Three Borrowed Pillars

In August the Chinese Communist Party's main theoretical journal published a post-mortem on the German industrial model. It found that the most admired manufacturing economy in Europe had rested on three pillars, that all three were rented from somebody else, and that all three were repossessed inside four years. Whatever you think of the source, the framework is the sharpest audit tool available to anyone allocating capital right now. Because the order being built on artificial intelligence is renting at least one of the same three, and in most cases it is renting two.

Architecting Alpha, Part IX

In The Next Petrodollar I argued that delivered electricity, not chips and not crude, is the binding constraint on this cycle, and that because a kilowatt-hour cannot be shipped, capital now moves to the energy rather than the energy moving to the capital. This essay is the geopolitical half of that argument. It is about what happens to an industrial position when the inputs it does not own get repriced by someone who does, and about the one layer underneath electricity that I promised to write about in April and never did.

The prompt came from an unusual place. On 16 August, 《求是》 (Qiushi), the theoretical journal of the CCP Central Committee, ran an international-affairs piece by Zheng Chunrong, who directs the Centre for German Studies at Tongji University. The title translates as Observations on Germany's Development Predicament and Political Deadlock. There is no English edition of it. Most of the world's commentary on China reads the Xi article at the front of each issue and stops; the analytical work sits further back. And the sentence that matters in this one is a balance sheet:

依托美国的安全保障，得以压缩国防开支，将更多财政资源投向本土经济建设；依托俄罗斯稳定廉价的能源供给，得以有效控制工业成本，巩固出口制造业的比较优势；依托中国庞大的市场，得以持续拓展外贸渠道，释放高端制造产能。

"Relying on American security guarantees, it could compress defence spending and channel more fiscal resources into domestic economic construction; relying on stable, cheap Russian energy, it could effectively control industrial costs and consolidate the comparative advantage of its export manufacturing; relying on China's vast market, it could continuously expand foreign trade channels and release high-end manufacturing capacity."

Security, energy, market. Three pillars, none of them owned. For thirty years that looked like comparative advantage and got taught in business schools as the Mittelstand model. Then the security guarantee became conditional, the cheap energy was sanctioned away across twenty-one rounds, and the market access was narrowed by Germany's own de-risking policy. The model did not fail because Germany forgot how to build things. It failed because three inputs it had never owned were repriced by their owners inside the same presidential term.

The claim, and how it fails

An industrial position is only as durable as its least-owned pillar. Three pillars decide it: security, meaning who guarantees the perimeter and controls the chokepoints your inputs travel through; energy, meaning who supplies the joules and sets the escalation clause; and market, meaning who is permitted to buy from you and sell to you. In the AI order all three pillars have changed substance. Security has migrated from alliance guarantees to control of the mineral midstream. Energy has migrated from cheap molecules to delivered electrons. Market has migrated from export destinations to export licences on compute. The substance changed; the audit did not. And almost every participant in this buildout, the United States included, is renting at least one pillar while calling it efficiency.

Four ways this is wrong. If mutual interdependence is genuinely stabilising, then renting is a permanent equilibrium rather than a deferred bill, and the German arrangement lasting thirty years is evidence for that rather than against it. If substitution arrives quickly, rare-earth-free motors, sodium chemistries, aluminium for copper, then the midstream chokepoint dissolves and the states now building floors under mineral prices will look like they subsidised a problem that solved itself. If the energy pillar unbinds, through efficiency, or because memory rather than megawatts turns out to be the constraint, then the second pillar is a 2026 story and not a decade. And if Germany died of politics rather than structure, which is a serious argument, then the pillars are a narrative device and governance is the only real variable. I take all four at the end.

Part I. The autopsy, and its arithmetic

Zheng's framing device is neat. The Economist called Germany "the sick man of Europe" in 1999; Schröder's Agenda 2010 labour reforms cured it; the same magazine applied the same label again in 2023. His argument is that the two diagnoses describe entirely different diseases. The first was a labour-market problem a government could legislate its way out of. The second is three missing pillars plus four internal ailments, and no German government currently has the votes to address any of them.

The internal ailments are worth listing because they are not unique to Germany: path dependence, where an industrial chain, technical system and business model were all bound to the internal combustion engine so tightly that the electric transition left the country's flagship sector behind; infrastructure under-investment, named explicitly as the power grid, gas and hydrogen networks and digital backbone, "unable to match the needs of the energy transition and digitalisation"; regulatory load, where German firms reportedly added 325,000 jobs in three years purely to process compliance paperwork; and a skills gap.

Then the discipline, because this is a state journal and I checked every number I intend to borrow.

The political claims hold up exactly. The AfD did take 20.8 percent and 152 seats in the 2025 Bundestag election, its best ever result, and it now leads national polling at roughly 28 percent against the CDU/CSU's 21. Government satisfaction really did collapse to the lowest level ever recorded in the ARD series, with 84 percent dissatisfied. And on 3 June 2026 Germany failed to win a UN Security Council seat for the first time, taking 104 votes against Portugal's 134 and Austria's 131, with the Chancellor conceding "we applied with conviction, we did not achieve our goal."

The economic claims are dressed. The article gives German GDP as minus 0.9 percent in 2023 and minus 0.5 percent in 2024. The federal statistical office reports minus 0.3 and minus 0.2. The contraction is real and the exaggeration is roughly threefold. The 530,000 unfilled skilled positions is the 2023 figure; the 2025 number is about 370,000, and it fell mostly because the economy weakened rather than because the shortage was solved. The claim that 2026 will be stagnation sits against a European Commission forecast of plus 0.6 percent.

That asymmetry is the most useful thing in the document. A state's published analysis of a rival model had a sharp diagnostic frame and inflated supporting evidence, and the inflation ran in exactly the direction that made the political conclusion land harder. Which is the correct way to use it: take the framework, audit the arithmetic, discard the conclusion. The conclusion Zheng reaches is that partisan fragmentation makes structural reform impossible and that external competition must be used to force domestic reform. That is an argument for unified party leadership addressed to cadres. The three-pillar frame underneath it is simply good analysis, and it does not care who wrote it.

Part II. The same issue, read as a balance sheet

Qiushi issue 16 is not a Germany issue. Read as a whole it is a state doing its own audit in public, and the arrangement of articles is the argument.

Three items back from the Germany piece sits a joint survey by the journal's economics desk and the China Iron and Steel Association, asking how traditional industries should transform. Its central line is a direct answer to German path dependence: new quality productive forces, it says, are "not about demolishing the old blast furnace and starting a new stove, but a renewal revolution on the vast existing stock." It reports more than 900 million tonnes of steel capacity through ultra-low-emissions retrofit, and describes AI process controllers running above 90 percent prediction accuracy, moving the industry from "watching the flame to adjust the furnace" toward "one-click steelmaking." Its cautionary example is not Germany. It is the American Rust Belt.

Two items further on, a Tsinghua computer scientist argues that the token should be treated as a factor of production, simultaneously the metering unit of intelligent services and the micro-level carrier of value creation, with output modelled as model intelligence multiplied by call volume multiplied by conversion efficiency. Set aside whether the economics is right. A state theoretical journal admitting a new factor of production into the canon alongside land, labour and capital is a sovereign act of accounting, and it is the same act this series has been tracking since Regime Change: something unrecognised gets moved onto the books, and whoever prepositioned before the recognition gets paid.

And in the following issue, two analysts at the China Institutes of Contemporary International Relations read the year's oil war and conclude that energy questions are increasingly hijacked by geopolitical objectives, and the globalisation of energy supply chains has been set back.

That is the hinge. Efficiency was the organising principle of the order that is ending: buy the input from whoever makes it cheapest, and treat the resulting dependency as mutual insurance. Security is the organising principle of the order beginning: buy the input from whoever cannot cut you off, and pay the premium. Germany optimised perfectly for the first principle and was invoiced under the second. Every pillar gets repriced when that switch flips, and the switch has flipped.

Part III. The energy pillar, tested in public

This year ran the experiment.

On 28 February, US and Israeli strikes hit Iran's nuclear and missile programmes, the Supreme Leader was killed, and Iran closed the Strait of Hormuz. It is still closed as I write, roughly six and a half months later. Pre-war the strait carried about 20 million barrels a day, close to a fifth of global oil supply. On 12 August, ten vessels crossed against roughly 130 daily transits before the conflict. Brent opened the first quarter at 61 dollars and closed it at 118, which the EIA notes was the largest quarterly increase on an inflation-adjusted basis in data going back to 1988. It touched 126 on 30 April. The IEA has cut its 2026 demand forecast by 940 thousand barrels a day, now expects world oil demand to decline by 2.5 million barrels a day this year, reports more than 10 million barrels a day of Gulf output shut in, and records US diesel above 200 dollars a barrel in early September, 94 percent above pre-war levels.

That is what it looks like when a rented energy pillar is repossessed at gunpoint. It is also, and this is the part worth sitting with, the old pillar. The molecule war is being fought over the input of the last order, and the United States is structurally better placed in it than at any point in fifty years, producing a record 13.8 million barrels a day.

The new pillar did not move the same way, because it cannot be seized. It has to be queued for, and I gave the queue arithmetic in Part V. What that essay did not say is how old the thing being queued for is. Seventy percent of American transmission lines are more than 25 years old, large power transformers average more than 40, and outages have more than doubled against the prior six-year period. Roughly one in three siting applications filed in the last five years was cancelled, about half were delayed six months or more, and delay costs developers around 200,000 dollars per megawatt. The venture fund MaC set out the underlying relationship most cleanly in its energy report this year, annotating the empty quadrant of a scatter plot of income against electricity consumption, across which the fit is 0.83, with a single line: high-income low-energy countries do not exist. No rich economy has ever skipped this pillar, and the one the United States built is now middle-aged.

And here the German lesson becomes uncomfortably transferable. Germany's energy pillar did not collapse into an industrial statistic. It collapsed into a political one: an inflation shock, then an energy-intensive sector shrinking, then a party at 28 percent. The American energy pillar is now being repriced into household bills during an election year. Residential rates rose nearly 5 percent across 2024 and 2025 and are forecast to rise about 4 percent again; utilities filed a record 31 billion dollars of rate increase requests; PJM capacity prices rose roughly tenfold on projected data centre growth. The president of the national association of state utility consumer advocates put it with some relish: "It's been a long time coming, but now we're the belle of the ball."

In fairness, the causal chain from data centres to any particular household bill is contested; gas prices, transmission investment and weather all contribute, and some studies find no universal residential effect. But the political economy does not wait for the econometrics. A pillar that shows up on a voter's monthly statement is a pillar that gets legislated.

Part IV. The pillar I owe you

In April I published a piece on my newsletter subtitled "the game of critical minerals in the New World, Part 1." I argued that copper is the purest read available on the physical economy because it has no narrative: no central bank holds it, no religion attaches to it, it gets used when people build things and not otherwise. Part 2 never appeared. This is it, and the reason it took five months is that I was looking at the wrong layer. The price is the symptom. The pillar is the processing.

First the symptom, updated, because I got the shape of it wrong. Copper did not peak in January. It set records in January, May, August, and again on 7 September at 14,533 dollars a tonne on the LME. Exchange stocks fell for 42 consecutive days, the longest run since 2014, to around 205,000 tonnes, with nearly half of the remainder already earmarked for withdrawal, and the cash to three-month spread hit a five-year high. The supply side explains most of it: Grasberg, roughly 3 percent of world supply, lost an estimated 270,000 tonnes this year and has had full restart pushed to early 2028; El Teniente is still recovering from last year's collapse; Cobre Panamá has processed stockpiles but is not reopened.

And then the detail that made me rewrite this section. The closure of Hormuz disrupted sulphuric acid shipments, and sulphuric acid is what you leach copper with. Regional prices ran to somewhere between 820 and 1,200 dollars a tonne. An oil chokepoint on the other side of the world became a copper story through a reagent nobody tracks. That is what coupled systems do, and it is the argument for reading pillars rather than markets.

Now the pillar itself. The scarce thing in minerals has never been the rock; it is the plant that turns rock into something a factory can use. On the IEA's numbers, China has accounted for over 90 percent of all growth in global copper smelting, taking its share of world capacity from around 15 percent to 50 percent by 2025. Across the rest of the list the concentration is higher still: roughly 99 percent of gallium refining, 96 percent of graphite, 91 percent of rare earths, 78 percent of cobalt, 70 percent of lithium. And the direction is the wrong one: the average share held by the top refining country rose from 70 percent in 2023 to 72 percent in 2025. Diversification is being announced faster than it is being built.

The control regime is now explicit. China's April 2025 licensing controls on seven medium and heavy rare earths remain in force, and the effect is measurable: yttrium exports to the United States fell from 333 tonnes before the restriction to 17 tonnes in the eight months after. The much broader October 2025 expansion, which covered processing equipment, battery and cathode materials, rare earth technologies, and an extraterritorial rule reaching any foreign product containing Chinese rare earths, was suspended for one year after the leaders met in South Korea. Those suspensions expire on 10 and 27 November 2026. Every capital allocation decision in the physical AI stack is currently being made inside an eight-week window on somebody else's option.

The Western response is the part that tells you the principle has changed. When the US Department of Defense took a stake in MP Materials in July 2025, the headline was the 400 million dollars and the roughly 15 percent ownership. The mechanism was the price floor of 110 dollars per kilogram on NdPr, committed for ten years, alongside a ten-year commitment to buy the entire output of a magnet plant that does not yet exist. Equity stakes in Lithium Americas and Trilogy followed. In January a Section 232 proclamation covering more than sixty processed critical minerals imposed no tariffs at all, and instead directed American negotiators to pursue agreements including price floors. At Evian in June the G7 set a target of cutting single-supplier dependency for rare earths and magnets to under 60 percent by 2030 and named price floors as a legitimate instrument, though in fairness it built no mechanism and committed no money, so treat the target as intent rather than infrastructure. One more correction while I am at it, because commentary gets this wrong constantly: refined copper still carries no Section 232 tariff. The 50 percent applies to semi-finished products.

A government setting a decade-long floor under a commodity price is not industrial policy in the ordinary sense. It is a state declaring that the market will not clear a strategic input at a price it can live with, and underwriting the difference. That is the efficiency principle being retired in a contract.

Part V. Run the audit on the position you hold

The reason to borrow a framework is to point it at yourself. So point it at the trade almost everyone reading this is long.

The security pillar of the AI buildout is a midstream it does not own. Every magnet in every motor, every transformer, and increasingly every tonne of smelted copper, sits behind a refining base concentrated between 70 and 99 percent in one jurisdiction, governed by a licensing regime whose current relaxation has an expiry date. The energy pillar is rented from a grid whose queue runs in years and whose hardware runs in decades, and the rent is now appearing on voters' bills. The market pillar is rented in both directions and least discussed of the three: who is licensed to buy your compute, and who is licensed to sell you theirs, is a policy variable now, not a commercial one.

Then run it on your own company, which is the version that pays. Who guarantees your perimeter: a single cloud, a single model provider, a single jurisdiction whose export rules you do not vote on. Who supplies your joules, on what contract, with what escalation clause, and what happens to your unit economics if that clause triggers. Who is permitted to buy from you, and does that set change if a trade posture changes. Germany's mistake was not renting a pillar. Everybody rents something; nobody is vertically integrated into physics. Germany's mistake was renting all three at once, from three different landlords, and booking the savings as comparative advantage for thirty years.

This is why the work my firms do sits deliberately across the intelligence, bits, atoms and electrons layers rather than inside any one of them. Not because anyone can own all four, which nobody can, but because the only way to know which pillar you are renting is to be standing at the seams where they join. LNCELOT's line has not changed: we do not build the infrastructure, we build the systems inside it. The rest of that stays off the page.

The three positions

Operator: own one pillar outright, and make it the one that would kill you. A cheap owned pillar beats three cheaper rented ones, because the rented ones reprice together and precisely when you can least afford it. For most companies in this cycle the answer is energy, which means contracted or owned generation on a horizon longer than your hardware depreciation. For some it is the mineral or component input, which means multi-year offtake rather than spot purchasing. For a few it is market access, which means a second jurisdiction you can actually serve from. Pick the one whose loss ends the company, and stop renting it this year.

Investor: underwrite the midstream and the conversion step, not the deposit. The scarce asset in minerals is separation, refining, magnet-making and smelting capacity, not ore in the ground, and the same is true one layer up, where the scarce asset is interconnection and delivery rather than generation. What is new is that states are now de-risking exactly that layer with floors, offtakes and equity, which is the recognition mechanism from Regime Change arriving in minerals: an asset that could not previously be financed becomes financeable because a sovereign has agreed to stand under its price. Position before the recognition, not after it, and read the November expiry dates as the calendar they are.

State: decide whether your jurisdiction is a pillar or rents one. Security has displaced efficiency as the organising principle, which means the questions are no longer about cost of production but about who could interrupt you and how long the interruption would last. Every jurisdiction holding a refining step, a chokepoint, a grid interconnection or a permitting regime holds a pillar somebody else needs, and most of them are still pricing it as a utility rather than as a term of trade.

How this could be wrong

Substitution is the strongest counter. If rare-earth-free motors, sodium-ion chemistries and aluminium substitution scale on the timelines their backers claim, the midstream chokepoint erodes from the demand side, the price floors look like a subsidy for a problem that solved itself, and the security premium being paid today is destroyed value. Engineering around a chokepoint has beaten controlling one more than once.

Interdependence may simply hold. The German arrangement lasted three decades, and mutual dependence is genuinely stabilising right up until it is not, which makes the timing unforecastable rather than the direction wrong. A related and sharper version: Germany may have died of politics rather than structure. A country with 84 percent government dissatisfaction and a fragmented parliament cannot execute a hard reform under any pillar configuration, in which case the three pillars are a symptom and governance capacity is the only variable that matters.

The energy pillar may unbind, through efficiency gains outrunning demand, through memory rather than megawatts turning out to be the binding input, or through behind-the-meter generation letting the refinery bring its own boiler and never join the queue. Those were the failure modes I attached to The Next Petrodollar and they apply here unchanged.

And the honest one about method. I have used a Chinese state theoretical journal as an analytical frame. It is written for cadres, its purpose includes demonstrating that a fragmented democracy cannot reform itself, and I showed above that it inflated the German economic evidence roughly threefold in the direction that made its conclusion land. None of that makes the framework wrong, and none of it would make the framework right either. Use the structure, check every number, keep the conclusion at arm's length. If one of the four failure modes above breaks first, write to info@selfbuiltsystems.com and tell me which. Short and specific gets an answer.

Three pillars: who guarantees you, who powers you, who is allowed to trade with you. Germany rented all three, for thirty good years, and the invoice arrived in four. The useful question is not whether you are renting. It is which one, from whom, and what the notice period is.

Architecting Alpha is published in the spirit of bold conjecture and ruthless criticism. Every claim is linked to its source. Where a source is a state publication, its numbers are checked against the primary statistics and corrected in the text rather than repeated.