RWA, Meet RWL, Part 5: When the Asset Is Already Digital

Part 5 of “RWA, Meet RWL: Knowing What You Own in the Tokenized World,” a series about what tokenized assets represent, how their data reaches the blockchain, and why none of this eliminates the need for trust.

Parts 2 through 4 deliberately stepped away from the asset hierarchy introduced in Part 1. Before expanding that hierarchy, I wanted to look at the information used to describe tokenized assets, the people and institutions responsible for that information, and what happens when those trust structures fail. Now we can return to the taxonomy itself.

The original split between intangible legal claims and tangible physical assets still works for most conventional RWAs. But it does not cover the entire tokenized world. Some of the things now represented, accessed, coordinated, or traded through tokens were digital from the beginning. Others are services, licenses, network rights, or credentials that may not really be assets at all.

Series Navigation

Part 1: “Real-World Asset” Is Not an Asset Class
Part 2: The Oracle Problem
Part 3: Permissionless Does Not Mean Trust-Free
Part 4: No Blockchain Solution Fixes Real-World Lies
Part 5: When the Asset Is Already Digital

I thought the taxonomy was more or less done. In Part 1, I divided tokenized assets into two large families. There were intangible legal claims, such as stocks, bonds, bank deposits, invoices, royalties, and other rights created through contracts or law. Then there were tangible physical assets, such as gold, buildings, aircraft, machinery, art, wine, and all the other stuff that actually exists somewhere and can theoretically be inspected.

That seemed like a reasonable high-level split. And it is. But it still leaves out a growing collection of things that don’t fit very comfortably into either side.

What about data? What about an AI model? An autonomous agent? An hour of GPU time? Storage capacity? A software license? A digital identity credential? An in-game object? A vote in a decentralized network? A token that rewards someone for contributing compute, bandwidth, predictions, or machine intelligence?

Some of these are assets. Some are services. Some are licenses. Some are payment mechanisms. Some are governance rights. Some are evidence that a fact or status was certified by somebody else. And some, come on now, are really just speculative crypto tokens with “AI” sprinkled over the marketing materials. So we need a third branch.

3. Digital-Native Assets and Network Rights

The best name I can come up with for this category is Digital-Native Assets and Network Rights.

That’s admittedly a little clunky. But the clunkiness reflects the problem. Unlike a gold token or tokenized apartment building, the underlying thing may not exist independently in a vault, county registry, warehouse, or filing cabinet. It may originate as software, information, network capacity, or a right that exists only within a particular digital system.

The Organisation for Economic Co-operation and Development, or OECD, has made a related distinction between tokens representing pre-existing off-chain assets and tokens that are “native” to a blockchain and live exclusively on the ledger. But even that split isn’t enough for everything developing now. A dataset might sit on an ordinary cloud server while a token controls access to it. GPU service might be coordinated on-chain while the actual chips remain in somebody’s data center. An AI model might be digital, but the legal rights to use or reproduce it still depend on ordinary contracts and intellectual-property law.

In other words, many of these things are hybrids. They’re digitally native in one sense and firmly attached to old-fashioned legal and physical dependencies in another.

And before we go any further, here’s the most important point: the fact that something has a token does not tell you what kind of thing it is.

A token can be money. It can be a security. It can be a receipt. It can be a license. It can be a prepaid service credit. It can be a vote. It can be a credential. It can be a reward mechanism. It can be a claim against future revenue. Or it can be nothing more than an object people are willing to trade because they hope somebody else will pay more for it later. The token is a technical formWhich brings us to another. It is not the economic or legal substance.

3.1 Data and Information Rights

Data is probably the most obvious missing category. We’ve spent years hearing that data is “the new oil,” although that analogy has always been a little strained. Oil is depleted when it’s used. Data can be copied, combined, reused, corrected, leaked, stolen, sold to multiple parties, and used to create entirely new information. Sometimes the same dataset becomes more valuable when more people use it. Sometimes wider distribution destroys the whole reason it was valuable in the first place.

Still, data clearly has economic value. Companies buy and sell commercial datasets, market feeds, geospatial information, scientific data, consumer research, business intelligence, training datasets, model outputs, and all sorts of proprietary databases. The AI boom has only made the question more obvious. Who owns or controls the material used to train a model? Who has the right to license it? Did the people who produced the underlying content consent to that use? Is the dataset accurate, complete, biased, outdated, duplicated, contaminated, or outright stolen?

Tokenization can provide a mechanism for registering a dataset, controlling access, recording licenses, or distributing revenue. But “tokenized data” still doesn’t have one meaning.

A token might represent:

Ownership of copyright or another intellectual-property interest
An exclusive license to commercialize a dataset
A limited right to access or query the data
A subscription to a continuing data feed
Permission to run an algorithm against private data without downloading it
A share of future revenue generated from the data
A reward for contributing data to a larger pool

Those are wildly different rights, which is why simply calling all of them “data tokens” does almost nothing to clarify what the holder actually receives. Ocean Protocol’s own documentation illustrates the difference pretty clearly. It describes a Data NFT as representing the copyright or an exclusive license associated with a data asset, while its datatokens function as licenses to access the data or service. So one token may represent the claimed base intellectual-property right, while another provides permission to use the thing under specified terms.

That does not mean a datatoken holder “owns the data.” It may just mean the holder is allowed to access it.

And even the claimed ownership token comes with an obvious qualifier. Ocean’s documentation says the owner is entitled to the related rights assuming the claim is valid. There it is again. Blockchain can record a claim beautifully. It cannot independently establish that the person making the claim had the right to do so.

This gets especially messy with AI training data. A token could provide access to a well-documented, properly licensed dataset. Great. It could also provide access to a giant pile of material scraped from somewhere else by people who didn’t have the right to package and resell it. The access control can work perfectly in both cases.

So for tokenized data, the usual question, “What do I own?” expands into several questions:

What data actually exists?
Where is it stored?
Who created or collected it?
Who has the legal right to license it?
What uses are permitted?
Can the same access rights be sold to unlimited buyers?
Can the data be updated, deleted, corrected, or revoked?
What happens if privacy law, a court, or a contractual restriction prevents continued use?

A data token can make access programmable. It cannot make the underlying rights less complicated.

3.2 Compute, Storage, Bandwidth, and Digital Capacity

Next up is digital capacity. This includes compute, storage, bandwidth, rendering, AI inference, and similar services. These may feel like digital assets because the customer experiences them entirely through software. But underneath them are still GPUs, CPUs, disks, networks, power supplies, cooling systems, facilities, technicians, and companies or individuals responsible for keeping everything running.

This is where the broader idea of decentralized physical infrastructure networks, often called DePIN, starts to overlap with tokenization. The network coordinates physical resources through digital markets and incentives. The deliverable may be digital, but the infrastructure remains stubbornly real.

Akash Network, for example, describes itself as an open marketplace where users buy and sell computing resources. Providers offer capacity, customers lease it, and the network helps coordinate the transaction. That doesn’t mean the customer owns part of a server. The customer is buying the use of compute capacity under a particular arrangement.

Filecoin’s documentation describes storage deals as contracts in which one party agrees to store another party’s data for a specified period. Again, the core economic object is a service commitment. You’re not necessarily buying a hard drive. You’re buying an obligation to store and, depending on the arrangement, retrieve data.

The Render Network offers another version. Its documentation says GPU work can be purchased by burning RENDER tokens in exchange for Render Credits. Those credits are then used to request computing work. That looks less like ownership of a productive asset and more like a programmable service credit connected to a marketplace.

Would you own:

Which brings us to another taxonomy problem. Should a prepaid hour of compute be called an asset?

Accounting and finance people can argue over that one. A transferable right to future service may certainly have value. But it’s still not the same thing as owning the machine providing the service. The token may provide access to capacity, serve as the network’s payment mechanism, reward resource providers, support staking, or do several of those things at once.

And these systems reintroduce many of the same real-world issues from the earlier parts of this series. Is the claimed GPU actually available? Does it meet the advertised specification? Is the service fast enough? Is the provider in a jurisdiction where the customer is permitted to process the data? What happens if the provider disappears halfway through a job? Who is responsible if confidential data is exposed? Can a network claim to have massive capacity when much of it is unreliable, oversubscribed, or economically unavailable at the price users expect?

A proof system may establish that some storage or computation occurred. It still may not answer whether the service was useful, lawful, secure, timely, or suitable for the customer’s actual purpose.

So perhaps the right term here isn’t “digital asset” at all. It may be tokenized service capacity or a digital service entitlement. Less exciting for a pitch deck, maybe. More accurate though.

3.3 AI Models, Software, and Autonomous Agents

Now we get to “AI tokens.” What’s an AI token? Well, that depends entirely on who is using the phrase. And sometimes, I suspect, whether they’re trying to explain a product or just get you excited enough to buy something.

An AI token might be:

A utility token used to pay for model access or inference
A reward token for supplying compute, data, or model output
A governance token for an AI-related network
A staking token used to back a validator, model, or subnet
A license to use model weights or software
A share of revenue associated with an AI service
A token associated with a particular autonomous agent
A speculative token whose only meaningful connection to AI is its name

That is not an asset class. It is a marketing label covering several unrelated economic structures.

Bittensor is a useful example because its documentation explicitly describes its independent subnets as markets that produce “digital commodities,” including compute, inference, storage, predictions, data, and models. Miners produce whatever the subnet is designed to reward, validators evaluate the work, and the network distributes incentives through TAO and subnet-specific tokens.

Interesting model. But owning TAO does not automatically mean you own an AI model. Owning a subnet token does not necessarily mean you own the compute, data, or predictions generated inside that subnet. The token’s function relates to the network’s incentive, staking, pricing, and governance mechanics. Those mechanics may create value, but they are not the same as title to a particular digital commodity.

What would actual ownership of an AI model mean anyway?

Would you own:

The source code?
The trained model weights?
The training dataset?
The right to run the model?
The right to make copies or derivative versions?
The API business that provides access?
A share of the revenue produced by the model?
The brand name under which it is sold?

Those things can all be separated. An AI model is not one clean object in the same way that a particular building or aircraft is. It is assembled from software, training processes, data, model weights, configuration, infrastructure, and often continuing services supplied by third parties. A token could represent rights to one layer while giving the holder no rights at all to the others.

Autonomous agents add another level of confusion. A token could be associated with a specific agent, but what does that mean? Is the agent a piece of software, a running instance, a wallet, a brand, a collection of prompts and tools, or an ongoing service controlled by a developer? If the developer replaces the model, changes the prompts, adds tools, removes capabilities, or forks the agent into ten copies, which one does the token represent?

And then we get back to the whole theme of this series. Who is responsible for what the agent does? Who controls its credentials? Who can shut it down? Who receives its revenue? Who pays when it makes a mistake? Does a token holder have governance rights, economic rights, both, or neither?

Saying “the community owns the AI agent” is easy. Defining what that sentence means in law and in actual system control is much harder.

3.4 Intellectual Property and Programmable Licensing

Intellectual property already appeared elsewhere in this taxonomy because royalties and licensing revenue are contractual cash flows. But digital IP deserves a section of its own because tokenization can break ownership, licensing, attribution, derivative rights, and royalties into separate programmable pieces.

Possible underlying rights include:

Copyright ownership
Patent rights
Software licenses
Music and media catalogs
Character and brand rights
Rights to create derivative works
Model or dataset licenses
Royalty participation

Story’s documentation, for example, describes an IP Asset with an associated NFT representing ownership over the IP, along with modules for licenses, derivative works, disputes, and royalty flows. Its system also allows separate License Tokens to represent permission to use the IP under defined terms.

Again, ownership and permission are different. I might own the copyright to an image. You might own a commercial license allowing you to print it on shirts. Someone else may have the right to create a derivative version. Another person might be entitled to five percent of the revenue. Tokens could represent any of those relationships.

And here’s where purely digital material behaves differently from a physical collectible. A painting can be forged, but the original physical object remains singular. A digital file can be copied perfectly in seconds. The scarcity usually exists in the recognized edition, registry entry, contractual right, or official association with the creator. It does not necessarily exist in the underlying bits.

An NFT can make one token scarce, but it cannot make the image impossible to copy. Nor does the NFT itself prove that the person who created it owned the copyright. Someone can tokenize somebody else’s work just as easily as they can upload it to an ordinary website. The blockchain may give us a permanent record of who made the unauthorized claim first. That is not quite the same thing as solving the problem.

3.5 Gaming Assets, Digital Objects, and Virtual Property

There is also a broader category of purely digital objects. These include digital art, in-game items, domain names, virtual land, membership passes, collectibles, and other scarce or controlled objects that exist primarily inside software or a registry.

Some fit comfortably into the collectibles category from Part 1. Others behave more like licenses or platform permissions.

Suppose you “own” a sword in an online game. What do you really own?

Maybe you own a token that the game recognizes as giving your account the right to use the sword. But the publisher still controls the software, visual artwork, game rules, and servers. It may be able to change the sword’s capabilities, ban your account, discontinue the game, or alter whether the token is recognized at all.

Virtual land raises the same issue. It can be scarce within a platform because the platform defines a limited map and recognizes certain entries as property-like rights. But the land does not exist independently of the platform’s software, governance, user base, and continued operation.

A domain name is another interesting comparison. It has value, can be transferred, and may be central to a business. But it is ultimately a registry-dependent right to use a particular name under a particular system and set of rules. You don’t own the words in the abstract.

So these assets may be digitally native, but they are not necessarily independent. Their value can be entirely dependent on a platform continuing to recognize and support them.

That sounds a lot like the statutory and registry-dependent rights discussed in Part 1. The difference is that the governing authority may now be a protocol, software company, DAO, game publisher, or digital registry instead of a government agency. Different ruler. Same basic dependency.

3.6 Credentials and Attestations That May Not Be Assets

Here’s another area that tends to get mixed into tokenization even though much of it should not be called an asset at all.

Identity credentials, professional qualifications, educational records, compliance certifications, reputation scores, and proof-of-participation records can all have tremendous economic value. But that does not mean they should be transferable.

I cannot legitimately sell you my college degree, pilot certificate, security clearance, professional license, or clean driving record. Well, I suppose I could sell you a piece of paper or token claiming to represent it. That would be the problem.

The World Wide Web Consortium’s Verifiable Credentials work provides standardized ways to express claims made by an issuer, such as an educational certificate or license. Its decentralized identifier standard provides a method for identifying people, organizations, devices, data models, and other subjects without necessarily relying on one centralized identity provider.

These tools can help answer questions such as:

Who issued this credential?
Was it altered?
Does it refer to the person or entity presenting it?
Has it expired or been revoked?

Useful. Very useful, actually. But the credential is evidence of a claim or status. It is not necessarily property. Making it transferable would often destroy its meaning.

This distinction will become especially important for AI agents. Agents may need identities, credentials, permissions, spending limits, operating histories, and reputation. A business may need to know which company or person is responsible for a particular agent and what it is authorized to do.

That does not mean the agent’s credential should trade like a stock. Not everything valuable is an asset, and not everything recorded on-chain should become a market.

3.7 Network Utility, Governance, and Incentive Rights

Finally, we have the giant bucket of tokens used to make digital networks function.

These tokens may be used to:

Pay transaction or service fees
Purchase access to a product
Reward people who supply resources
Stake behind validators, providers, or subnets
Vote on protocol changes
Allocate incentives
Provide collateral inside a network
Signal demand, quality, or confidence

The OECD has described utility tokens as rights to access particular goods or services, often functioning like prepayment or a voucher rather than representing ownership of an asset. That’s a useful distinction, although modern network tokens often do several jobs at once.

A token could simultaneously serve as payment, staking collateral, a governance vote, and an incentive reward. That makes it harder to classify and harder to value. Is its price supported by demand for the underlying service? By speculation? By the need to stake? By expectations of governance power? By token issuance rules? By the hope that the network eventually becomes important?

And what exactly do governance rights get you? Owning a governance token does not necessarily give you equity in a company. It may not give you a legal claim against revenue, assets, or management. It may let you vote on selected protocol settings, assuming enough other holders participate and assuming the people operating the software actually implement the result.

That can be meaningful, but it is not the same thing as owning common stock. This is why “utility token,” “governance token,” “AI token,” and “DePIN token” are not enough. They describe a theme or one possible function. They don’t fully describe the holder’s legal rights, economic exposure, control, or remedies.

So, Are These RWAs?

Some are. Some aren’t. Some sit directly on the border. A token that licenses access to a dataset stored on ordinary servers may look a lot like an intangible legal claim tied to an off-chain asset. A tokenized right to GPU service is a digital service entitlement backed by physical infrastructure. A token representing copyright ownership is an intangible legal claim, even if the copyrighted work is digital. A governance token native to a protocol may have no real-world backing at all.

That suggests a more complete taxonomy:

Traditional or Off-Chain Things Represented on a Ledger

Money and financial claims
Receivables and contractual cash flows
Government- or registry-dependent rights
Physical commodities and property
Collectibles and other singular objects

Digital-Native Assets and Rights

Data and information rights
Software, AI models, and digital IP
Digital objects and virtual property
Network utility, governance, and incentive rights

Digital Service Entitlements

Compute
Storage
Bandwidth
Rendering
AI inference and other processing services

Credentials and Attestations That Aren’t Really Assets

Identity credentials
Professional and educational qualifications
Compliance status
Reputation and operating history
Proof of participation or authorization

And then, of course, there are hybrids everywhere. An AI inference token looks digital, but the service depends on GPUs, power, data centers, software, model licenses, network access, and operators. A data token may control a digital dataset, but the right to use that data depends on contracts, privacy rules, copyright, and the behavior of whoever hosts it. A virtual object may exist on-chain, but its usefulness depends on a game or platform continuing to recognize it.

We keep coming back to the same thing. Blockchain can help coordinate, record, transfer, and automate rights. It may create entirely new markets. It may make services easier to buy and resources easier to monetize. Those are real possibilities, and some are genuinely exciting. But it does not relieve us of the need to identify the underlying thing.

The Questions Don’t Change

Whether a token represents a building, an invoice, a dataset, an hour of GPU time, an AI model, a digital sword, or a vote in a network, the label tells you almost nothing by itself.

You still need to ask:

What actually exists?
What does the token represent?
What legal, contractual, or technical rights do I receive?
Is the right transferable?
Who controls the underlying system?
Who can change the rules?
Who verifies performance?
What happens if the service, platform, model, or network disappears?
What remedy do I have if the promised thing was never legitimate?

“AI token” doesn’t answer any of that. Neither does “data token,” “utility token,” “governance token,” “DePIN token,” or even “RWA token.” They are starting points for more questions, not answers.

So maybe the broader lesson from this entire series is that we should stop treating tokenization as if it defines the asset. It defines a way of recording, transferring, accessing, or coordinating something. The something is still what matters.

Forget about the category label for a moment. Forget about whether it’s real-world, digital-native, AI-powered, decentralized, programmable, or any of the other words we’re going to pile onto these things.

The same question remains: What do I actually own?