The Most Valuable Thing You Own Might Be Worth Zero on Your Books

This isn't because you're careless with the books. It's because the definition of the word "asset" is having the ground swapped out from under your feet.

The current international accounting standard (IAS 38) defines an asset as a **resource** controlled by an entity as a result of past events, from which future economic benefits are expected to flow. Note where the emphasis falls: first comes the resource, then whether it can be controlled. The resource is the substance; control is merely attribution.

Follow that logic downstream and you arrive at conclusions that sound jarring today:

**Control usually rests on legal ownership.** IAS 38 states plainly that control generally arises from legal rights enforceable in a court; without legal rights, demonstrating control is harder. It then goes further: a team of skilled employees, training, and specific technical talent generally do not constitute intangible assets; customer relationships and loyalty unprotected by legal rights likewise usually don't count. **【verified】** In other words, the old paradigm in principle refuses to recognize "people" and "relationships" as assets.

**Recognition depends on whether cost can be measured reliably.** There are only two recognition criteria: future economic benefits are probable, and cost can be measured reliably. Brands, mastheads, newspaper titles, and customer lists—whether externally purchased or internally generated—have all subsequent expenditure charged to profit or loss. **【verified】** The result is a systematic bias: **the things grown internally, which are most likely to be genuine assets (brand, process, relationships, organizational capability), are precisely the hardest to record as assets; only what is bought from outside goes onto the books easily.**

**Liquidity is treated as a proxy for value.** Only if something is separable and can be sold on its own does it count as "identifiable." In 2019 the IFRS Interpretations Committee drew a hard line: paying solely for future access to cloud software does not create a software asset—it's merely a service contract. **【verified】** Ownership is an asset; access is an expense.

This "resource ontology" has governed tax, M&A, and equity compensation for decades. It always asks the same question: **Is this thing mine?**

Generation is now nearly free. Content, code, and model output can be replicated infinitely at zero marginal cost. The old paradigm had seen information goods before—it just treated them as a special case, patched with patents and copyright. AI turned that special case into the general rule. When the product itself is no longer scarce, the premise of "the scarce object" loosens. **【inference】**

The sharpest example is model weights. In January 2025, DeepSeek released R1, publishing the weights and distilled models, with performance matching the top reasoning models of the day. **【verified】** If frontier capability can be publicly replicated, then "owning the weights" no longer means owning an asset—value has to land somewhere else: compute supply, data pipelines, distribution channels, evaluation systems, compliance, and liability.

The same goes for data. Data is non-excludable, costs almost nothing to copy, and a single item has almost no certain return; the same dataset is worth wildly different amounts across models, tasks, and points in time. **Data without boundaries is a liability (storage, compliance, privacy risk), not an asset.** **【inference】**

So the genuinely scarce thing surfaces: when output is no longer scarce, what's scarce is "whether this thing is correct, who made it, and who's accountable when it breaks." Value migrates from the object to the **boundary layer** around it—the layer that can **establish exclusivity, be independently verified, and carry clear attribution**. **【inference】**

To distinguish "stuffing data into the old definition" from "replacing the old definition," you have to look at two things: whether the asset's **constituent requirements** (control, measurement, separability) and its **source of value** (the object itself, or the boundary-and-control layer around it) have changed.

Recognizing data and models as intangible assets by analogy to IAS 38 is **deepening**. Pricing data with a better valuation model is **deepening**. Securitizing real-world assets to add liquidity is still **deepening**—merely an extension of the old paradigm's line from "asset" to "tradeable."

There are only four genuine **revolutions**:

1. **Value migrates from the object to the boundary-control layer** (verification, exclusivity, attribution);

2. **The sources of control generalize**, extending from legal ownership to compute, interfaces, evaluation, provenance, reputation, and assumption of liability;

3. **The measurement basis changes**, from "the original cost is on file" to "it can be independently verified after the fact by a third party";

4. **The logic of depreciation is rewritten**, from linear amortization to update-driven capability depreciation and drift depreciation.

Measure by the old standards and you will systematically miss the real assets—while overvaluing the objects currently being commoditized.

A few facts worth placing side by side:

The body of IAS 38 and paragraphs 15 and 20 show how the old paradigm excludes "people, relationships, and internally generated brands." **【verified】** The 2019 IFRIC decision on SaaS drew the line that "access ≠ asset." **【verified】** The IASB's intangibles project discussed stakeholder feedback in February 2025 but has not yet reached a decision—standard-setters are already responding. **【verified】**

The market side is blunter. Ocean Tomo's oft-cited figure: intangible assets account for roughly 90% of S&P 500 market value (about 17% in 1975)—the precise number still needs to be checked against a primary source. **【single source · pending primary-source verification】** In October 2025, OpenAI completed a roughly $6.6 billion secondary share sale at a valuation of about $500 billion—a figure that balance-sheet assets cannot begin to explain. **【verified · multi-source cross-check】** By late 2025, BlackRock's tokenized money market fund BUIDL had surpassed $2 billion in assets and more than $100 million in cumulative dividends. **【single source · pending primary-source verification】** The implication of the latter is this: **for the first time, an asset's boundary can be written as code, automatically enforced, and composed with others.** This is not discovering an asset—it is manufacturing one.

If all this feels too "new," just look at the hardest counterexample: Nvidia represents competitive physical capital—GPUs, power, sites, manufacturing capacity. The old asset is still the hardest asset, and it never left the stage.

Here is a definition you can put straight to use:

The old definition asks, "Is this resource mine?" The new definition asks, "**Can I establish exclusivity over this flow, prove it, and be accountable for it?**" The asset turns from a noun (a thing) into a verb (boundary engineering). **【inference】**

From this, the four requirements are rewritten: control shifts from "legal ownership" to "enforceable exclusivity"; measurement shifts from "reliable cost measurement" to "verifiability"; separability shifts from "can be sold on its own" to "can be orchestrated, licensed, and accessed"; scarcity shifts from "physical scarcity" to "complementarity with verification, distribution, trust, and liability."

The categories of assets also expand into five coexisting kinds: competitive physical capital (GPUs, power), non-rival cognitive capital (weights, data, algorithms), **process capital** (workflows, evaluators, memory, provenance ledgers), **relationship/trust capital** (delivery records, reputation, capacity for liability), and **agency capital** (agent systems, whose value is sustained only through operation plus supervision).

The natural position of the one-person company is in the third, fourth, and fifth categories—not the first and second—because the former require capital thresholds, while the latter are easily commoditized.

**First, re-inventory what you actually own.** Not what's on the books, but "if I had to start over tomorrow, which three things could I not rebuild quickly?" Most likely it's the workflow, the evaluator, and the delivery record—not the data on the hard drive.

**Second, fit every flow with a three-piece kit: exclusivity, verification, attribution.** Use interfaces and keys for exclusivity, replayable evaluators for verification, and clear provenance and liability statements for attribution. Evaluation standards are boundaries; source records are property rights. **【inference】**

**Third, convert data from a liability into an asset.** Don't hoard it—build context and combinations around specific tasks; if data can't be verified or attributed, delete it decisively.

**Fourth, re-estimate depreciation by the cadence of updates.** Your capability assets will be replaced by stronger models (capability depreciation), invalidated by interface changes (drift depreciation), and your verifiers themselves will expire (verification depreciation). **Useful life is set by the pace of updates, not by contract term.**

**Fifth, and most important: your opportunity is not "finding undervalued assets" but "designing boundaries nobody else has yet."** Assetization was never discovery—it is a process manufactured jointly by institutions and technology; AI makes that process programmable for the first time.

When copying becomes free, possession loses its meaning. What truly belongs to you is no longer what you own, but the ring of boundaries you can define, prove, and take responsibility for.

The old ledger will keep failing to see it. The market won't.