI, transformers for grids, batteries for storage — exactly when throughput demand on those inputs is highest. Competition also acts on the AI capex source itself: the belief that being second in capability carries strategic costs makes the capex less elastic to all prices, reinforcing loop one's malign arm. This loop amplifies every other loop's costs, and unlike the others it has no self-correction — its stabilizing arm is the recognition of mutual dependence, which the great-power relationship now provides only episodically.

Loop five — the macro-absorption loop, stabilizing, contingent. The AI build-out, whatever else it is, is a massive investment demand shock, and investment booms are absorbed differently depending on whether the economy has slack: with idle capacity, capex expands output; with full employment and constrained supply, it bids up prices — power, construction, skilled labor — and the boom partially finances its own inflation. The loop stabilizes when productivity from deployed AI systems arrives in time to raise output against the added demand; it destabilizes when the productivity arrives late or small, leaving the demand without the supply. The loop's parameter is the productivity payoff of AI, its timing and magnitude — the second great unknown, and unlike the capex elasticity, it is a quantity economists have spent decades failing to predict for general-purpose technologies in advance.

The inventory's headline: three stabilizing loops, two destabilizing ones, and — this is the capstone point — the two destabilizing loops act fastest while all three stabilizing loops act slowly or contingently. The system's characteristic behavior, unless the parameters are favorable, is overshoot: the fast loops push before the slow loops can pull. That asymmetry of timescales, not any single loop's sign, is the system-level finding.

03 Coupling Analysis: How the Loops Interact

The map is static; the system is the couplings. Four couplings dominate, and each pairs a fast loop with a slow one in a way that concentrates risk.

Coupling one: loop one and loop two — the rate treadmill against the build queue. The destabilizing demand-finance loop acts in months: capex decisions, power contracts, price signals all move quickly. The stabilizing build-capacity loop acts in years to decades. The coupling means that the system's demand grows on financial-market timescales while its supply grows on infrastructure timescales, and the gap between those timescales is filled by scarcity pricing — electricity price spikes, construction cost inflation, interconnection-premium capital rationing — which feeds back into the inflation channel the fast loop already owns. The system is, in effect, running its demand on a clock its supply cannot match, and the difference is booked as inflation. This is the mechanism by which the AI build-out becomes a monetary phenomenon, and it is why central banks now watch interconnection queues the way an earlier generation watched oil rigs.

Coupling two: loop four and loop three — competition throttles the hedge. The destabilizing competition loop acts on the substitution loop's instruments: transformers, batteries, minerals, and increasingly the advanced chips that make AI systems efficient per watt. Trade fragmentation raises the cost and lengthens the lead time of exactly the equipment the stabilizing substitution loop needs to relieve energy pressure. The coupling converts a strategic policy — supply-chain security — into a macroeconomic cost, and it means the system cannot be stabilized by energy policy alone: the substitution hedge is governed as much in trade dockets as in rate cases. The summary's observation that international energy relations have contributed to globalization while creating energy vulnerability at the same time (source: Wikipedia summary — Energy security) is the standing description of this coupling's deeper version, and the AI era adds a new expression: compute dependencies alongside fuel dependencies.

Coupling three: loop one and loop five — capex elasticity versus productivity timing. Both great unknowns couple directly. If capex is inelastic to rates because the capability race dominates, and productivity arrives late, the system runs demand without supply for years, and every slow stabilizing loop is pressured to deliver early: build-capacity must build faster than any modern precedent, substitution must clear permitting faster than administrative law allows, and the competition loop's costs land on a system already stretched. If capex is elastic and productivity arrives on time, the system behaves almost conventionally — a strong investment cycle with a soft landing. The two parameters are correlated in the malign direction and uncorrelated in the benign one: a capex retrenchment (elastic capex) typically coincides with disappointed capability expectations (late productivity), which is why the benign fork of loop one and the benign fork of loop five rarely co-occur. The system's best case and its second-worst case share a trigger, and which one materializes depends on why capex slowed — discipline after success is the benign version; exhaustion after over-promising is the malign one.

Coupling four: loop four and loop one — competition sets the race's stakes. The competition loop acts on the demand-finance loop's core parameter: if being second in AI capability carries unacceptable strategic costs, capex becomes inelastic to financing cost by policy rather than by commerce, and the malign fork of loop one becomes structural rather than cyclical. This is the coupling that makes the system geopolitical rather than merely macroeconomic: the interest elasticity of AI capex — the system's most important parameter — is partly set in Beijing and Washington, not on trading floors. It also creates the system's one genuine stabilizer at the strategic level: a U.S.–China crisis-communications or mutual-restraint mechanism for AI, of the kind both governments have explored, would convert the race's stakes from existential to positional, re-introducing rate sensitivity into capex and with it the fast loop's self-correction. Arms-control logic, transposed to compute, is thus macroeconomic stabilization policy at system scale.

04 Historical Analogy and Its Limits

The obvious analogies — the railway booms, the electrification era, the dot-com cycle — each capture one loop and miss the system, and the discipline of stating what differs matters more here than in any single-domain analysis this series has run.

The railway booms of the nineteenth century are the closest structural analogue: transformative infrastructure, demand projections that outran settlement, capital markets financing long-horizon assets, and repeated episodes of overbuilding, consolidation, and financial crisis. What is similar: the co-movement of a general-purpose infrastructure build with speculative finance; overbuilding gluts that follow scarcity pricing; the state's role in land grants and standards standing behind private capex. What is different: the railways competed for nothing at the systems level — there was no great-power rival racing to lay the same track, and no monetary authority setting a policy rate against which the entire capex was evaluated. Why it matters: the railway analogy predicts the hog cycle — loop two's oscillation — but says nothing about the competition loop or the macro coupling, which is where the present system's distinctiveness lives. The railways oscillated within a national economy; the AI-energy system oscillates inside a geopolitical race, and the race changes the oscillation's damping.

The 1970s energy-inflation era is the analogy for the inflation channel: oil shocks feeding headline inflation, inflation feeding monetary tightening, tightening feeding recession, recession reducing demand until energy prices relented. What is similar: the energy price to inflation to rates to real-economy transmission is textbook and was learned in that decade at great cost. What is different, and the difference is total: the 1970s shocks were supply-side — the demand for energy was flat and the supply moved. The present system's pressure is demand-side — the supply of power is constrained and the demand is growing. Why it matters: the 1970s playbook, demand suppression through monetary tightening, works against supply-driven inflation; it works slowly and expensively against demand-driven energy pressure, because the demand source — the capability race, loop one — may not be interest-elastic. A central bank facing the 1970s had a working tool; a central bank facing this system has a tool whose efficacy depends on the unmeasured parameter, and the 1970s analogy is therefore actively dangerous as a guide: it recommends the one instrument whose main effect, if capex is inelastic, is to compress the rest of the economy while leaving the pressure source intact.

The dot-com cycle is the analogy for loop five's fork: transformative technology, massive capex, delayed productivity, retrenchment, and then — a decade later — the productivity payoff arriving at scale. What is similar: the general shape of over-promise, crash, and belated vindication; the pattern of infrastructure overbuild whose capacity is later absorbed. What is different: the dot-com capex was small relative to the economy and its power demands were trivial; it did not couple to the grid, to energy prices, or to the inflation channel at all. Why it matters: the dot-com analogy predicts loop five's benign resolution — productivity arrives late but arrives — while the physical scale of the AI build-out means the system's macro and grid footprints are now large enough that the retrenchment phase, if it comes, will itself be an energy-and-inflation event, not merely a Nasdaq one. The dot-com crash was macroeconomically survivable precisely because its wires did not reach the energy system. The present system's wires do, which means its downside scenario is not a repeat of 2000 but something with an energy-price signature at its core.

05 Effects, Orders, and the System's Distributional Ledger

The system's effects propagate in orders, and at system level the standard first-, second-, third-order framing needs one amendment: the orders run concurrently, because the loops close faster than any single chain completes.

First order, now observable in structure: AI capex flowing into data-center construction; interconnection queues lengthening; power-market price signals tightening in constrained regions; grid-equipment and chip lead times extending. Second order: the build-capacity policy paradigm hardening — permitting, procurement, financing institutions reorganizing around throughput, as developed in this series' policy analysis; electricity price pressure feeding into measured inflation via the energy and increasingly the services channel; and rate-policy responses raising financing costs for every long-duration asset in the economy, not least the clean-energy build-out itself, which is the system's quietest irony: the AI demand that accelerates the clean build-out simultaneously raises its cost of capital through the inflation channel. Third order, labeled explicitly as model-based projection: if the malign fork of loop one persists — capex inelastic through high rates — the system reaches a regime where monetary policy loses traction against a strategically-mandated demand source, forcing fiscal or industrial policy to do the demand management that rates cannot; if the benign fork holds, the system converges to the strongest productivity wave since electrification, with surplus grid capacity — the overbuild from the fear case — becoming the cheap-power base for the next industrial era, in a repeat of fiber's afterlife after the dot-com bust. Both third orders are plausible. Neither is a forecast. The scenario structure below organizes them.

The distributional ledger, because a system analysis without one is a model without its costs. Within economies: the system's inflation channel is regressive — energy and electricity prices weigh heaviest on low-income households — while its asset channel is progressive only for those who hold the equity of the firms building it; the gap between those two channels is the system's political vulnerability in every democracy that hosts it. Across economies: the summary's observation that uneven distribution of energy supplies among countries creates significant vulnerabilities (source: Wikipedia summary — Energy security) extends to compute, chips, and grid equipment; developing economies face a system in which the scarce inputs to their own transitions are bid up by the great powers' race, and in which the choice of supplier is no longer commercial but geopolitical. Across generations: the system is financing present capability competition with debt and with grid build-out whose costs ratepayers carry forward regardless of whether the AI revenue materializes — a transfer from future electricity customers to present AI capability, enacted through utility rate cases one at a time, never announced as a policy, and therefore never debated as one.

06 System-Level Scenarios: A, B, C

At system level, the three scenarios are regimes, not episodes: coherent configurations of the loop parameters that, once entered, persist because the loops reinforce them. They are conditional constructions, not forecasts; no probabilities are assigned, though what would distinguish them is stated with each.

Scenario A — Coherent Build-out, the stabilized regime. The system's fast loops discipline themselves: AI capex proves moderately rate-elastic as capability expectations mature, so the demand-finance loop's benign fork operates; productivity from deployed systems arrives on a horizon that lets loop five absorb the investment demand; the build-capacity loop delivers — permitting compresses, interconnection throughput rises, manufacturing scales — so the slow stabilizing loops catch up with the fast destabilizing ones, and scarcity pricing decays into ordinary commodity cyclicality. Trigger: the conjunction of demonstrated AI productivity, surviving permitting reform, and a U.S.–China relationship stable enough that the competition loop stops escalating. Transmission: energy-price volatility fades as an inflation driver; rates normalize; the clean build-out's cost of capital falls with them, and the transition accelerates for macro reasons rather than despite them. Indicators to monitor: measured productivity-relevant AI adoption in firm-level data; interconnection-completion rates; the ratio of announced to contracted data-center capacity; trade-flow stability in grid equipment; core inflation's energy-services contribution. Consequence: the system settles into the strongest investment-led expansion since the postwar electrification programs, with the geopolitics of compute institutionalized rather than resolved — the race continues, but within a stable macro frame.

Scenario B — Squeezed Oscillation, the persistent-incoherence regime. The parameters split: capex stays inelastic while productivity stays late, and the build-rate stays throttled — so the fast loops push, the slow loops fail to pull, and the system oscillates at a higher mean level of energy-price and financing-cost pressure. This is not collapse; it is chronic strain: recurring regional power scarcity, recurring inflation contributions from energy-services, rates held higher than the rest of the economy warrants, and a political economy that frays accordingly. Trigger: the malign fork of loop one without the malign fork's full consequence — capability competition stays hot, but deployment revenue grows fast enough to keep the capex funded, just not fast enough to relieve the system. Transmission: the hog cycle runs at full amplitude — scarcity, overbuild, glut, retrench, scarcity — with each oscillation leaving stranded costs on ratepayers and each trough weakening the manufacturing base the next peak needs. Indicators: cyclical energy-services inflation; data-center interconnection cancellations followed by fresh queues; capacity-factor spreads between constrained and unconstrained grid regions; the gap between AI capex announcements and utility-certificated load. Consequence: a decade of the transition underperforming its own physics — the build-capacity paradigm's institutions eroded by oscillation rather than defeated by any single event, and the summary's uneven-distribution vulnerability (source: Wikipedia summary — Energy security) deepening as procurement chases whichever supplier is least constrained this cycle.

Scenario C — Fragmented Escalation, the destabilized regime. The competition loop dominates: a Taiwan-strait crisis, a chip-chokepoint rupture, or a mutual-goods trade rupture severs the supply chains the entire system runs through, and the severance converts the slow stabilizing loops' inputs into contested terrain overnight. Compute, transformers, and minerals all reprice at strategic-scarcity levels simultaneously; the substitution loop and the build-capacity loop stall for want of equipment; the AI capex source, now a security program rather than a commercial one, is fiscalized — public money substituting for private as the market's discount rate becomes irrelevant to strategic demand; and the energy-inflation channel runs hot under scarcity that monetary policy cannot touch. Trigger: geopolitical shock to any of the system's physical chokepoints, or a deliberate escalation of export controls into a blockade logic. Transmission: energy security becomes compute security becomes food-and-fuel security, in a cascade of the vulnerabilities the reference summary names; inflation reverts to 1970s-style supply-driven dynamics — and the 1970s playbook, with all its costs, becomes correct again, because the shock is now supply-side. Indicators: export-control scope expansion; strategic stockpiling of chips, minerals, and grid equipment; hyperscaler capex migrating from market-disclosed to security-classified channels; grid-construction cost indices decoupling from general construction costs. Consequence: the system's loops all destabilize together, and the stabilization task migrates from macroeconomic policy — which cannot reach the chokepoints — to geopolitical management, which is to say: the system's end-state is decided in the same arena where its most destabilizing loop was born.

07 Indicators and the Bottom Line

Ten indicators — two per loop — will reveal which regime the system is entering before any commentary does. For loop one: the growth rate of hyperscaler capex against policy-rate movements, which is the interest-elasticity parameter revealing itself in data; and the ratio of announced to contracted data-center capacity, which separates intent from load. For loop two: interconnection-queue completion rates, the build-capacity loop's cleanest throughput measure; and grid-equipment lead times, the substitution loop's slowest input and a leading indicator of the entire slow-stabilizing cluster. For loop three: the share of new capacity additions that are firm clean generation, which measures the hedge's actual deployment against its political promotion; and the energy-services contribution to core inflation, which is the inflation channel's live gauge. For loop four: the breadth of export-control scope and the pace of strategic-stockpile accumulation, which measure the competition loop's escalation; and trade-flow concentration in chips, minerals, and grid equipment, which measures how much of the system runs through how few doors. For loop five: firm-level productivity data from sectors with heavy AI deployment, arriving eventually in official statistics; and the spread between AI-related and aggregate private investment growth, which shows whether the boom is broadening into general capital deepening or remaining a sectoral race.

The bottom line, in the evidence-strength register the standard requires.

What we know: AI capex, electricity demand, energy prices, inflation, interest rates, and great-power supply-chain competition are each measurable quantities with demonstrated causal links between adjacent pairs — the wires are real, and each has been observed operating in this series' component analyses; energy security is the association between national security and energy availability, with uneven supply distribution creating documented vulnerabilities (source: Wikipedia summary — Energy security); and the timescale asymmetry — fast destabilizing loops, slow stabilizing ones — is a structural property of the couplings, not a prediction.

What we think we know: the system's dominant behavior will be overshoot and oscillation rather than smooth adjustment, because the fast loops outpace the slow ones; the interest elasticity of AI capex and the timing-magnitude of AI productivity are the two parameters on which regime selection turns; and the 1970s analogy, however intuitive, mis-specifies the problem as supply-driven when the present pressure is demand-driven — a distinction with direct policy consequences.

What we do not know: the value of either governing parameter at any useful horizon; whether the competition loop's escalation is a phase or a settling point of the international system; whether the build-capacity loop's institutional throughput can be compressed enough to close the timescale gap; and whether the system's political coalitions — in every democracy hosting the build-out — will tolerate the distributional ledger the loops write.

What to watch next: the ten indicators above, with hyperscaler capex's rate-sensitivity and grid-equipment lead times the two that will move first in any regime change.

Final verdict on signal versus noise: this is the strongest structural signal in the current landscape, the capstone finding of this series. The system is not a coincidence of four news beats; it is a single machine whose wires — capex to demand, demand to energy, energy to inflation, inflation to rates, rates to capex, and competition around every physical joint — now all carry load. Machines with this topology do not behave like their parts. They overshoot, they oscillate, and they select regimes on the values of one or two parameters. The two that govern this one — how rate-sensitive the race is, and when the productivity arrives — are exactly the quantities that no participant knows and every participant is betting on. That is the definition of a system at its inflection, and it is why the analysts of 2036, looking back, will read the 2026 energy pages and the 2026 AI pages as one story, whichever way it ended.