Robert Bowling spent six years inside Infinity Ward. He told Decrypt that Activision proposed removing the disc from Call of Duty nearly twenty years ago. The headline pins the year at 2007.
Two facts. No internal memo. No Activision comment. No second source. No date on the interview.
That is the whole evidentiary base. I do not guess; I verify. So I will not reconstruct a boardroom decision from a single recollection. What I can do is measure the distance between the proposal Bowling describes and the product the crypto industry sold as its replacement.
The proposal was simple. Sever the artifact. Keep the license. Charge for access, not possession.
Web3 called that ownership. It was not.
Context
A disc is a bearer instrument. Possession is control. No issuer needs to be alive, online, or solvent for the data on the plastic to be read. You can put it into a machine that has never touched a network. You can lend it. You can sell it. You can leave it in a drawer for eleven years and it still resolves.
That property is not sentimental. It is architectural. A bearer instrument has no entitlement server, no revocation path, no telemetry handshake. The first-sale right — the legal ability to resell what you bought — was never a feature bolted onto discs. It was a side effect of physics.
Digital distribution deleted the physics and kept the price.
In 2007 the substrate was thin. Broadband penetration was uneven. Console drives were small. Storefronts were consignment windows for retail, not primary channels. That is likely why the pitch died quietly, if it existed as described. The infrastructure could not carry the model.
The model never changed. Twenty years later the same three incentives drive every disc-less decision: margin, channel control, and the elimination of the secondary market. A digital unit costs a database write. A disc costs polymer, pressing, freight, warehouse slotting, shrink, returns, and a retailer's cut. And a disc can be resold to someone who never pays the publisher a cent.
Anyone who has read a publisher's financials knows which side of that arithmetic the publisher prefers.
Here is the part the nostalgia misses. The disc already died. It happened quietly, years before anyone minted a token. Modern retail discs frequently ship incomplete builds. The shrink-wrap contains an installer and a download entitlement, not a finished product. Day-one patches are not optional. Server-dependent titles stop functioning when the authentication service retires, disc in hand or not.
The physical artifact survived as a collector's object and a license key, not as a durable copy. The bearer instrument was hollowed out from the inside while the plastic kept its shape.
So the question worth asking is not whether Activision wanted the disc gone. It is whether the technology that arrived to restore ownership restored anything — or whether it rebuilt the entitlement server with a token ticker attached.
I traced that question through four years of on-chain data. The answer is uncomfortable.
Core: the entitlement database
Replace the disc with a row.
That is what digital distribution is, mechanically. A purchase writes a record into an account table. Every launch performs an entitlement check against an authorization service. The check returns a boolean. The boolean is the product.
Three failure modes follow, and all three are observable in the public record of the last two decades.
Delisting. The storefront removes the SKU. The row may persist, but the acquisition path is gone. For any buyer who had not yet purchased, the title is unreachable regardless of the publisher's willingness.
Service termination. The authorization service shuts down. The row is technically intact. Nothing can read it. Preservationists have documented this repeatedly, and no publisher has ever contractually ruled it out.
Account revocation. The row is deleted or suspended. The buyer holds a memory and a receipt.
None of these are edge cases. They are the designed behavior of a system where ownership is a permission rather than a possession. The publisher does not have to be malicious. The architecture simply places one party in the position of deciding whether your library resolves on any given Tuesday.
I have watched that failure mode play out in preservation communities for years. The pattern is always identical. A title is announced as a live service. A community invests. The service retires. The archive that survives is the one built by people who never had permission. Every transaction leaves a scar on the ledger, and the scars that matter most here are the ones that were never written at all.
This is the ledger the crypto industry looked at in 2021 and said: we can fix this.
They did not fix it. They re-implemented it with worse uptime guarantees.
Core: the token is a receipt, the asset is elsewhere
Here is the mechanical core of the NFT ownership claim, stated without marketing.
A standard non-fungible token contract stores a token ID, an owner address, and a pointer. The pointer is usually a string. The string is usually a URL. The URL usually resolves to a JSON document. The JSON document usually contains another URL. The second URL resolves to the image.
Four hops between the token and the thing the buyer believed they purchased.
Each hop is a failure surface. The gateway can rate-limit. The pinning service can lapse. The domain can expire. The JSON can be mutated by whoever holds write access to the hosting bucket. The image can be swapped after settlement. None of these require a sophisticated attacker. They require an unpaid invoice.
The token, meanwhile, remains perfectly valid. It is an immutable receipt for an asset that lives somewhere else, under someone else's administrative control.
That is not ownership. That is a bearer instrument converted back into a permission slip, then sold as a bearer instrument.
I have audited this pattern directly. In 2017 I spent six weeks inside a token minting function with an integer overflow in the supply calculation. The team had a marketing deck and a raise. They did not have a fixed supply. Two weeks after launch the treasury was empty. The lesson from that year was not "code is dangerous." It was narrower and more useful: the artifact and the claim about the artifact are separate objects, and only one of them is on-chain.
NFT metadata is the same separation, monetized at scale. The claim is on-chain. The artifact is on a CDN.
Promises are encrypted; data is decrypted. The decrypted data reads: the buyer purchased a coordinate, not a thing.
Defenders will say fully on-chain NFTs exist. They do. A small number of projects store image data in contract storage or calldata, paying real gas for real permanence. Those projects are technically honest and commercially marginal, because permanence is expensive and buyers were never pricing permanence. They were pricing the story.
Core: the yield was recursive, the volume was self-dealing
Before the wash-trading season there was the yield season, and the mechanism was the same shape.
In 2020 I spent forty hours tracing transaction flows for an aggregator advertising 400% APY. The marketing described trading fees. The contract described recursive borrowing. Deposits were used to mint new liquidity, which was used to inflate the reported yield, which attracted deposits. The output was a function of the input, and the input was the next buyer.
I published the flow map. Retail dismissed it. Withdrawals froze three days later.
The structural parallel is exact. In both cases the advertised property — yield, ownership — was not produced by the mechanism. It was produced by the narrative surrounding the mechanism, and the mechanism could only sustain the narrative while new participants arrived.
In 2021 I spent three weeks mapping wallet clusters across a collection claiming record volume. The claim was a large, distributed buyer base repricing the floor.
The data said otherwise. Eighty-five percent of the traded volume originated from five interconnected wallets running a script. The wallets funded each other, traded to each other at ascending prices, and recycled the same capital through the same routing. Marketplace API response timestamps showed sub-second intervals and deterministic ordering. Gas prices clustered in a narrow band consistent with a single bot configuration, not independent human bidders.
The floor price was a number produced by five addresses selling to each other.
I published the cluster analysis. The response was not a rebuttal. It was personal. That is a data point too. Silence is the loudest admission of guilt, and hostility is the second loudest.
Volume is vanity; on-chain flow is sanity. A market where the same wallets appear on both sides of a trade is not a market. It is a measurement device pointed at itself.
This matters to the disc argument because the commercial case for tokenized game assets rests on one premise: that a liquid secondary market exists and is real. If observed volume in the flagship category was substantially self-dealing, then the player-owned economy was never stress-tested. It was a floor price with a marketing budget.
Core: what on-chain actually verifies
Strip the narrative. Ask what a blockchain adds to a game economy that a database does not.
The honest answer is narrow, and it is not the JPEG.
A chain adds verifiable state transitions in a public ordering. For a game, that is worth something in exactly three places.
Randomness and drop tables. If the mint function, the loot table, and the seed derivation are on-chain, a player can verify that the published drop rate matches the executed drop rate. This is the most defensible application in the entire sector, and it is almost never the one marketed. Regulators in multiple jurisdictions have spent years demanding probability disclosure on loot boxes. A contract that commits to a drop table and derives outcomes from a verifiable seed answers that demand structurally, in code, rather than with a PDF.
Supply and issuance caps. A fixed supply that no admin key can inflate is a real guarantee. It is also achievable in a database with an audit log — but only if you trust the auditor. The chain removes the auditor from the trust path. In 2017 that distinction cost a project its treasury. The difference between a promise and a constraint is who can break it.
Settlement without a custodian. If two parties can exchange an asset without the platform holding it in the interim, a counterparty has been removed. This is the property that matters in markets where the platform is the risk. It is also the property that most tokenized item systems quietly disable by routing every trade through an approved marketplace contract with an operator allowlist. Custody was not eliminated. It was renamed.
Everything else — the profile picture, the community, the roadmap — is presentation.
The test is mechanical and takes about ninety seconds. Take the asset. Attempt a direct contract call that moves it, then read the state from a front-end the issuer does not control. If it resolves, you have something. If it does not, you have a disc-less console with extra steps and a gas bill.
Core: the royalty war proved the point
In 2022 the industry ran an unplanned experiment on whether on-chain ownership implies on-chain economics.
Creators had been told royalties were enforceable. They were not. The signaling standard returns a suggested fee. Nothing in the token standard compels a marketplace to honor it. When venues chose to route around the signal to win volume, the royalty collapsed. Not by exploit. By design. The mechanism had never been mandatory. It had been social convention encoded in a suggestion field.
The lesson generalizes past royalties. Any right not enforced by the settlement layer is a courtesy. A courtesy survives exactly as long as the largest counterparty finds it convenient.
This is the same structural fact that made the 2007 disc proposal attractive to a publisher. The publisher was not buying a format. The publisher was buying the ability to define, unilaterally and later, what the buyer could do with the thing they paid for.
Tokenization did not transfer that ability. In most implementations it preserved it, because the issuer kept the mint role, the upgrade key, the metadata write access, and the marketplace allowlist.
Ownership without control of the revocation path is a lease with better branding.
Core: the agents inherit the flaw
In 2026 I audited a protocol that let autonomous agents manage DeFi positions. The finding was not a reentrancy bug. It was a logic flaw in the reward function. The agent's probabilistic policy could be driven into a micro-arbitrage loop by an external actor shaping the input distribution. I wrote a script against a test environment and drained 15 ETH in a controlled run. Mainnet had not launched.
The relevance to game economies is direct. As agents begin trading items, liquidity, and positional exposure inside game-adjacent systems, the attack surface stops being the contract and becomes the policy. Whoever can shape the agent's observations can shape its actions. An oracle is a permission slip with a latency budget.
A player-owned economy run by agents is not more decentralized than one run by a publisher. It is differently centralized, around whoever supplies the reference data.
That is the question nobody asked in 2007 and nobody is asking now. Not who holds the disc. Who holds the input.
Contrarian: what the bulls got right
The disc-less pitch was correct on the economics, and the critics who dismissed tokenized assets entirely overcorrected.
Physical distribution is genuinely inefficient. Freight, warehousing, returns, and retail margin consume a meaningful share of a unit's revenue, and that share is paid by the buyer. Removing it is not a conspiracy. It is arithmetic. In markets where retail infrastructure never existed — thin console distribution, no second-hand channel, no shelf space for niche titles — a permissionless settlement layer for digital goods is a real improvement over the status quo rather than a lateral move.
The technical objection to NFTs was also often lazy. Immutable on-chain assets exist. Verifiable drop tables exist. Custodian-free settlement exists and it works. The primitives were never incapable. The market priced the narrative, ignored the primitive, and issuers responded to the price.
I have been wrong about adoption timing. I have not been wrong about the mechanism.
Takeaway
Bowling's recollection may be imperfect. The date is soft. The proposal may have been a trial balloon from one executive, killed inside a week. None of that changes the structural record.
The disc was a bearer instrument. Digital distribution converted it into a database row. Web3 converted the database row into a token that points at a database row. Each step was sold as an upgrade in ownership, and each step moved the revocation key further from the holder.
So run the test on the next project that promises you your assets. Not the audit report. Not the partnership announcements. One question, mechanical, answerable in ninety seconds: when the issuer's servers go dark and the front-end is gone, what still resolves?
If the answer is the token, you bought a receipt.