Hook
On October 5 — a Monday, the dateline carrying no year, which my audit instinct flags before anything else in the document — Morgan Stanley's equity strategy desk sent a note to institutional clients with a thesis that has hardened into reflex across the sell-side: the US equity pullback is a window, not a warning. Earnings are strong. Capital goods names lead the upward revisions. Heavy-asset cyclicals look attractive. Buy the dip.
I have read this genre for two decades, and it has a structural tell. It argues from the numerator — earnings, revenue, capital expenditure — while the market's actual clearing mechanism runs on the denominator: the discount rate. Morgan Stanley references the denominator exactly once, observes that rising bond yields have offset a strong earnings season, and then never returns to it. That single omission is the entire story. It matters more for crypto than for equities, because crypto holds the longest-duration assets that trade anywhere.
Context
The note's scaffolding is straightforward, and I want to lay it out precisely before I dismantle it. Its argument rests on four claims that do not sit as comfortably together as the desk implies: a valuation claim, an earnings claim, a driver claim, and an interest-rate claim.
The valuation claim is that equities have fallen materially since early summer, which the desk frames as a better entry environment. The earnings claim is that profits are strong and, more important, being revised upward, with capital goods at the front of the revision queue. The driver claim is that this strength comes from a concentrated set of engines: artificial intelligence demand, record capital expenditure from the cloud hyperscalers, and a "strong macroeconomic environment." The interest-rate claim is that rising bond yields have offset the earnings strength, slowing the market's momentum since mid-August.
The quantitative anchor is a Bloomberg Intelligence consensus figure: S&P 500 constituent earnings are expected to grow 25% year-over-year in Q3, following 34% in Q2. Read those two numbers in sequence. 34%, then 25%. The desk calls this "strong." It is strong in absolute terms. It is decelerating in trend terms. A growth rate that falls nine percentage points in a single quarter is a growth rate that has turned. Presenting a decelerating series as evidence of durability is the kind of framing choice that survives only until the next print.
Here is where the architecture becomes interesting. An equity price is a discounted stream of future cash flows. The numerator is the cash flow. The denominator is (1 + r)^t, where r is the discount rate and t is time. When the risk-free rate rises, r rises, the denominator grows, and the present value of every future cash flow shrinks. The longer the duration — the further out the cash flows sit — the more violent the compression. A capital goods company with near-term earnings and a short payback period suffers less than a software company whose value is back-loaded into a terminal assumption. This is why rising yields are never style-neutral. They are a mechanical rotation signal: out of long duration, into short duration.
Morgan Stanley recommends cyclical, heavy-asset names. That recommendation is coherent only if rising yields are growth-driven — the reflation case — and not tightening-driven or supply-driven. The note never states the assumption, never tests it, and never prices the alternative. That is a logic gap, not a rounding error. And the gap widens the moment you move from equities to crypto.
Core
Crypto is where the logic gap becomes a canyon. If rising discount rates mechanically punish long-duration assets, then the asset class with the longest duration in existence should be the most punished. A token that pays no dividend, holds no terminal cash flow, and derives its entire value from a discounted expectation of future network activity is, by construction, the purest long-duration instrument ever listed. The Morgan Stanley note never mentions this. It does not have to; crypto is not its mandate. But anyone holding crypto should perform the translation the desk skipped, because the same denominator that is compressing equity multiples is compressing token valuations, and it does so with more force.
Here is the part that most crypto participants get wrong. They read a macro note like Morgan Stanley's and conclude that "strong earnings" is bullish for risk assets, including crypto. That conclusion inverts the mechanism. Strong earnings are a numerator signal. Rising yields are a denominator signal. When the two move in opposite directions, the market prices the denominator first, because the discount rate touches every asset simultaneously, while earnings touch only the specific issuers that report them. The denominator is systemic; the numerator is idiosyncratic. Crypto is a systemic-duration asset. It inherits the denominator's verdict, not the numerator's.
I have watched this mechanism run in real time. During the 2022 unwind, I spent three months reconstructing Three Arrows Capital's isolated margin positions, tracing liquidation cascades through Anchor and Venus to establish where the insolvency actually originated. The consensus narrative blamed the protocols. The on-chain record said otherwise: the failure was internal leverage mismanagement, a duration mismatch between borrows and collateral that no amount of protocol redundancy could absorb. That is the discipline the Morgan Stanley note lacks. The ledger remembers what the interface forgets. A dashboard can display a healthy collateral ratio right up until the discount rate moves, at which point the ledger reveals that the ratio was always a function of a rate assumption nobody stress-tested.
That is the first place crypto diverges from the equity story. In equities, the discount rate is set by a central bank and observed in the Treasury market. In DeFi, the discount rate is a governance variable. This is not a metaphor. Look at how lending protocols price risk. On Aave and Compound, the interest rate curve — the borrow rate as a function of utilization — is not discovered by market clearing. It is chosen. Governance sets the base rate, the slope, the optimal utilization point, the reserve factor. The "market" rate is an administered rate wearing the costume of a market rate. I have argued for years that these models are arbitrary in exactly this sense: they respond to a governance vote, not to the balance of real supply and demand for credit. When the global risk-free rate rises, an on-chain borrow rate that was calibrated in a zero-rate regime does not automatically adjust. It stays where governance left it, and the gap between the on-chain rate and the true cost of capital becomes an unhedged exposure sitting on every levered position in the system.
This is the denominator problem in its crypto-native form. In TradFi, when the discount rate rises, every valuation model updates because the input is a live market price. In DeFi, the discount rate is a parameter, and parameters do not update themselves. The ledger remembers what the interface forgets — and what the interface forgets, in this case, is that the borrow rate on a governance dashboard is a policy choice, not a market signal. A protocol can display a stable 3% borrow rate for months while the risk-free rate triples underneath it, and the dashboard will show nothing unusual until the positions that depended on that 3% rate are liquidated.
The on-chain risk-free rate has a name, and it is the stablecoin yield. When the US Treasury bill rate rises, the cost of holding a zero-yield token rises with it, because the opportunity cost is now the T-bill rate available through tokenized treasuries and stablecoin lending markets. This is the cleanest transmission channel from the macro denominator into crypto, and it is observable in real time. The basis trade — long spot, short perpetual futures — compresses when funding rates fall, and funding rates fall when the risk-free rate makes the carry unattractive. Watch the funding rate on a major perpetual, and you are watching the crypto market's own discount rate in motion. When funding goes negative while spot holds, the market is telling you that the denominator has overtaken the numerator.
Consider what this means for the capital goods rotation on-chain. The tokens that survive a rising denominator regime are the ones whose cash flows are denominated in a unit that appreciates with the risk-free rate — stablecoins, tokenized treasuries, protocols earning real spread. The tokens that do not survive are the ones whose cash flows are denominated in a unit that depreciates as rates rise — governance tokens with no fee capture, points, airdrop futures. The rotation is not thematic. It is a flight to duration.
Now connect this to the capital goods rotation that Morgan Stanley is actually recommending. The desk's real signal — the one buried under the "buy the dip" framing — is a style rotation: out of light-asset growth, into heavy-asset cyclical. Capital goods lead the earnings revision queue because capital expenditure is rising, and capital expenditure is rising because the AI buildout requires physical infrastructure: servers, power, cooling, fabs, transmission. The market is repricing the physical substrate of computation.
Crypto has a direct analogue, and it is the most important structural trade in the asset class right now. The rotation on-chain runs from narrative assets to cash-flow infrastructure. On one side sit the tokens whose value is pure expectation — memecoins, unlaunched narratives, collections priced on floor dynamics rather than usage. On the other side sit the protocols whose value is a discounted stream of observable fees: layer-2 sequencers that collect real transaction revenue, DePIN networks that sell physical compute and bandwidth, restaking layers that charge for security, and RWA protocols that earn spread on tokenized collateral. The first group is the longest-duration asset in the market, because its "cash flows" are entirely terminal. The second group has actual near-term cash flows, which means it is shorter duration, which means it is less exposed to the rising discount rate. The rotation Morgan Stanley is describing in equities is happening in crypto for the same mechanical reason, and most participants are misreading it as a sentiment cycle.
The AI capex narrative sharpens this further, and it is where I have direct professional exposure. In 2026 I spent four months co-authoring the technical specification for a zero-knowledge payment channel designed for autonomous machine-to-machine commerce — AI agents transacting without human intermediation. The design constraint we imposed was deliberately conservative: no "AI-native" tokenomics, no novel incentive layers, only proven cryptographic primitives and backward-compatible settlement. The reason is precisely the denominator logic. An AI agent payment layer is infrastructure. Infrastructure is priced on throughput and reliability, not on narrative. If you build it with a cash-flow profile that depends on a token appreciating, you have built a long-duration asset and dressed it as a utility, and the discount rate will find it.
The cloud capex that Morgan Stanley cites as the earnings engine is the same capex that funds decentralized compute markets. When hyperscalers commit record capital expenditure to AI infrastructure, they validate the demand curve that DePIN compute networks are trying to serve at the margin. That is a genuine, if indirect, tailwind. But it is a numerator tailwind. It says nothing about the denominator. An AI compute token can have rising real demand and still fall in price if the risk-free rate rises faster than its fee growth. Demand and price are not the same variable, and the gap between them is exactly the discount rate.
This is where I want to be precise about what the Morgan Stanley note gets right, because a forensic reading requires granting the valid parts. The desk is correct that earnings revisions are concentrated in capital goods and that this concentration is a leading signal for a capex cycle. It is correct that valuations have compressed, which improves forward returns if earnings hold. It is correct that the driver is concentrated in AI and cloud spending, which is a real and measurable capital flow, not a sentiment. The error is not in the facts. The error is in the causal chain. The desk treats "strong earnings" as sufficient for "higher prices," when the correct statement is that strong earnings are necessary but not sufficient, and sufficiency depends entirely on the path of the denominator.
Crypto's version of this error is more expensive, because crypto leverage is more reflexive. In equities, a rising discount rate compresses multiples gradually, through the re-rating of forward estimates. In crypto, a rising discount rate hits positions that are marked to a governance-set borrow rate, and the adjustment is not gradual — it is a liquidation cascade. I traced exactly this dynamic in the Three Arrows reconstruction: the positions looked solvent on the interface because the interface was pricing collateral against an oracle and a borrow rate that had not yet repriced. The ledger remembers what the interface forgets, and what it remembered in 2022 was that a collateral ratio computed under a stale discount rate is not a collateral ratio at all. It is a latency artifact.
The MakerDAO analysis I did during the 2020 DeFi Summer is the mirror image of that lesson, and it is worth restating because it is the constructive case. When the ETH/USD oracle was manipulated and the DAI peg came under pressure, the mainstream panic narrative predicted systemic failure. I traced the liquidation threshold calculations directly in the Solidity and found the opposite: the protocol's conservative collateralization ratios and redundant oracle design absorbed the shock. The system held not because the discount rate was stable, but because the parameters had been set with enough margin to survive its movement. That is the design lesson the AI-capex crypto trade needs to internalize. The question is not whether demand for decentralized compute will grow. The question is whether the protocols capturing that demand have set their parameters — their borrow curves, their liquidation thresholds, their emission schedules — with enough margin to survive a denominator regime they did not calibrate for.
I have seen this parameter-lag failure mode before, in a different context. In early 2017 I spent six months auditing the early draft of Ethereum's Slasher protocol, before mainnet launch. The critical finding was a consensus divergence in the finalized proof-of-work state transition function — under high latency, two honest nodes could compute different finalized states and split the chain permanently. The mechanism was a parameter that assumed a latency bound the network could not guarantee. The discount rate problem in DeFi is the same class of bug. A borrow curve assumes a rate regime. When the regime moves and the parameter does not, the system does not gracefully adjust — it diverges, and the divergence settles as liquidations. The Slasher memo I submitted was rejected initially and validated later. Parameter assumptions always look safe until the environment violates them.
The emission schedule point deserves its own treatment, because it is crypto's hidden second denominator. A token's discount rate is not just the global risk-free rate. It is the risk-free rate plus a token-specific inflation term. When a protocol emits new tokens to incentivize usage, it is raising the discount rate applied to its own existing supply. This is the mechanism that makes so many "real yield" narratives fail. A protocol can report rising fee revenue — a rising numerator — while simultaneously emitting tokens at a rate that exceeds the fee growth, which raises the denominator faster than the numerator. The net present value of holding the token falls even as the protocol's usage metrics rise. The interface shows a growing protocol. The ledger shows net dilution. The ledger remembers what the interface forgets, and what it forgets here is that emissions are a discount rate, not a marketing budget.
The capital goods rotation, translated on-chain, is therefore not simply "buy infrastructure tokens." It is "buy the infrastructure tokens whose near-term cash flows exceed their emission rate." That is a much narrower set, and identifying it requires exactly the kind of code-level analysis the equity note never performs. You have to read the emission schedule in the contract, not the APR on the dashboard. You have to compute net issuance against fee capture, not gross revenue against total value locked. You have to check whether the borrow curve that prices the protocol's leverage was calibrated under a regime that no longer exists. This is audit work, not sentiment work.
There is a second, quieter way the denominator shows up on-chain, and it concerns the retail user directly. The DEX aggregator promises the "best route," and the interface displays a slippage figure that looks acceptable. What the interface does not display is the value extracted by MEV bots between the user's intent and the user's settlement. I have audited this layer, and the finding is consistent: for the average retail swap, the realized price is worse than the quoted price by a margin that exceeds the aggregator's advertised fee savings. The "best route" is a routing optimization on the visible surface. The invisible surface — the mempool, the ordering, the sandwich — extracts more than the routing saves. The ledger remembers what the interface forgets, and what the interface forgets is that the price you see is not the price you get. This is the same structural error as the equity note: the visible numerator, the quoted rate, is presented as the whole story, while the hidden denominator, the extracted value, determines the outcome.
The on-chain expression of the denominator regime is the tokenized treasury bid. When the risk-free rate is high enough, capital that would otherwise sit in volatile tokens migrates to tokenized T-bills, because the yield is real, dollar-denominated, and free of duration risk. This migration is measurable on-chain, and it is crypto's equivalent of the equity market's rotation out of growth. The protocols that benefit are the ones issuing the tokenized treasuries and the stablecoin rails that carry the flows. The protocols that suffer are the ones whose token was the destination for capital that now has a better, safer alternative. The denominator does not announce itself. It shows up as a slow reallocation of stablecoin supply, visible only in the ledger.
History is unambiguous about what happens to long-duration assets when the discount rate regime shifts. The 2021 to 2022 transition from a zero-rate regime to a tightening regime compressed crypto's market capitalization by more than two-thirds, and the compression was not evenly distributed. Tokens with near-term cash flows fell less than tokens whose value was entirely terminal. The market was not making a sentiment judgment. It was repricing duration. The same repricing is available to be read in the current regime, and the sell-side note that frames the current environment as a buying opportunity does not perform the duration decomposition that would tell you which assets are actually short-duration and which are merely pretending to be.
Contrarian
Here is the contrarian angle, and it cuts against both the Morgan Stanley thesis and the crypto consensus that has formed around it.
The consensus reading of a note like this is binary: either "risk-on, buy the dip" or "risk-off, the discount rate wins." Both readings share a hidden assumption — that the discount rate and the earnings cycle are independent variables that will resolve in one direction. They are not independent. The entire AI capex cycle that Morgan Stanley cites as the earnings engine is itself financed by the same credit conditions that set the discount rate. Record cloud capital expenditure is funded by corporate debt issuance and equity, both of which are sensitive to the risk-free rate. If the discount rate rises enough to compress equity valuations, it also raises the cost of the capital expenditure that generates the earnings. The numerator and the denominator are not racing on parallel tracks. The denominator is the fuel for the numerator. When the discount rate rises past a threshold, it does not just discount the AI capex earnings — it slows the AI capex itself.
This is the blind spot that neither the sell-side note nor the crypto consensus prices. The note assumes the AI capex cycle is exogenous to interest rates. It is not. And the crypto version of the same blind spot is more dangerous, because crypto's capex analogue — DePIN buildouts, L2 infrastructure investment, restaking security budgets — is financed by token issuance, which is itself a discount rate. A rising global risk-free rate compresses token prices, which raises the real cost of the token-financed buildout, which slows the buildout, which reduces the future fees that justified the token price. The reflexive loop runs in both directions. In a rising-rate regime, it runs down.
There is a second blind spot, and it is structural. The note's recommendation of heavy-asset cyclicals assumes that capital goods earnings are more durable than growth earnings because they are nearer-term. But near-term capital goods earnings are, by definition, a function of a capital expenditure cycle that has already been committed. The order book is real. The risk is not the current cycle; it is the next one. If the discount rate keeps rising, the capex commitments that produced this quarter's capital goods earnings become next cycle's stranded capacity. Near-term cash flows are only safe if the cycle they depend on does not turn. The note prices the cash flows and ignores the cycle. Crypto's infrastructure tokens have the same profile: their current fees reflect a buildout that has already been funded, and their duration advantage evaporates the moment the funding cycle turns.
There is a third blind spot, and it is the one that most crypto investors will not want to read. The flight to duration that I described — out of narrative assets, into cash-flow infrastructure — is itself a crowded trade, and crowded trades have a discount rate of their own. When everyone rotates into the same "safe" infrastructure tokens, the entry price rises, the expected return falls, and the token becomes long-duration again, because you are now paying a premium for cash flows that were cheap when nobody wanted them. The rotation is only profitable if you arrive before the crowd. Once the crowd arrives, the rotation has already repriced the denominator, and the "safe" assets carry the same duration risk as the ones they replaced. There is no duration-free asset in a rising-rate regime. There is only duration you have priced and duration you have not.
Takeaway
The forward-looking question is not whether Morgan Stanley is right about the pullback. It is whether the market has begun pricing the denominator's second-order effect — the one where rising rates slow the very capex cycle that generates the earnings. Watch the emission schedules and the borrow curves, not the price charts. Watch whether the AI compute tokens' net issuance falls below their fee capture, because that is the only condition under which a long-duration asset becomes short-duration. And watch the on-chain lending markets for the tell that the interface is hiding: a borrow rate that has not repriced while the world's risk-free rate has. When the ledger and the dashboard disagree, believe the ledger. It is the only one that settles.