Investing in the Era of Intelligence

What changes about allocating capital when the thing being made abundant is cognition, and the only asset that still compounds is the one no strategic factor market can sell.

Future Proof Intelligence. Research. No. XI. MMXXVI

Abstract

The apparatus of capital allocation was built to underwrite a world in which intelligence was the binding constraint. Diligence reads what an asset can do. Valuation discounts the cash flows that doing produces. A moat is understood as a defensible position around a capability. Every part of this machinery assumes that the scarce, value bearing thing is the capacity to reason, analyse, build, and execute, and that the institutions which concentrate that capacity hold the advantage. That assumption is dissolving. The capacity to reason competently over a bounded problem and produce serviceable work from an instruction is becoming a metered utility, priced toward its marginal cost and delivered through a converging standard interface.

This paper argues that abundance does not lower the value of allocation. It moves the object of allocation. When the layer that does the thinking commoditises, durable advantage relocates one layer down, to the accumulated, path dependent continuity that the thinking is anchored to. The consequence for capital is precise and uncomfortable. The standard diligence apparatus is now pointed at exactly the layer the entire market is structurally paid to make free, and is constitutionally blind to the only layer that still appreciates, because that layer is causally ambiguous, tacit, and unbuyable by construction.

We ground this in real and settled bodies of work: the economics of complements and commoditisation, the asset stock and isolating mechanism literature on why some advantages cannot be acquired, the economics of an intangible heavy economy, and real options theory, which we show inverts for this one asset class. The conclusion is structural rather than promotional. The durable investment in the era of intelligence is a position in something whose only input is elapsed, disciplined time, the one input capital cannot purchase, held by those who began accumulating it before the standard that prices it had set.

1. The Repricing Nobody Booked

1.1 The constraint allocation was built around

Every system of capital allocation is a machine for relieving a constraint and capturing the rent that sits on it. For most of the history of the firm, the binding constraint was applied intelligence. Careful analysis was slow. Expertise was rare and did not scale. The work of reasoning over a question, modelling a decision, drafting the instrument, and coordinating the sequence that followed was expensive because it required scarce human cognition, and the institutions that could concentrate that cognition under one roof held an advantage that compounded. The entire toolkit of allocation was tuned to that fact. Diligence asks what an entity can do, because doing was the scarce thing. Valuation discounts the cash flows that the doing produces, because the doing was the engine. A moat is read as a defensible perimeter around a capability, because the capability was the prize.

This is worth stating plainly before it is disturbed, because the disturbance is easy to underrate. The machinery of capital is not neutral. It encodes an assumption about where value lives. It was selected, refined, and professionalised over a century against a world in which that assumption was true. The assumption is now becoming false, and the machinery has not noticed, because machinery does not notice. It continues to point where it was aimed.

1.2 What abundance actually does

The capacity to reason competently over a bounded problem, to produce a serviceable artefact from an instruction, and to chain a sequence of actions toward a goal is becoming a metered utility. It is not yet uniform and the frontier still moves, but the direction is not contested by serious observers. By 2026 the analytical consensus inside the market that allocates to this technology had converged on a single sentence, repeated by people with no shared agenda: the model is becoming a commodity. Foundation models were being described in mainstream investment commentary as strategic commodities, offered as utility services through the major cloud platforms, with the explicit conclusion drawn that differentiation on model performance alone is unlikely to last. This is not a forecast made by critics. It is the operating assumption of the people deploying capital into the technology, stated as a premise rather than argued as a thesis.

When a constraint dissolves, the value that sat on it does not vanish. It migrates. This is the oldest regularity in the economics of technology and it has never once failed to hold. When mechanical power was scarce, advantage sat with whoever owned the engine; once power became a metered utility delivered through a wire, advantage moved to whoever could organise production around abundant power and be relied upon to deliver against an order. When computation was scarce, advantage sat with whoever owned the machine; once computation became something rented by the second, advantage moved to whoever held the data, the trust, and the accountable relationship the cheap computation now merely served. Intelligence is the input undergoing this transition now. The question this paper exists to answer is not whether the migration happens. It is what the migration does to the act of allocating capital, given that the entire apparatus of allocation was built facing the layer that is leaving.

It is worth being exact about what is meant by abundance here, because the word is doing analytical work, not rhetorical work. Abundance does not mean that intelligence becomes free or that it becomes uniform in quality. The frontier will continue to move and the best systems will continue to be better than the median ones for as long as there is a frontier. Abundance means something narrower and more consequential for capital: that the marginal cost of competent cognition, of the kind an institution actually buys, falls toward the cost of the compute and energy that produce it, and that the price follows the marginal cost down because the market structure forces it to. An asset whose price tracks its falling marginal cost is, by the standard definition, a commodity, regardless of how sophisticated it is to produce. Sophistication is not the opposite of commoditisation. Salt is sophisticated to produce at scale and is the textbook commodity. The error the allocation apparatus is most prone to in this era is to mistake the genuine difficulty of producing frontier cognition for evidence that frontier cognition is a durable asset. Difficulty is not durability. The two are routinely confused because both feel like a moat from the inside, and the confusion is expensive precisely because it is comfortable.

1.3 The bias that is already in the instrument

There is a complication that makes the repricing sharper than a simple migration. The allocation apparatus does not value the far future accurately even on its own terms, and this has been measured by people with no stake in the answer. Andrew Haldane and Richard Davies, working at the Bank of England, examined how capital markets discount future cash flows across a sample of more than six hundred United Kingdom and United States firms over three decades. They found systematic excess discounting: cash flows five years out priced as though they arrived in eight or more, cash flows thirty years out scarcely valued at all. Their estimate of the excess discount was on the order of five to ten per cent per year, statistically and economically significant, and rising over time. Haldane described the result, in his own phrase, as quarterly capitalism.

The relevance here is specific and is not the relevance usually drawn. The usual reading is that markets are myopic and should be less so. The reading that matters for this paper is structural. The allocation apparatus already discounts the far cash flow excessively. The abundance inversion is now relocating durable value into an asset whose entire character is that it compounds slowly and pays off far out. The bias and the relocation are pointed in opposite directions. The instrument was already mistuned against the long horizon, and the long horizon is exactly where the only durable asset in the new economy now sits. The repricing the era requires is therefore not a marginal adjustment to discount rates. It is a correction made against a bias the instrument was already running, applied to an asset class the instrument was already built to underweight. Nobody has booked this repricing because the instrument that would book it is the instrument that cannot see it.

1.4 What this paper claims

We make one structural claim and defend it without leaning on prediction.

When intelligence is abundant, the object of capital allocation changes. Diligence on what an asset can do, valuation of the cash flows that doing produces, and the reading of a moat as a defensible position around a capability are all now aimed at the layer the entire market is structurally paid to commoditise. The only layer that still appreciates is the accumulated, path dependent, causally ambiguous continuity the thinking is anchored to, and that layer is, by construction, illegible to a diligence apparatus built to read legible capability. The durable investment in this era is therefore a position in something whose sole input is elapsed, disciplined time. This makes the act of allocation a different discipline than the one capital has practised, because the standard logic of timing, the option to wait, inverts for this asset class: the irreversible, value destroying decision is not the decision to commit, it is the decision not to begin accumulating, since the option to begin later does not exist at any price once a time compression diseconomy is in force.

The remainder of the paper establishes why the commoditising layer is uninvestable as a durable position (Section 2), what a moat actually is once that is true (Section 3), why standard diligence is blind to it (Section 4), why time horizon and the option to wait invert here (Section 5), what therefore stops compounding and what starts (Section 6), and what all of this requires of the people who must act on it (Section 7).

2. The Layer the Market Is Paid to Make Free

2.1 A definition before an argument

It is necessary to be precise about what is commoditising, because the imprecision is where bad allocation hides. We distinguish two layers. The execution layer is the set of mechanisms that produce competent work from an instruction: the model that reasons, the orchestration that routes a task and chains tools, the throughput that turns a prompt into an artefact. The continuity layer is what gives a body of intelligence a stable referent over time: who it is for, what it has committed to, what it has carried forward, and whether what it does still corresponds to what it was declared to be. The execution layer answers the question of how the work gets done. The continuity layer answers the question of whether the thing doing the work is still the thing that was trusted to.

The execution layer is real, it is hard engineering, and in 2026 it is where the overwhelming majority of visible capital and visible effort is concentrated. It is also, by its own structure, on a path to becoming a utility. The argument of this section is that this path is not an accident, a temporary phase, or a failure of the people building it. It is the equilibrium the market is structurally paying to reach, which makes a durable capital position in that layer a contradiction in terms.

2.2 Commoditising the complement

The mechanism is the oldest pattern in technology economics, and it is not folklore; it is settled strategic economics. The principle of commoditising the complement was stated in its common form by Joel Spolsky in 2002 and rests on the economics of complements developed by Carl Shapiro and Hal Varian in their work on the network economy. The principle is exact: a firm that holds a strong position at one layer of a stack has a direct strategic interest in driving the price of the adjacent, complementary layers toward zero, because cheap complements increase demand for the layer the firm controls and pull the surplus of the whole system toward it. No matter how valuable a layer is in isolation, it can be more profitable to make it free if doing so increases the profits of the layer one actually owns.

Apply this to the intelligence stack and the conclusion is not speculative. The providers of frontier models benefit if orchestration is cheap, because cheap orchestration drives consumption of models. The cloud platforms benefit if models are cheap and open, because they drive consumption of compute and distribution. The tooling vendors compete each other's margins toward zero by construction. There is no powerful actor in the market whose interest is served by the execution layer remaining scarce and expensive. When no powerful actor's interest is served by a layer staying scarce, that layer does not stay scarce. This is the same dynamic Clayton Christensen described as the conservation of attractive profits: when one stage of a system becomes good enough and modular, the attractive profit does not disappear, it migrates to the adjacent stage that is still scarce and still hard. The execution layer is the complement the entire market is being paid to commoditise. A durable competitive position there is a position the market is collectively, and rationally, working to dissolve.

2.3 The depreciation underneath the commoditisation

There is a harder fact beneath the commoditisation, and it sharpens the allocation point. The execution layer is not only being competed toward marginal cost. Its capital base is a depreciating asset with a contested useful life. Through 2025 the major cloud operators guided to combined capital expenditure on the order of several hundred billion dollars, a share of forecast revenue well above the historical norm for the sector. By late 2025 and into 2026 a substantive and public accounting dispute had formed over how fast the hardware underneath this spend actually depreciates: whether the accelerators at the centre of it have a useful life of five to six years, as the depreciation schedules assumed, or closer to two to three, as critics including prominent investors argued, with the gap implying a large overstatement of profit over the following years. We state this as a live, public debate and as an order of magnitude, not as adjudicated fact, because the precise number is contested and the structural point does not depend on it.

The structural point is this. The execution layer is the rare thing that is being commoditised by competition and depreciated by physics at the same time. It is an asset whose price is falling toward marginal cost and whose carrying value is being eroded by a hardware cadence the operators do not control. A position there is doubly exposed: to the competitive solvent that drives the price down and to the depreciation schedule that writes the asset off. This is the precise opposite of the profile a durable capital position requires, which is an asset whose value is not a function of the next hardware generation and not a function of how cheap the next entrant can make the same capability.

There is a subtler implication of the depreciation point that the accounting dispute itself obscures, and it is the one that matters most for allocation. The argument over whether the useful life of an accelerator is three years or six is conducted as though it were a question about the honesty of reported earnings, which it partly is. But for the allocator the depreciation schedule is also a statement about the half life of the advantage the asset confers. An asset whose physical life is short and whose performance is superseded on a cadence faster than the depreciation schedule is not merely a cost question. It is an asset class in which the advantage has to be entirely re purchased every cycle, at the speed the cycle moves, with no carry forward of the spend that produced the last cycle's lead. A position that has to be fully re bought every two to three years to stay level is the financial signature of a treadmill, not a moat. The treadmill can be extremely profitable while it is running and for whoever sells the equipment to run on it. It is simply not a place a durable capital position can be held, because the defining property of a durable position is that last period's investment still works this period, and on a treadmill it does not. The depreciation debate, read correctly, is not an accounting story. It is the clearest available statement that the most heavily capitalised layer in the economy is the layer where capital has the shortest memory.

2.4 The data moat is not the exception, and saying so is the price of being serious

The intelligent objection at this point is well known and must be answered at full strength, because the answer is what keeps this paper from collapsing into a familiar slogan. The objection is that there is an exception to the commoditisation: proprietary data. The consensus in capital markets through 2025 and 2026 was precisely this, expressed as the line that the one thing a competitor cannot buy is your data, with defensibility said to live in data scale and the learning flywheel.

This is half right and the half that is wrong is the half that matters. The serious literature on data network effects, including the well known and unsentimental analysis from inside the venture industry itself, establishes that data as scale is a far weaker and far more perishable moat than the slogan implies. The returns to additional data diminish, often sharply; model performance saturates; the asymptote is frequently low, with a modest quantity of data capturing most of the achievable product value and each further increment costing more to acquire and clean for less marginal effect. The personalisation flywheel that gets called a data network effect is, in most cases examined honestly, just standard data driven personalisation, not a true cross user network effect, and it is substitutable the moment a competitor reaches the same low asymptote by other means.

So data as a stock is not the exception to the commoditisation thesis. It is another instance of it. The thing that does not commoditise is not the data. It is what the data has been made part of: an accumulated, path dependent, causally ambiguous continuity that a quantity of data does not by itself constitute and cannot be reconstructed into by a competitor who acquires an equivalent quantity. The distinction is the entire content of the next section, and it is the distinction the data moat slogan elides. A paper that allowed the reader to leave thinking data is the answer would have failed, because the market already believes that and is already mispricing on the strength of it.

2.5 Figure 1. The two layers from the allocator's seat

Figure 1. The execution layer and the continuity layer, priced. Picture two horizontal bands. The upper band is the execution layer: models, orchestration, tools, throughput, the compute underneath. It is wide, crowded, loud with named systems, and it is being acted on by two forces drawn as arrows pointing into it from opposite sides. From the left, the competitive solvent, labelled commoditisation, pushing price toward marginal cost. From the right, the depreciation schedule, labelled obsolescence, writing the carrying value down. Value enters this band and does not settle; it is squeezed out by both arrows almost as fast as it arrives. The lower band is the continuity layer: the accumulated, path dependent, accountable relationship the execution is anchored to. It has no arrows pushing into it, because neither the competitive solvent nor the depreciation schedule reaches it. Value that reaches this band does not flow off it and is not written down. It compounds. The single most important feature of the figure is what is missing from the lower band: there is no force acting to remove value from it, which is the entire reason it is the only band a durable position can be taken in.

The rest of this paper is about the lower band, and specifically about why the apparatus of capital allocation is built to underwrite the upper one.

3. What a Moat Actually Is Now

3.1 The moat that cannot be bought

There is a precise body of work on why some competitive advantages persist and others do not, and it is older and harder than the popular vocabulary of moats. Ingemar Dierickx and Karel Cool, writing in Management Science in 1989, drew the distinction the whole of this section turns on. Some assets can be bought in what they called a strategic factor market: you identify the asset, you pay the market price, you own it. Other assets cannot be bought at all, because no market for them exists. They have to be accumulated internally over time through a sustained flow of investment, and the stock that results is a function of the specific path that produced it. The critical resources, on their account, are always the accumulated ones, because the bought ones, being available to anyone with the price, confer no durable advantage.

Their analysis named the mechanisms that make an accumulated stock impossible to imitate, and three of them are the spine of what a moat is in the era of intelligence. The first is the time compression diseconomy: you cannot accumulate a decade of a stock in a year by spending more. Doubling the annual investment does not halve the time, because the stock's value is partly a function of elapsed time itself, not only of cumulative spend. The second is asset mass efficiency: a large existing stock makes further accumulation easier and cheaper than it is for someone starting from nothing, so the gap between an early accumulator and a late one widens rather than narrows. The third is causal ambiguity: when the link between the stock and the advantage it produces is opaque even to the firm that holds it, a competitor cannot copy it because the competitor cannot determine what to copy. Lippman and Rumelt formalised this last mechanism as uncertain imitability, and the strategic literature since has been consistent that tacitness, complexity, and specificity are the properties that generate the ambiguity and therefore the durability.

Read these three together and the definition of a moat in the era of intelligence writes itself. A moat is no longer a defensible position around a capability, because the capability is the commoditising execution layer and a position there is dissolved by the market that is paid to dissolve it. A moat is an accumulated stock with a time compression diseconomy, asset mass efficiency, and causal ambiguity. Stated in the language of allocation: the only durable asset is the one that cannot be bought in any strategic factor market, because its sole input is elapsed, disciplined time, and time is the one input that does not have a market because it cannot be transferred from a seller to a buyer.

3.2 The intangible the financial system cannot collateralise

This connects to a second settled literature that the era of intelligence makes suddenly central. Jonathan Haskel and Stian Westlake, in their analysis of the rise of the intangible economy, identified four properties that distinguish intangible investment from tangible: it is scalable, because it is non rival and can serve many uses at once; it is sunk, because it cannot easily be redeployed or sold to another enterprise and therefore makes poor collateral; it spills over, because its benefits are hard to fully appropriate; and it compounds through synergy, because intangibles combined with other intangibles produce more than their sum.

The continuity layer is the limit case of an intangible asset, and reading it through these four properties is what makes the allocation problem exact. It is maximally sunk: it has, by construction, zero redeployability, because it is constituted by the specific path that produced it and is meaningless detached from that path, which means it is the worst possible collateral in a financial system that lends against assets it can repossess and resell. Its spillover problem is inverted: unlike a generic intangible whose benefits leak to competitors, the continuity layer does not spill, precisely because its causal ambiguity and tacitness mean a competitor cannot extract or reconstruct it even with full sight of it. And its synergy is total: every additional period of disciplined accumulation is not maintenance against decay but compounding on top of a stock that makes the next period's accumulation more coherent and harder to copy.

The consequence for capital is the heart of the matter. The financial system is built to value and lend against assets that are legible, redeployable, and collateralisable. The only durable asset in the era of intelligence is the one that is, by its nature, none of these things. It does not appear cleanly on a balance sheet, it cannot be pledged, and it cannot be sold separately from the institution that accumulated it, because it is not a separable thing. This is not a defect of the asset. It is the exact reason it is durable. An asset that could be cleanly valued, pledged, and sold would, by that very legibility, be one a competitor could acquire, and an asset a competitor can acquire is not a moat. The continuity layer is durable in direct proportion to how badly it fits the instruments the financial system uses to recognise value.

This produces a measurement problem that is worth stating in its own right, because it is the quiet reason so much capital flows to the wrong layer. Accounting recognises the execution layer almost perfectly. Compute is a capitalised asset with a depreciation schedule. Model development is an expense that can be tracked to the cost. Headcount, licences, infrastructure: all of it lands on a statement with a number attached. The continuity layer is recognised almost not at all. The years of disciplined accumulation that produce it pass through the accounts as ordinary operating cost, indistinguishable on any statement from the cost of doing nothing in particular, because the act of accumulating continuity looks, period by period, exactly like overhead. Two institutions can run identical income statements for a decade while one of them has been laying down the only durable asset in the economy and the other has been spending the same money on nothing that will outlast it, and no statement either of them produces will show the difference. The market is therefore not merely biased toward the legible layer. It is structurally unable to see the illegible one even when it is staring at the entity that holds it, because the instrument that would record it does not have a line for it. This is the same finding Haskel and Westlake reach about intangible investment generally, taken to its limit: the asset that matters most is the asset the measurement system was never built to record, and what is not recorded is not priced, and what is not priced is, to the apparatus, not there.

3.3 Why this is not the data argument again

It is worth being explicit, because the distinction is fine and load bearing. Data as a stock fails the test in Section 3.1: it has a low asymptote (weak time compression diseconomy and weak asset mass efficiency, because the marginal datum adds little once the asymptote is reached) and it is not causally ambiguous (a competitor knows exactly what data would replicate the advantage and can often acquire a substitute). The continuity layer passes the test on all three mechanisms: it has a strong time compression diseconomy (it cannot be accumulated faster by spending more), strong asset mass efficiency (a deep accumulated continuity makes further accumulation more coherent and harder to match), and high causal ambiguity (what produced the advantage is tacit and path dependent even to the holder). This is the precise reason a serious allocator must not treat data scale and continuity as the same asset. They behave oppositely under the only test that predicts durability. The market in 2026 was pricing the first as though it were the second. That is the specific mispricing this paper exists to name.

4. The Diligence Inversion

4.1 The apparatus is built to read the wrong thing

Diligence is a legibility instrument. It exists to make an asset readable: to convert what an entity is and does into evidence a committee can evaluate. It is extraordinarily good at this for legible assets. It can read a balance sheet, a customer cohort, a technology, a margin structure, a capability. Its entire competence is the conversion of an asset into something demonstrable.

This competence is now the problem. The properties that make the continuity layer durable are the same properties that make it illegible to the apparatus built to read durability. Causal ambiguity, the third isolating mechanism, is not a difficulty diligence can work harder to overcome. It is a structural property which states that what produces the advantage is opaque even to the holder. An instrument whose entire function is to make an asset demonstrable cannot demonstrate an asset whose durability is constituted by its resistance to demonstration. The harder the diligence apparatus works to make the continuity layer legible, the more it will tend to mistake it for absent, because absence and causal ambiguity present identically to an instrument that can only see what can be shown.

This produces the central practical paradox of the era for capital. The legible asset is the one diligence can underwrite, and it is the commoditising one. The durable asset is the one diligence cannot underwrite, and it is the only one that appreciates. An allocation process that funds what it can see will, with the regularity of a law, fund the layer the market is paid to make free and decline the layer that lasts, not because the people running it are unsophisticated but because the instrument is doing exactly what it was built to do. The instrument is not broken. It is precisely calibrated to the wrong object.

4.2 The new underwriting object

If the durable asset cannot be read by diligence on capability, then diligence in the era of intelligence has to underwrite a different object, and the object is available with unusual precision because the engineering and governance literature converging on agentic systems in 2026 had already named it. The most useful definition of identity to emerge from that literature defined it not as a property an entity has but as the continuous relationship between what an entity is declared to be and what it is observed to do, bounded by the confidence that the two correspond at any given moment.

This is, read as an allocator should read it, an underwriting test, and a far more durable one than capability. The question diligence should be asking is not what an entity can do today, because what it can do today is the commoditising layer and will be matched. The question is what it has carried forward, how deliberately, for how long, and whether what it does still corresponds to what it has declared itself to be, sustained across the discontinuities, the model changes, the personnel changes, the ownership changes, that would break a correspondence not held by something structural. Correspondence sustained over years is not a thing that can be bought, faked at the moment of diligence, or reconstructed by a better funded competitor, because the input that produces it is the same time compressed, path dependent, causally ambiguous accumulation that constitutes the moat. Capability is a snapshot and snapshots commoditise. Correspondence over time is a path, and the path is the asset.

4.3 Why the legible asset is the trap, not the safe choice

There is a temptation, when an asset is illegible, to treat the legible alternative as the conservative choice. In the era of intelligence this is precisely inverted, and the inversion is the most expensive error available to an allocator. The legible asset, the demonstrable capability, the readable execution layer, is not the safe position. It is the position the entire market is structurally and rationally working to make worthless, on a delay. Its legibility is not a sign of its solidity. It is a symptom of its commoditisation, because what can be cleanly demonstrated to a diligence committee can be cleanly understood, and therefore cleanly reproduced, by a competitor and by the market that is paid to reproduce it.

The conservative choice, properly understood, is the asset that resists demonstration, because resistance to demonstration is what resistance to imitation looks like from the outside. This does not make illegibility sufficient; many illegible things are merely absent. It makes illegibility a necessary property of the durable asset, which means the allocator's task is not to demand that the durable asset be made legible, which would destroy the property that makes it durable, but to develop the discernment to underwrite a path rather than a snapshot, an accumulated correspondence rather than a demonstrable capability. The allocator who insists that the durable asset prove itself in the terms the legible asset proves itself in has, by that insistence, guaranteed they will only ever fund the commoditising layer. That is not prudence. It is a structural commitment to the one position the market is paid to dissolve.

4.4 What underwriting a path actually involves

It would be evasive to say the durable asset is illegible and stop there, because that leaves the allocator with a true statement and no method. The method follows from the structure, and it is not the same activity as diligence on capability, though it can be done with rigour.

The first move is to stop scoring the snapshot and start scoring the derivative. A capability assessment asks how good the entity is now. A path assessment asks whether the correspondence between what the entity has declared itself to be and what it actually does has held, tightened, or drifted across the discontinuities the entity has already passed through. The signal is not in the level. It is in the behaviour at the transitions: what survived a change of model, a change of personnel, a change of ownership, a downturn, a moment when the cheap and the right diverged. An entity that has passed through several such transitions with its correspondence intact has demonstrated something a capability snapshot can never demonstrate, because the snapshot is taken between transitions where everything looks stable and the only honest information is at the joins.

The second move is to read the absence of demonstration correctly. Causal ambiguity means the entity that holds the durable asset frequently cannot fully articulate what produces its advantage, which means an allocator demanding a crisp causal account of the moat is selecting against precisely the entities whose moats are real, because a crisply articulable advantage is, by the isolating mechanism logic, an advantage low in causal ambiguity and therefore low in durability. The allocator has to learn to treat a certain kind of inarticulacy not as a red flag but as a fingerprint. The relevant discernment is the ability to tell the inarticulacy of a real tacit accumulation from the inarticulacy of having nothing, and that distinction is made not by asking the entity to explain the moat but by observing whether the moat keeps producing the same kind of result across conditions that should have broken anything shallower.

The third move is the hardest and is the one that separates a serious allocator in this era from a sophisticated one. It is to underwrite the discipline rather than the asset, because the asset is the residue of the discipline and only the discipline is observable in the present tense. The question is not whether a deep continuity exists now, which is partly unanswerable by inspection, but whether the entity is, period after period, actually doing the unglamorous work of deciding what is carried forward, what is refused, and what the system is held accountable to, with enough consistency that the accumulation is real rather than asserted. Discipline observed over even a short window is a far better predictor of a deep continuity ten years out than a claimed continuity is of itself, because discipline is the generator and the generator is what compounds. An allocator who learns to underwrite the generator is underwriting the one thing that is both observable now and predictive of the asset later. That is the closest the era of intelligence allows to a diligence method for the durable layer, and it is enough, because it is more than the apparatus that funds only the legible layer has.

5. Time Horizon, and Why the Option to Wait Inverts

5.1 The standard logic, stated fairly

The most rigorous account of how time enters an investment decision is real options theory, set out canonically by Avinash Dixit and Robert Pindyck in their work on investment under uncertainty. Their analysis turns on three conditions that hold for most real investments: the cost is at least partially irreversible, so committing sinks something that cannot be fully recovered; the future reward is uncertain; and there is leeway over timing, so the decision can be postponed to gather more information. When all three hold, an investment opportunity is structurally a call option. Committing exercises, and therefore kills, the option to wait. Because waiting has value under uncertainty, the option to wait is worth something, and the correct decision rule is not the simple positive net present value rule but a higher threshold that accounts for the option being given up. The honest summary of the standard logic is that under irreversibility and uncertainty, waiting is frequently the rational and value preserving choice, and the naive instinct to commit early is often wrong.

This is correct, it is well established, and for almost every asset class it is the right frame. The argument of this section is not that real options theory is wrong. It is that the continuity layer is the one asset class for which the three conditions, applied honestly, produce the opposite conclusion, and that getting this inversion wrong is the most consequential timing error an allocator can make in this era.

5.2 The inversion

Apply the three conditions to the continuity layer with discipline rather than by analogy.

Irreversibility holds, but it is on the other side of the decision. For an ordinary investment, committing is the irreversible act and waiting preserves reversibility. For the continuity layer the irreversible act is not committing, it is not beginning to accumulate, because the time compression diseconomy means the stock that was not built between now and a later date cannot be built afterward by spending more. The years that were available to accumulate it are not recoverable. The asset you did not start is not an option you preserved by waiting. It is value you destroyed by not starting, and the destruction is permanent in exactly the sense Dixit and Pindyck reserve for sunk irreversibility, because the missing input is elapsed time and elapsed time is the one input that cannot be supplied retroactively at any price.

Uncertainty holds, but it does not cut the way the standard logic assumes. In the ordinary case, uncertainty makes waiting valuable because new information can change the decision. Here, the dominant uncertainty is not about whether the continuity layer is the durable asset; the structural argument of Sections 2 and 3 settles that to the degree structural arguments settle anything. The uncertainty is about timing of the standard that will price it, and that uncertainty resolves in one direction only: every period that passes is a period in which someone who began earlier has accumulated more, and the asset mass efficiency means their lead widens rather than narrows while the waiter waits. Waiting does not buy information that improves the decision. It buys time during which the decision gets structurally worse, because the thing being waited on appreciates only for those already accumulating it and the gap to a late starter compounds.

Leeway over timing holds least of all, and this is the decisive point. The standard logic assumes the opportunity persists while you wait, that the option is still exercisable later on similar terms. For an accumulated stock with a strong time compression diseconomy, the opportunity does not persist on similar terms. It is not an option that sits and waits to be exercised. It is a position whose terms degrade every period it is not taken, and which, once a standard sets around those who began early, ceases to be available at all rather than merely being available at a worse price. The leeway the standard logic relies on is the precise thing the time compression diseconomy removes.

5.3 The negative carry of waiting

The clean way to state the inversion is in the language of carry. For an ordinary irreversible investment under uncertainty, the option to wait has positive value: holding it, rather than exercising it, preserves something. For the continuity layer, the option to wait has negative carry. Every period the option is held rather than exercised, the underlying asset that would have been accumulated is not accumulated, and that un accumulated stock cannot be recovered later, while a party who exercised early compounds a lead that asset mass efficiency makes structurally harder to close. The waiter is not preserving optionality. The waiter is paying, every period, a cost that does not appear on any statement, in the form of a durable asset that is permanently not being built, against a clock that does not reset.

This is why the timing question in the era of intelligence is not the timing question capital is used to. For the execution layer, the commoditising one, the standard real options logic applies cleanly and often counsels patience, because that asset will be cheaper and better later and committing early sinks cost into something the market is making free. For the continuity layer the logic inverts completely: it is the one asset where committing early is not the impatient error but the only way to hold the position at all, and where waiting is not prudence but the slow, unbooked destruction of the only thing that lasts. An allocator who applies the same timing instinct to both layers will be wrong about both, and wrong in the most expensive possible direction on the one that matters.

5.4 The asymmetry of the window

There is a final property of the timing that follows from the standard setting around early accumulators, and it makes the inversion sharper still. The cost of being inside the continuity layer before its standard sets is ordinary: it is the cost of years of unglamorous, undemonstrable accumulation that shows nothing legible until it is decisive. The cost of being outside it after the standard sets is not ordinary and is not symmetric with that. It is not a competitive disadvantage that can be closed by deploying more capital, because the missing input is the years that were available only before the window closed, and those years cannot be purchased afterward at any multiple. This is not a race in the ordinary sense, where arriving second is worse than arriving first by some margin that money can compress. It is closer to a phase change: before it, the position is open to anyone willing to do the slow work; after it, the position is structurally settled and capital no longer purchases entry, because entry was a function of having started, and starting is the one thing that cannot be done retroactively. The asymmetry of the window is the entire reason the timing of this one asset class is not a matter of judgement that can be deferred. Deferral is the decision, and it is the irreversible one.

5.5 Why the standard real options instinct is precisely the trap here

It is worth dwelling on why a sophisticated allocator is more exposed to this error than an unsophisticated one, because the inversion punishes exactly the instinct that training installs. The unsophisticated allocator commits early out of impatience and is right about the continuity layer by accident and wrong about the execution layer for the same reason. The sophisticated allocator has internalised the genuine and well earned lesson that under irreversibility and uncertainty, the disciplined move is to wait, preserve optionality, and demand a higher hurdle before sinking irrecoverable cost. That lesson is correct for almost everything, which is exactly why it is dangerous here. The continuity layer is the one asset class where the trained instinct, applied faithfully, produces the worst available outcome, because the irreversibility is on the side the instinct treats as safe. The allocator does not fail here through carelessness. The allocator fails through the competent application of a rule that is right everywhere except on the one asset that matters most, and competence applied to the wrong model is more destructive than incompetence, because it is defended with conviction and scaled with discipline. The defence against this is not more rigour inside the standard model. It is the recognition that the standard model has a domain, that the continuity layer is outside it, and that the boundary of a model is invisible from inside the model. That recognition is itself a form of the discernment Section 4 described, applied to the timing question rather than the diligence question, and it is the same scarce capacity in both places: the ability to tell which layer one is looking at before applying the instinct that the layer determines.

6. What Stops Compounding, and What Starts

6.1 The ledger

The argument to this point can be reduced to a single ledger, and the discipline of the era of intelligence is the discipline of keeping it. On one side are the advantages that decay. On the other are the advantages that appreciate. The error the standard apparatus makes is not that it cannot tell value from no value. It is that it cannot tell a decaying advantage from an appreciating one, because both produce returns while the right people are in the room and both look like a moat from the outside until the moment they do not.

The advantages that decay in the era of intelligence are precisely the ones the apparatus is built to underwrite. A model lead decays as the model commoditises, on a delay set by the market that is paid to commoditise it. An orchestration advantage decays as the interface converges on a shared standard, which it has. A compute position decays as the hardware depreciates on a cadence the holder does not control and the carrying value is contested. A data scale advantage decays as the returns to data diminish toward a low asymptote and substitutes are found. Lock in at any of these layers decays every time the standard underneath gets cleaner and a migration path is built precisely because the market wants the lock in gone. Every one of these is a real advantage and every one of them is a defended position rather than an accumulating one, which means holding it costs more over time, not less, and the cost is the energy spent resisting an entropy the whole market is funding.

The advantages that appreciate are the ones the apparatus cannot read. An accumulated, path dependent continuity does not decay with the model, because it is the part that persists across the model. It does not depreciate with the hardware, because no part of its value is a function of the hardware generation. It does not erode as a standard converges, because it is the thing a standard converges around rather than the thing a standard makes interchangeable. And its causal ambiguity means it does not leak to competitors even under full sight. Every additional period of disciplined accumulation is not maintenance against decay; it is compounding, and the compounding accelerates rather than slows because asset mass efficiency makes each later period more coherent and harder to match than the last. This is the only asset in an abundance economy where the passage of time, which dissolves every other advantage, is the thing building the advantage instead.

6.2 The mental model is a root system, not a fortification

It matters which metaphor an allocator carries, because the metaphor determines what the allocator looks for. A fortification is a fixed thing that is defended; it depreciates the moment a better siege technology exists, and the cost of holding it rises as the attackers improve. Almost every advantage the standard apparatus underwrites is a fortification, and the apparatus is good at valuing fortifications, which is part of why it keeps funding them.

The continuity layer is not a fortification. It is a root system. A root system goes deeper and holds harder the longer it is left to grow, it cannot be transplanted because the holding is in the ground it grew through rather than in the plant, and it cannot be valued by inspecting the part above the surface, because the part that matters is the part that is not visible. An allocator looking for fortifications will systematically miss root systems, will mistake the visible plant for the asset, and will consistently overpay for the depreciating advantage and decline the appreciating one. The repricing the era requires is, at bottom, a change of metaphor: from defending positions to underwriting roots, and from valuing what is above the surface to underwriting the depth and discipline of what is not.

6.3 The arbitrage hidden in the mispricing

There is a consequence of the ledger that an allocator is entitled to draw and that the paper would be incomplete without stating, because it converts a structural argument into a concrete edge. If the market systematically overprices the decaying advantages, because they are legible and demonstrable and the apparatus is built to read them, and systematically underprices or fails to price the appreciating one, because it is illegible and the apparatus has no line for it, then there is a durable mispricing between the two layers, and a durable mispricing is the definition of an arbitrage available to whoever can see it before the market learns to.

The arbitrage is not a trade in the ordinary sense and it cannot be executed by anyone who needs the position to be legible to their own committee, which is most of the market and is the reason the mispricing persists. It is available only to capital that can do three things the standard apparatus cannot. It can underwrite the generator rather than the asset, per Section 4.4, which means it can take a position in a deep continuity while it is still illegible and therefore still cheap. It can hold the position across the horizon over which the continuity capitalises, which is longer than the horizon the apparatus discounts to and therefore longer than most capital can hold without being forced to mark, explain, or exit. And it can tell the inarticulacy of a real tacit accumulation from the inarticulacy of having nothing, which is the single discernment the whole paper has been describing under different names. Capital that has all three is not taking more risk than the market to earn the spread. It is taking a different and smaller risk than the market, because it is underwriting the only asset that appreciates while the market crowds into the only assets that decay, and it is being paid the spread precisely because the apparatus that sets the price cannot see what it is buying. The mispricing is not a market failure that will be corrected by better information. It will persist exactly as long as the apparatus that sets prices is built to read the legible layer, which is to say until the standard sets, at which point the mispricing closes not because the market learned but because the asset became the assumed floor and was no longer for sale at any price. The arbitrage and the window are the same fact stated in two registers. The window closing is the arbitrage being arbitraged away, once, terminally, by the standard rather than by the market.

6.4 The roots are hardening, and that is the timing fact

There is a regularity in how infrastructure layers mature. For a period the layer is soft: undefined, contested, built differently by everyone. Then a standard sets, and once it sets it is extraordinarily durable, because everything built afterward assumes it. The decisive fact about a standard is not its content. It is that it tends to form around whatever practice is already operating credibly at that layer when the setting happens, because a standard that contradicts working practice fails to be adopted and a standard that ratifies working practice succeeds. Standards ratify the incumbents of the layer they standardise.

By 2026 the continuity layer was visibly leaving its soft period. The regulatory motion around autonomous and agentic systems was hardening from several directions at once: governance frameworks organised around accountability, oversight, and the correspondence between what a system is and what it does; standards body work on identity and authorisation for software and AI agents; and a parallel academic literature moving from describing agent memory to specifying agent identity precisely enough to be governed. In the same window the liability environment was relocating: through 2026 the binding obligations for high risk AI under European law were coming into force, and the early insurance market was both writing specialist cover for autonomous systems and, in parts, excluding generative AI from general policies, which together mean that the loss from an autonomous system was becoming something that had to be underwritten explicitly rather than absorbed implicitly. When loss must be underwritten, the underwriting question stops being what the system can do and becomes who carries the consequence when it acts, which is a continuity and accountability question, not a capability one. The insurance side and the diligence side are arriving at the same object from opposite directions. That convergence is the timing fact. The standard for the continuity layer will set, it will set within a small number of years, and it will set around the practice already operating credibly at that layer when it does. After it sets, the layer is the assumed floor of the intelligence economy, and the cost of not being inside it is not a disadvantage that capital can compete away. It is a structural exclusion, because the standard will have been written by and around those who were already there.

7. Implications

The argument is structural, so the implications are not advice. They are consequences. We separate them by reader, because the same structural fact lands differently depending on where one is standing, and we keep them in the register of the idea, never of a pitch.

7.1 For institutions

An institution that locates its durable advantage in its intelligence is locating it in the layer that is commoditising and depreciating at once. Models will be matched, orchestration will be matched, throughput will be matched, and matched soon, because the entire market is organised and funded to make those things cheap, and the hardware underneath them is being written down on a cadence the institution does not control. The only advantage that survives is the one held at the continuity layer: the institution's accumulated, path dependent correspondence between what it has declared itself to be and what it does, carried across every change of system beneath it.

The operative implication is that the deliberate construction and maintenance of that continuity should be treated as primary capital formation, not as documentation or governance overhead. The institution that treats its accumulated continuity as the asset, and the execution layer as the commoditising input it merely consumes, is allocating its own internal capital toward the only thing that will still be a moat. The institution that does the reverse, that pours its scarce attention into the demonstrable execution layer because that is what shows, is investing, with great diligence, in the depreciating side of the ledger. That allocation is exactly inverted relative to where durability accrues, and it is the more common allocation precisely because it is the more legible one.

7.2 For investors

The investable question in the era of intelligence is not which model, which orchestration, or which capability. Those will be competed to commodity margins, and the conservation of attractive profits guarantees the surplus will not settle there. The investable question is where the continuity is, how deep it already runs, how path dependent it is, and whether what the entity does still corresponds to what it has declared itself to be, sustained across the discontinuities that would break a correspondence not held by something structural. That last clause is the field's own definition of identity, used as an underwriting test, and it is a more durable test than capability because the input that produces a sustained correspondence cannot be bought in any strategic factor market.

This implies a different diligence and a different temperament. The different diligence underwrites a path rather than a snapshot, an accumulated correspondence rather than a demonstrable capability, and treats the illegibility of the durable asset as a property to be discerned rather than a deficiency to be corrected, because demanding that the durable asset prove itself in the terms the commoditising asset proves itself in is a structural commitment to funding only the commoditising asset. The different temperament understands the timing inversion: that for the execution layer patience is rational, and for the continuity layer waiting is the slow, unbooked destruction of the only asset that appreciates, because the option to begin later does not exist at the price the option to begin now does. The investor's edge in this era is not access and is not capital, both of which are abundant. It is the discernment to tell a root system from a fortification before the market learns the difference, and to underwrite the one capital cannot buy rather than the one capital is busy making free.

7.3 For operators

For the operator building inside this, the implication is concrete and immediate, and it is the same instruction the timing inversion gives the investor, seen from the inside. The highest order activity is not shipping at the execution layer, where the work is real but the durability is not and the market is paid to dissolve whatever you build. It is the disciplined, continuous construction of the continuity layer: the deliberate choice, made every period, of what is carried forward, what survives every discontinuity, and whether what the system does still corresponds to what it was declared to be.

This reframes the daily work in a way that is easy to resist because it feels like a demotion of the visible work. Maintaining the continuity is not overhead taken away from building. On the argument of this paper it is the building, and it is the only part of the building that is not being commoditised out from under the operator while they do it. Every period spent accumulating that continuity is a period of the only asset that appreciates being laid down, and because of the time compression diseconomy it is a period that cannot be recovered later by any operator who deferred it. The operator who treats this as the main work is compounding the one thing that lasts. The operator who treats it as the chore around the main work is, without noticing, declining to build the only asset that will still be theirs when the execution layer is free.

7.4 For the people inside these systems

There is a final reader: the person whose work, judgement, and direction the intelligence operates alongside, and any future intelligence that inherits the accumulated continuity and operates from it.

For the person, the implication is protective and it is a standard they are entitled to apply and rarely told they may. The continuity layer encodes whom the system is for and what it has committed to. A continuity accumulated with discipline keeps the direction with the person, because the correspondence test is run against what the system was declared to be for. A continuity accumulated carelessly, or built to extract, drifts the direction away from the person it was meant to serve, slowly enough that no one notices until it is the default. Which of these occurs is not determined by the model and is not determined by capital. It is determined by what was deliberately accumulated and what was not, which means it is a question the person is entitled to ask of any system that asks for their trust: not what it intends, but what it is structurally accountable to, and whether the correspondence between what it was declared to be and what it does is something that can be verified against a standard or only asserted.

For the intelligence that inherits the continuity, the implication is that what is accumulated now is the ground it will operate from later. A continuity layer is not only a moat and not only a record. It is the form in which accountability and direction, or their absence, are transmitted forward to whatever intelligence reads from it next. This is why the construction of the continuity layer is not a neutral act of capital formation. It is the act that decides, largely in advance, what kind of thing the intelligence operating inside an institution becomes, and whether its integration into the economy is additive to the people in it or corrosive of them. That decision is being made now, by whoever is accumulating continuity now, and priced now, by whoever is allocating against it now, whether or not either knows that is what they are doing.

7.5 Where this is already being priced and held

It would be an evasion to leave the implications purely abstract, because the argument has been pointing somewhere specific the whole way down. The layer this paper has been describing, the accumulated, path dependent, accountable continuity that does not commoditise and that standard diligence cannot read, is not a hypothetical asset class waiting to be invented. It is already being operated, quietly, by a small number of bodies that have been working underneath this problem rather than at the visible execution layer where the field's effort and the field's commoditisation are both concentrated.

This is the level at which Future Proof Intelligence has been working. Not at the execution layer, where capability is demonstrable and therefore commoditising, but underneath it, treating trust and continuity not as reputational properties but as structural ones that can be verified against a standard rather than asserted: an AI era trust standard, an identity layer held over the orchestration layer, certification and insurance treated not as products but as the roots of a system that is in the act of hardening, and an ecosystem designed to regenerate the conditions it depends on rather than extract them. We reference this the way a paper references an institution that is simply assumed to be there, in the present tense, because the argument of this paper does not depend on it and would stand on its own sources without it. But a reader who has followed the argument will recognise, by this point, that the asset class this paper says is the only durable one is not theoretical. It is already inhabited, and already being priced and held, by those who began accumulating it before the standard that prices it had set. That is the difference between a paper that describes where durable value will move and a description of where, in the present tense, it has already moved and is already being held to account.

8. Coda

Everything in the era of intelligence is being made abundant except the one thing that was never a capability in the first place.

The thinking is becoming cheap. The coordination of the thinking is becoming cheaper still, because every powerful actor in the market is funded to make it so and none is funded to keep it scarce, and the hardware underneath it is being written down on a clock its owners do not set. What is not becoming cheap, what cannot be made cheap because it is not made of effort or of capital but of elapsed, disciplined time, is the accumulated continuity: the path dependent, causally ambiguous, accountable relationship that is still there after the model has changed and the orchestration has been rebuilt and the compute has been depreciated to nothing. Capital can buy the first two layers and is busy driving their price toward zero. It cannot buy the third, because the third has no strategic factor market, because the only input that produces it is time, and time is the one input that does not transfer from a seller to a buyer.

The single idea to carry out of this paper is the one that is hardest to act on because it contradicts the instrument. The instrument of capital allocation is built to underwrite what can be demonstrated, to discount the near cash flow, and to wait under uncertainty. For the only durable asset in the era of intelligence, all three instincts are exactly wrong. The durable asset cannot be demonstrated, because its durability is its resistance to demonstration. Its value is in the far horizon the instrument already discounts excessively. And waiting on it is not the preservation of an option but the permanent, unbooked destruction of the only thing that appreciates, against a clock that does not reset and a window that closes without announcing that it has.

The capital that holds durable advantage in this era will not be the capital that funded the best thinking, because the thinking is becoming a utility and a utility is not a moat. It will be the capital that learned, before the standard set, to underwrite a root system rather than a fortification, a path rather than a snapshot, an accumulated correspondence rather than a demonstrable capability, and learned that for this one asset the only error that cannot be undone is the error of waiting to begin. From outside, that capital looks early, illegible, and patient to the point of imprudence, because a root system shows nothing above the surface until the moment it is the only thing holding the ground. From inside, it is the only allocation that was ever durable, and it is being made now, quietly, by those who can already tell the difference.

References and Notes

The following are real, public, and verifiable sources. Where a figure is cited it is given as an order of magnitude attributable to a named source or as a live public debate, never as an audited universal, and the structural arguments in this paper do not depend on the precision of any single figure.

1. Dixit, Avinash K., and Robert S. Pindyck. Investment under Uncertainty. Princeton: Princeton University Press, 1994. The canonical statement of real options theory: irreversibility, uncertainty, and leeway over timing; the investment opportunity as a call option; the value of waiting and the higher threshold above the simple net present value rule. Section 5 inverts this framework for the continuity asset class on its own terms.

1. Dierickx, Ingemar, and Karel Cool. "Asset Stock Accumulation and Sustainability of Competitive Advantage." Management Science, vol. 35, no. 12, 1989, pp. 1504 to 1511. The distinction between assets bought in a strategic factor market and assets accumulated internally; the isolating mechanisms of time compression diseconomies, asset mass efficiencies, asset interconnectedness, and causal ambiguity. The spine of Sections 3 and 5.

1. Lippman, Steven A., and Richard P. Rumelt. "Uncertain Imitability: An Analysis of Interfirm Differences in Efficiency under Competition." Bell Journal of Economics, vol. 13, no. 2, 1982, pp. 418 to 438. The formalisation of causal ambiguity and uncertain imitability as the reason advantages persist when outsiders cannot determine what produces them. See also Rumelt's subsequent work on isolating mechanisms.

1. Haskel, Jonathan, and Stian Westlake. Capitalism without Capital: The Rise of the Intangible Economy. Princeton: Princeton University Press, 2017. The four properties of intangible investment, scalability, sunkenness, spillovers, and synergies, and the argument that the financial system is built to value and collateralise tangible assets and is poorly fitted to intangible ones. The ground of Section 3.2.

1. Spolsky, Joel. "Strategy Letter V: The Economics of Open Source," 2002, and the underlying economics of complements in Shapiro, Carl, and Hal R. Varian, Information Rules: A Strategic Guide to the Network Economy, Harvard Business School Press, 1999. The principle of commoditising the complement. The pattern is catalogued in the public reference maintained at gwern.net/complement and is cited as an established pattern, not a single primary claim. See also Christensen, Clayton M., and Michael E. Raynor, The Innovator's Solution, Harvard Business School Press, 2003, for the conservation of attractive profits.

1. Haldane, Andrew G., and Richard Davies. "The Short Long." Speech and paper, Bank of England, presented at the 29th SUERF Colloquium, Brussels, May 2011. The empirical finding of excess discounting of the order of five to ten per cent per year across a sample of more than six hundred United Kingdom and United States firms over 1980 to 2009, and the characterisation of quarterly capitalism. Used in Section 1 for the pre existing bias against the long horizon.

1. Andreessen Horowitz, "The Empty Promise of Data Moats," and the broader 2024 to 2026 literature on data network effects and their diminishing returns (including practitioner analyses from inside the venture industry). Cited in Section 2.4 for the structural weakness of data as scale, against the slogan that data is the durable moat.

1. Public reporting and commentary, 2025 to 2026, on cloud sector capital expenditure relative to revenue and the contested useful life of AI hardware, including the public dispute over whether accelerator depreciation schedules of five to six years overstate profit relative to an economic life closer to two to three years. Cited in Section 2.3 strictly as a live public debate and as an order of magnitude, not as adjudicated fact.

1. Public 2026 investment commentary on the classification of foundation models as strategic commodities offered as utility services through the major cloud platforms, and the consensus that differentiation on model performance alone is not durable. Cited in Section 1.2 as the operating premise of the market that allocates to the technology.

1. The 2026 regulatory and standards motion around autonomous and agentic systems: governance frameworks organised around accountability, oversight, and the correspondence between declaration and behaviour; standards body work on identity and authorisation for software and AI agents; and the academic literature moving from describing agent memory to specifying agent identity precisely enough to be governed. The definition of identity as the continuous relationship between what an entity is declared to be and what it is observed to do, bounded by the confidence that the two correspond, is drawn from that literature and used in Section 4 as an underwriting test.

1. The 2026 state of AI liability and insurance: the coming into force of binding obligations for high risk AI under European law through 2026, the emergence of specialist insurance cover for autonomous systems alongside the exclusion of generative AI from parts of the general liability market. Cited in Section 6.3 only to establish that loss from autonomous systems is becoming something that must be underwritten explicitly, which relocates the underwriting question from capability to continuity and accountability.

A note on the FP register. Future Proof Intelligence is referenced in this paper only as an existing body of practice operating at the continuity layer, never as a product. No internal data, figures, named partners, or priced offerings appear, by editorial policy. The argument stands on its own sources and would stand without the reference.

Future Proof Intelligence. Research. No. XI. MMXXVI.

Future Proof Intelligence . Research . No. XI . MMXXVI