The bond market was not the loudest thing in the room last week. The chip stocks were — and bitcoin fell with them, which is the detail nobody wanted to underline.
Arthur Hayes published a piece arguing that the coming wave of AI initial public offerings will function as a supply shock against the crypto bid: that the SpaceX and Anthropic listings will unlock shares, dilute early holders, crater the equity narrative, and push marginal risk capital toward a monetary asset that cannot be diluted. He titled it, with characteristic meme-craft, "D is for Dilution." The internet did what the internet does. What it did not do was check the arithmetic.
I want to take the thesis seriously — more seriously, in fact, than its author did — because buried inside a somewhat careless argument is a genuinely useful tool for positioning across the next cycle. And also a trap.
The context you need is a supply-side map, not a price chart. Over the past four years the AI complex has absorbed an enormous quantity of capital, and the precise shape of that absorption matters more than the headline number. Hayes puts AI-related debt issuance at roughly one and a half trillion dollars. I would flag that figure immediately: global hyperscaler AI capex in 2025 lands somewhere in the four-to-five hundred billion range, and if you layer in private credit, data-center REIT debt, and SPV financing structures, a trillion-plus cumulative by 2026 is directionally defensible. But it is an estimate rendered as a fact, and anyone building a position on it should treat it as a hypothesis, not a datum.
The more interesting number is Bitcoin's. After the April 2024 halving, block rewards sit at 3.125 BTC per roughly ten-minute block. Run the multiplication — 3.125 times six blocks an hour, twenty-four hours, 365 days — and you get about 164,000 new coins a year against a circulating supply near 19.8 million. That is an annualized issuance rate of roughly 0.83 percent, and it halves again on a schedule published in code sixteen years ago. This is not a forecast. It is a reading of a rule that no board, no underwriter, and no sovereign can amend unilaterally.
Set that against an equity register. The claim circulating is that SpaceX holds about 86 billion shares, of which some 8.6 billion could become tradable around June 2027. Note the internal contradiction, because it is instructive: 8.6 out of 86 is ten percent, not the "roughly half" that appears in some retellings. Either the total share count is wrong or the comparison was against free float, not total shares. The specific number is unusable; only the qualitative read survives — a lockup structure large enough to matter, arriving on a schedule the market will be able to see coming.
And that visibility is the whole point, and the part the bulls keep stepping over.
In my own work — I spent the better part of a year building liquidity heatmaps for a Bangkok-based desk, mapping stablecoin issuance against settlement volume — the lesson that kept repeating was that supply increases are almost never the proximate cause of a drawdown. They are the accelerant. What determines the price is whether demand can absorb the float. I have watched unlock schedules get published, priced, absorbed, and forgotten, with the token rallying through them because the demand side was expanding faster than the register. Supply is a fact. Price is a negotiation.
So the Zcash analogy, which Hayes leans on, is doing less work than advertised. ZEC launched with a credible capped supply and a slow-start emission, and buyers in 2016 still lost money for years. But the cause was not the emission curve. It was an initial valuation that priced in a future nobody had built, plus a use case that never arrived. Supply discipline prevents dilution; it does not manufacture demand. Transplanting that lesson onto AI equities requires the unproven premise that AI valuations are already over-extended — and that premise is asserted, not demonstrated.
Here is where I part company with the framing, and where I think the sharper read sits. The phrase "bitcoin has no insiders to unlock" is doing rhetorical work that the facts do not quite support. There is no vesting cliff, true. There is also a large population of dormant early addresses, a mining sector that sells structurally to fund operations, and custodial estates from failed exchanges that will eventually find their way to market on nobody's published timetable. The accurate formulation is narrower and stronger: bitcoin has no predictable, scheduled unlock — a distinction that matters enormously for anyone modeling forward supply.
Which brings me to the contradiction that should anchor every reader's skepticism. Maelstrom plans to stand aside until liquidity conditions improve. The author of the long-term bullish thesis is not long. That is not hypocrisy — it is information, and it is the most honest sentence in the entire piece. Where liquidity hides, narrative finds its voice; and here the narrative's own author is telling you where he thinks liquidity is not.
I have watched this pattern before. In 2022, in the wreckage of Terra, what mattered was not the algorithmic stablecoin's design flaws — those were visible to anyone who read the curve — but the balance-sheet overlap nobody was mapping between the lending platforms and the trading desks. Contagion moves through the plumbing, not the pitch deck. Today the plumbing question is simpler and uglier: if AI capex disappoints on ROI before any of these listings arrive, the dilution event never happens, and the liquidity release still does. Right conclusion, wrong road. That is the most dangerous kind of analysis to inherit, because it works until the day it doesn't, and you will never know which mechanism actually paid you.
And the decoupling thesis has a hole in it the size of this week's tape. Chip stocks led the market lower; bitcoin followed. Correlation approaching one in stress is not a hypothesis, it is a recurring observation — March 2020, 2022, and again in the risk-off tape we just printed. An asset that falls with the Nasdaq on bad AI news cannot simultaneously function as the hedge against an AI bubble. The ETF wrapper gave institutions the switching rail, which is real and structurally irreversible, but a rail is not a destination.
What I think is under-owned in this whole conversation is where the marginal dollar actually goes when AI equity loses its bid. Not to bitcoin first. To bills, to gold, to money-market funds — and on-chain, to the tokenized dollar, where the issuer collects the Treasury spread with mechanical certainty. The quiet beneficiary of an AI deflation scare is the stablecoin balance sheet, and it appears nowhere in the thesis.
So position accordingly. Treat the dilution framework as a coordinate system, not a signal. Watch three things: the actual listing structure (a direct listing means the supply arrives at once, not in 2027, and that changes everything), the absorption capacity of the bid as measured through ETF creations and stablecoin float, and whether AI capex guidance holds through the next two earnings cycles. Chasing ghosts in the algorithmic machine is a hobby. Reading the silence between the blockchain blocks — the absence of new stablecoin minting, the flat ETF flows, the miner wallets that stop moving — is the work.
The illusion of control in a fluid world is believing you can schedule a narrative. You cannot. You can only measure what is actually flowing, and wait for the moment when the crowd's story and the ledger's arithmetic finally agree.
