The Digital Age: Why Accounting Standards Overlook Your Most Valuable Assets?

Current International Accounting Standard (IAS 38) defines an asset as "a resource controlled by the entity as a result of past events and from which future economic benefits are expected to flow to the entity." The focus of this "resource ontology" is first on establishing the existence of a "resource," and then on determining its "control." According to this logic, control is typically tightly linked to legal ownership. IAS 38 explicitly states that control generally stems from legally enforceable rights, and that employee skills, training, specific technical talent, and even customer relationships and loyalty, lacking legal protection, usually do not qualify as intangible assets. This set of standards, in principle, excludes "people" and "relationships" from the scope of assets, and its limitations are becoming increasingly apparent.

Another key condition for asset recognition is that "cost can be measured reliably," meaning a high probability of future economic benefits flowing in and traceable costs. However, even for internally generated brands, mastheads, or customer lists, subsequent expenditures are invariably expensed. This institutional design creates a systematic bias: truly valuable assets incubated internally – such as brand influence, proprietary processes, customer relationships, and organizational capabilities – are the most difficult to recognize as assets. Conversely, only intangible assets acquired externally more easily meet the recognition criteria. Furthermore, the standards emphasize the "separability" of an asset, i.e., whether it can be sold individually. In 2019, the IFRS Interpretations Committee determined that paying only for access to cloud software does not constitute a software asset, but merely a service contract, further reinforcing the traditional notion that "ownership is an asset, access is an expense.

With the rapid advancement of AI technology, digital content, code, and model outputs can be replicated almost infinitely at near-zero marginal cost, fundamentally disrupting the underlying logic of traditional "scarcity." While older paradigms attempted to address the peculiarities of information goods through patents and copyrights, AI has transformed this "exception" into a "rule." When the output itself is no longer scarce, the premise of traditional definitions, which posit assets as "scarce objects," begins to crumble. A prime example is model weights: in January 2025, DeepSeek released its R1 model weights, whose performance rivaled top-tier inference models of the time. If cutting-edge capabilities can be easily replicated, then merely "owning weights" no longer equates to possessing core assets.

The value of data is also impacted. Data itself is non-excludable and has extremely low replication costs; a single data item rarely generates definitive revenue, and the value of the same data varies significantly across different models, tasks, and time points. More importantly, data without clear boundaries often represents a liability rather than an asset for enterprises, as it can entail high storage, compliance, and privacy risks. Consequently, true scarcity has shifted from the output itself to its surrounding "boundary layers" – mechanisms that establish exclusivity, enable independent verification, and clearly assign responsibility. In an AI-driven economy, value is migrating from the "object" itself to the boundary and control layers surrounding the "object.

To determine whether current discussions on digital assets are about "shoehorning data into old definitions" or "fundamentally replacing old definitions," the key is to observe whether the constituent elements of an asset (e.g., control, measurement, separability) and its source of value (whether it's the object itself, or the boundary and control layers built around the object) have undergone fundamental changes. For example, recognizing data and models as intangible assets under the IAS 38 framework, or using more advanced valuation models for data, or even securitizing assets to increase liquidity, all represent a "deepening" along the "asset → tradable" continuum of the old paradigm. They still revolve around the core question of "is this mine?

However, a true "revolution" in asset definition manifests in four core shifts. First, value migrates from the mere "object" to the "boundary control layer," meaning an asset's value is embodied in its verification, exclusivity, and attribution mechanisms. Second, the source of an asset's "control" becomes generalized, no longer confined to legal ownership, but extending to dimensions like computing power, interfaces, evaluation standards, provenance chains, reputation, and responsibility. Third, the "measurement benchmark" for assets changes from "verifiable initial cost" to "independently verifiable by a third party ex-post." Finally, the "depreciation logic" needs to be rewritten, transforming from linear time-based amortization to dynamic depreciation driven by technological updates and capability drift. If measured by old standards, enterprises will systematically miss out on truly emerging assets while potentially overestimating existing "objects" that are rapidly commoditizing.

Factually, the text of IAS 38, paragraphs 15 and 20, explicitly excludes "people, relationships, and internally generated brands" from intangible assets. The 2019 IFRIC interpretation of SaaS also drew a clear line that "right of access does not equal asset ownership." These instances reflect the lag of traditional standards when confronting new business models. It is noteworthy that the International Accounting Standards Board (IASB)'s intangible assets project discussed stakeholder feedback in February 2025, indicating that standard-setters are responding to market changes, although a final decision has yet to be made.

Market feedback, meanwhile, is more direct and forceful. Studies indicate that intangible assets now constitute approximately 90% of the S&P 500's market capitalization (compared to only 17% in 1975), suggesting a qualitative change in the composition of corporate value, though precise figures still require firsthand verification. More strikingly, in October 2025, OpenAI achieved a valuation of approximately $500 billion through a secondary share sale of around $6.6 billion, an astronomical figure far exceeding its book assets, with its core value clearly outside traditional categories. Furthermore, by late 2025, BlackRock's tokenized money market fund BUIDL surpassed $2 billion in assets and accumulated over $100 million in dividends, revealing the possibility that asset boundaries can, for the first time, be coded, automatically executed, and freely combined. This is no longer merely "discovering" assets but a completely new way of "manufacturing" them. However, Nvidia's success also reminds us that competitive physical capital (such as GPUs, electricity, manufacturing capabilities) remains hard currency, never having left the scene, and continues to play a foundational role in the digital economy.

In the AI era, we propose a more operational definition of an asset: an asset is a set of executable boundaries established around a "Flow," including exclusive boundaries, independent verifiers, and attributable responsible parties. Its value is jointly determined by "Flow Potential × Boundary Executability × Verification Credibility × Responsibility Attributability." The old definition asked, "Is this my resource?" while the new definition focuses on, "Can I establish exclusivity over this flow, prove its value, and take responsibility for it?" The essence of an asset is transforming from a static "noun" (object) to a dynamic "verb" (boundary engineering), reflecting its functional and constructive nature.

Based on this, the four key elements of an asset are redefined. First, "control" evolves from "legal ownership" to "executable exclusivity." Second, "measurement" shifts from "reliably measurable cost" to "independently verifiable by a third party ex-post." Third, "separability" is no longer "salable individually," but rather emphasizes its characteristics of "orchestrability, authorizability, and accessibility." Finally, "scarcity" transitions from "physical scarcity" to "complementarity with verification, distribution, trust, and responsibility." Consequently, asset categories expand to five coexisting types: competitive physical capital (e.g., GPUs), non-competitive cognitive capital (e.g., model weights, data, algorithms), process capital (workflows, evaluators, memory, provenance ledgers), relationship/trust capital (delivery records, reputation, accountability), and proactive capital (Agent systems, whose value requires continuous operation and oversight). For one-person companies, their core advantages often concentrate on the latter three categories, rather than the first two with high capital thresholds.

Enterprises should immediately re-evaluate their true asset lists, rather than relying solely on book records. Ask yourselves, "If everything had to start from scratch tomorrow, what three things could not be quickly rebuilt?" The answer is most likely long-established workflows, exclusive evaluation systems, and reliable delivery records, rather than mere data piled on hard drives. Second, for every critical information flow, a "three-piece set" should be equipped: exclusivity achieved through interfaces and keys, independent verification using replayable evaluators, and clear attribution through a transparent provenance chain and statement of responsibility. Here, evaluation standards serve as asset boundaries, and provenance records function as digital property rights.

Third, transform data from a liability into a true asset. This means stopping blind hoarding and instead building contextually rich and composable datasets around specific tasks; data that cannot be verified or attributed should be decisively purged. Fourth, re-estimate asset depreciation based on the pace of AI technology updates and market environment changes. For instance, an enterprise's capability assets may depreciate due to the emergence of stronger models (capability depreciation), become obsolete due to interface changes (drift depreciation), or even have their verifiers expire (verification depreciation). The useful life of an asset is no longer determined by contract duration but by the speed of updates it faces. Fifth, and most crucially, future opportunities lie not in "discovering undervalued assets" but in "designing executable boundaries that others do not yet possess." Assetization is never a passive discovery; it is a process collectively shaped and manufactured by institutions and technology. In the AI era, this manufacturing process becomes programmable and customizable for the first time.