Decent Newsroom
Sign In

Six Years to a Turbine

By Roger November 10, 2026
Cheaper tokens, a six-year queue for gas turbines, and a missile full of smuggled chips: the AI bill has arrived in physical form, and miners are paying part of it with coins.
Six Years to a Turbine

Six Years to a Turbine

At 09:31 this morning the Dow

Gas turbines on skids beside containerised data-centre modules at dusk was up 1,104 points, the Nikkei up 1,684 and the Nasdaq up 674. Gold traded at $4,125.70 an ounce. Bitcoin, in the same reading, was down $719 at $74,535.08. The snapshot is my own measurement, taken at 09:31 CET this morning.

A tape like that normally drags bitcoin along with it. This time the highest-beta asset in the market went the other way while almost everything else ripped. The explanation does not live inside bitcoin’s own chart. It lives in the order book for gas turbines, in the price of a single model call, and in the customs record of an Nvidia module pulled out of a cruise missile.

Three prices tell one story. The bill for the artificial intelligence build-out has arrived in physical form, and a portion of it is being settled in coins.

The seller who should not be selling

CoinDesk reported this morning that MARA, one of the largest listed bitcoin miners, sold 13,597 BTC for $1.1 billion across six weeks. The company repaid every loan secured by its coins and says it will sell more, with the proceeds going into AI datacenters (CoinDesk, 10 November 2026).

Divide the two disclosed numbers and the average realized price lands near $80,900 per coin, roughly 8.5% above this morning’s $74,535. That is my arithmetic on MARA’s own figures, not a company statement, and it says something the headline does not. The coins went into better prices than the ones we see now. This was not forced liquidation at the bottom. It was a decision taken while the window was still open.

Two readings fit the same facts and they point in opposite directions. Read one: a miner converting a volatile treasury into a contracted, cash-generating business is the most rational thing a miner has done in three years. Read two: the marginal holder of bitcoin just became a seller, because something else promised a better return per unit of risk.

The second reading has support outside MARA. BitGo reported third-quarter revenue down 51% while net income rose on treasury gains (The Block, 10 November 2026). Trump Media swung to a $409.7 million quarterly profit from a $54.8 million loss a year earlier, driven by crypto gains (The Block, 9 November 2026). In both cases the operating business shrank and the balance sheet carried the earnings. An income statement that has turned into a price chart is a warning, not a business model.

That is the quiet part of this cycle. When holding appreciates faster than doing, the reported profit stops describing the enterprise. It describes the position. Three companies in one morning showing the same pattern is not three coincidences; it is one regime.

On the other side of the trade sits Amundi, which raised its Strategy stake 44% to 1,903,725 shares, worth $291.4 million at 30 September prices (Bitcoin Magazine, 9 November 2026). Those shares closed at $153.09 that day and trade near $126 now, an 18% markdown the filing cannot show.

So the picture is not “institutions leaving”. It is a rotation inside the same conviction: out of the operating miner, into the treasury company; out of coins held as inventory, into megawatts held as contract.

The chart underneath

The technical picture matches the flow. James Check notes that bitcoin is holding $75k by its fingernails, and that just below sits an air pocket with very little traded volume and almost no on-chain cost basis. He puts the zone between $75k and $72k as the deepest this correction can go while remaining constructive (Checkonchain, 10 November 2026).

My own reading of the candles agrees on the shape. The all-time intraday high was $126,199.63 in October 2025; this morning’s price is 40.9% below it. Over the past ninety days bitcoin printed a high of $87,395.67 and a low of $62,535.24, and the last thirty days cost 10.9%; those figures come from my own analysis of daily and monthly candles this morning.

Line chart: ninety days of daily bitcoin closes with the 72k to 75k air pocket shaded

An air pocket is not a mystery. It is a place where nobody established a cost basis on the way up, because the move was too fast to trade. When price re-enters such a zone, the only bids are people who never owned it before, and they have no reason to defend a level they never paid. That is why Check calls the low-volume gap a place that tends to get back-filled rather than held.

Nothing in that description is bitcoin-specific. It is what a thin order book does when the marginal seller has a use for the money.

What a miner actually owns

Strip the branding off a miner and what remains is a shed full of machines attached to an electrical interconnection that somebody queued years for. The interconnection is the scarce asset, and it did not become scarce because of bitcoin. It became scarce because everything now competes for the same electrons.

The evidence is in the turbine queue. New gas turbine delivery in the United States has slipped into the 2030–2031 window, according to Wood Mackenzie senior analyst Aurora Tenorio, with global orders above 100 GW against annual manufacturing capacity of 60 to 70 GW (Reuters, 9 November 2026, via TFTC). Six years from signature to spinning metal.

Run the arithmetic on that gap. At 100 GW of annual orders against a 65 GW midpoint of capacity, the queue grows by roughly 35 GW every year the imbalance persists. Carried five years to 2031, that is on the order of 175 GW of accumulated deficit. The projection is mine; the two inputs are Tenorio’s.

Bar chart: annual gas turbine orders of 100 GW against 60 to 70 GW of capacity, accumulating a deficit to 2031

Markets do not wait for the queue to clear. Crusoe signed with ProEnergy on 16 April 2026 for thirteen aeroderivative turbine generator sets totalling 650 MW, delivered by summer 2027, and named speed to power as the deciding factor (TFTC, 9 November 2026). Aeroderivative means a jet engine on a skid; 50 MW per unit when you divide it out. Refurbished aircraft engines are now the fast lane to a datacenter.

Here is the part that matters for anyone holding bitcoin. What a miner sells is not the machine. It is the position in the queue, the land and the substation — the three things an AI developer cannot manufacture with capital alone. MARA is not abandoning bitcoin. It is monetizing the one input that became scarce in a world where the scarce input changed.

There is a symmetry worth naming. Bitcoin mining was always an arbitrage on wasted energy: find the stranded kilowatt, pay almost nothing for it, convert it into a monetary asset. Datacenters are the same trade with a customer attached. The miner already knew how to buy power nobody else wanted, and that knowledge is now worth more rented out than applied to hashing.

The second price: what a task costs

The same logic shows up in the model business, in a place most coverage misses. Anthropic cut the price of Claude Fable 5.2 by 20% per token, from $10 and $50 per million to $8 and $40. Cost per task on GDPx, a benchmark of professional work, rose from $67.16 to $88.15 (Latent Space AINews, 10 November 2026, citing SemiAnalysis).

Cheaper tokens, more expensive work. The run logs explain it. Where the previous generation tried about twenty test cases, the new one sometimes writes hundreds or thousands and spawns subagents to check its own output. It took first place at 16.4%. You pay for the checking either way.

Bar chart: token price fell 20 percent while cost per professional task rose 31 percent

That is not a pricing anomaly. It is the division of labour arriving inside a single product. The part of the work that got cheap is generation. The part that got expensive is verification, because verification is the only part whose cost scales with the consequence of being wrong.

The market is already splitting along that seam. Meta’s Muse Ember, released today, costs $0.10 per million input tokens and $0.50 per million output, one-eightieth of Fable’s $8 and $40 (Simon Willison, 10 November 2026). Ember is described as built for the work agents repeat all day: tool calls, routing, extraction, long document parsing. Not for the thinking that has to survive an audit.

Anyone who has run an agent for a week recognises the shape. Most calls are clerical, and a small remainder decides whether a number enters a database or a patient record. Paying eighty times the price for the clerical calls is how budgets die. Paying the cheap rate for the deciding call is how audits die.

Notice what this does to the cost structure of an economy that automates. The expensive step is no longer the answer. It is the check on the answer, and checks cannot be batched away, because each one has to be tied to a specific claim about a specific thing. Software made copying free. Verification stayed expensive, and it now sets the price of everything built on top of it.

The third price: geography under control

The third bottleneck is political, and it is the one that reveals the most. Ukraine’s military intelligence has catalogued 6,284 foreign-made electronic components inside 213 Russian weapon systems, including an Nvidia Jetson Orin recovered from an S-71 Monochrome air-launched cruise missile with autonomous optical targeting (TFTC, 9 November 2026, citing HUR’s War&Sanctions database). The routing runs through China, Kazakhstan, Kyrgyzstan and the UAE, through shell companies and consumer-grade re-export.

Set that against the American side of the same regime. Commerce approved Chinese chip buyers whose listed addresses reporters found to be empty offices, and arranged for the United States to collect a 25% tariff on the Nvidia chips headed there (Cory Klippsten, 9 November 2026). A control regime that takes a cut of the permitted flow is not a control regime. It is a toll booth with a security review attached.

The same reflex now reaches a market that did not exist five years ago. Senator Elizabeth Warren, about to chair Senate Banking, plans oversight of prediction markets on insider trading and platform manipulation (Bitcoin Magazine, 9 November 2026). New venue, familiar instinct: whatever trades, supervise; whatever supervises, tax.

Then enforcement escalates down the stack. On 6 November OFAC designated eight Canadian over-the-counter crypto businesses, the third wave of Operation Economic Outcast, alleging they converted Iranian rials into digital assets; Treasury Secretary Scott Bessent said Canada “cannot be a safe haven” (TFTC, 10 November 2026). The reach is extraterritorial, and it now lands on peer-to-peer intermediaries rather than on banks with compliance departments.

The mechanism is old and not complicated. A prohibition is a price ceiling set at zero on a lawful route. The rent it creates pays for the smuggling infrastructure. Once that infrastructure exists, the state must police the routes, which creates new rents, which requires new agencies. Nothing in the loop ever reduces the flow of chips into a missile. It raises the price of the components and moves the paperwork to another jurisdiction.

The case against all of this

The strongest objection deserves the space. Miners are not visionaries; they are power tenants with short, interruptible contracts, and a datacenter tenant needs fifteen years of firm capacity. The AI pivot chases the top of a capex cycle with borrowed conviction, at the moment the cycle turns. The turbine shortage will ease faster than a five-year queue suggests, because solar, batteries and behind-the-meter gas do not need a 200-tonne Frame unit and a permitting hearing. And if MARA is selling, that is deleveraging, which is the opposite of distress.

That is a serious case and parts of it are probably right. MARA clearing every secured loan before a drawdown is conservative, not panicked; a miner without collateral obligations is the one that survives the flush. The interruptible-contract point also cuts against my own argument: a power position is worth more as an option than as a fifteen-year lease, and options get repriced rather than rented out.

But the objection does not touch the price signal. Whether the build-out is 1995 or 1999, the turbine queue is real today, and capital is moving toward it from wherever it can be found. Miners are simply where it was findable first, because they held the scarce input and a liquid asset to sell. Whether the capex is well spent is a separate question from where the money comes from. Both can be true at once: a bubble being financed, and a real constraint being discovered.

What the three prices say together

Put them on one page. Bitcoin falls on a risk-on day because a holder with a better use for the money is selling. A model gets 20% cheaper per token and 31% more expensive per job because verification is where the labour went. A chip that cannot legally be sold arrives in a missile because the ban pays for the route that defeats it.

Underneath all three sits one price that is missing: the price of time. When the interest rate is administered rather than discovered, the structure of production lengthens according to the administrator’s preference, not the saver’s. The digital layer gets overbuilt because its projects discount furthest into the future. The physical layer starves because its returns arrive too slowly for a balance sheet measured quarterly. The correction then does not present itself as a crash. It presents itself as six years of lead time on the machine that makes electricity.

That is where the honest limit of this piece sits. I cannot tell you whether the datacenter build-out is 1995 or 1999, and neither can anyone selling you a model of it. What the three prices do tell you is the currency of the settlement. Not dollars. Watts, verified outputs and jurisdiction.

The consequence

A monetary system that cannot price time will keep mispricing energy, and energy is now the largest line item in the most capital-hungry industry on earth. Bitcoin does not build a turbine. It does not shorten the queue, and it will not make a Jetson harder to smuggle.

What it does is refuse to add a fourth administered price to the three the world is already paying. Every unit that can be diluted lengthens the queue for real things, because it shifts the return from building to borrowing. The miners selling coins to buy megawatts are not walking away from the monetary argument. They are acting on it: they hold the one asset whose supply cannot be expanded to finance the expansion of everything else.

Sources

Write a comment