Skip to main content
● LIVE

Why AI tokenization could reshape digital asset markets

Mohana Priya By Mohana Priya
14 Min Read

AI tokenization could reshape digital asset markets because it turns compute, models, and training data into tradable claims that settle onchain. Capital is already moving: Story Protocol raised $140 million for intellectual property markets, while Nous Research and Prime Intellect drew major backing for openly owned models. Tokenized AI now has real revenue, real hardware, and real investors behind it.

Key Highlights

  • Story Protocol launched its mainnet in February 2025 after raising $140 million, including an $80 million round led by a16z crypto in August 2024
  • Nous Research raised $50 million at a $1 billion valuation in April 2025 and Prime Intellect took $15 million from Founders Fund in February 2025 to build openly owned models
  • Reported H100 rental rates fell from roughly $8 per hour in 2023 to about $2 per hour in 2025, squeezing the economics of GPU tokens like Render and Akash
  • The EU AI Act’s general purpose AI obligations took effect on 2 August 2025 and California signed SB 53 on 29 September 2025, both pushing labs toward documented training data
  • Virtuals Protocol’s agent token ecosystem approached $5 billion in value in January 2025 before falling more than 80 percent

The short answer to why AI tokenization could reshape markets is this: the scarce asset in artificial intelligence has moved. Between 2023 and 2025 the trade was compute, and tokens like Render and Akash sold exposure to graphics cards. That trade is finished. The scarce asset now is legal permission, meaning the rights to a model’s weights, the data it trained on, and the revenue those two things generate. Whoever builds the accounting layer for those rights controls the next cycle.

Why AI tokenization could reshape markets after the compute bubble

The first AI token cycle was a hardware story with a crypto wrapper. Render migrated from Ethereum to Solana in November 2023 and tied token supply to completed rendering work through a burn and mint model. Akash sold server capacity to anyone who could not get Nvidia allocation. Both worked because GPU supply was tight.

Then supply caught up. Rental rates for H100s, reported near $8 per hour in 2023, fell toward $2 per hour by 2025 as new data centers came online and Blackwell shipped. A token that represents a commodity with a falling spot price is not an investment. It is a derivative on hardware depreciation.

That failure cleared the ground for a different pitch. Ownership of the model, not the machine, is where the margin sits. A trained frontier model costs $100 million or more to build and cannot be copied cheaply. Bittensor’s 21 million TAO supply cap and its subnet structure, which pays emissions directly to teams producing useful model output, was the first serious attempt to price that work onchain.

Model ownership becomes something you can buy

The clearest 2025 signal was not a token price. It was where venture money went. Nous Research raised $50 million at a $1 billion valuation in April 2025 to train open models. Prime Intellect took $15 million from Founders Fund in February 2025 to run decentralized training. Both are betting that a model owned by its users can attract capital and researchers away from single lab control.

Crypto’s version of that bet is the subnet. Bittensor passed its first halving in late 2025 with emissions cut in half, and its subnets compete to produce models and data in narrow domains such as forecasting and protein work. A subnet token is not equity, but it behaves like a revenue share on a small research team.

The contrast is instructive. Virtuals Protocol, which tokenized AI agents on Base, saw its ecosystem value approach $5 billion in January 2025 before falling more than 80 percent. Agents without a product went to zero. Subnets with measurable output held better, because the market started pricing work instead of narrative.

Royalty streams are the piece crypto keeps missing

Data owners already get paid. Reddit signed a licensing deal with Google in February 2024 reported at $60 million per year. News Corp struck a five year deal with OpenAI in May 2024 worth a reported $250 million. Those numbers prove training data has a price and that publishers will litigate to defend it.

What does not exist is a public ledger of who owns what and who is owed how much. Story Protocol launched mainnet in February 2025 to be that ledger, using programmable license terms that attach to an asset and pay royalties when it is used.

The bottleneck is provenance. A royalty system only functions if you can prove a model ingested a specific work. That needs training data attestation and content credentials, and the standard already exists in the C2PA specification backed by Adobe, Microsoft and Google.

Regulation is quietly becoming a tailwind

For three years regulators were the reason to avoid AI tokens. That flipped in 2025. The EU AI Act’s obligations for general purpose AI models took effect on 2 August 2025 and require providers to publish a detailed summary of their training content. California followed with SB 53, signed on 29 September 2025, which imposes transparency duties on frontier developers.

Read that as a compliance mandate for the infrastructure crypto has spent two years building. A lab that must document its training corpus needs a registry. A publisher that licenses text needs an auditable usage log. Onchain royalty vaults are expensive and clumsy, but they are also the only neutral, timestamped record that two parties who do not trust each other can share.

Securities treatment is moving too. The CLARITY Act passed the House in July 2025 and the SEC’s Project Crypto signaled a taxonomy that separates tokens tied to active networks from passive investment contracts. A token representing a licensing claim with a paper trail fits the first bucket far better than a memecoin does.

What would break the thesis

Two things. First, the labs could keep licensing offchain. Microsoft, Amazon and Google already pay publishers through private contracts and internal databases, and none of them need a token to do it. If the biggest buyers never touch a public registry, onchain royalties stay a niche for independent creators.

Second, dilution. Bittensor’s emissions schedule, like every proof of work or proof of stake network, hands new supply to whoever shows up. Subnet tokens paying for research today can be diluted into irrelevance by 2028 if demand does not scale with issuance.

So watch the volume that actually settles. Story Protocol’s royalty vault activity, not its token price, is the number that matters. If licensed AI usage starts routing real dollars through those vaults, the infrastructure thesis is confirmed. If vault balances stay under a few million dollars through 2026, crypto is a solution looking for a counterparty.

Frequently Asked Questions

Why are AI tokens like Render and Akash losing their appeal?

The first AI token trade was a bet on graphics card supply. Reported H100 rental rates fell from about $8 per hour in 2023 to about $2 per hour in 2025, which squeezed the economics of GPU tokens.

What is the scarce asset in AI tokenization now?

It has moved from compute to legal permission, meaning the rights to a model’s weights, the data it trained on and the revenue those two things generate. Story Protocol raised $140 million and Nous Research raised $50 million, a sign of where investors see value.

What happened to Virtuals Protocol’s agent tokens?

Its agent token ecosystem approached $5 billion in value in January 2025, then fell more than 80 percent. The swing is a warning about how fast sentiment can turn in AI token markets.

How do new AI laws affect AI tokenization?

The EU AI Act’s general purpose AI obligations took effect on 2 August 2025, and California signed SB 53 on 29 September 2025. Both push labs toward documented training data, which makes clear rights and records more valuable.

The TCB View

TCB is bullish on AI tokenization, but only on the narrow slice that carries a legal claim and a cash flow attached to it. We see the winners as provenance and royalty infrastructure plus publishers sitting on licensed corpora that regulators now force labs to document, and the losers as GPU rental tokens and agent memecoins that repriced 80 percent lower in 2025 for good reason. The real risk is not regulatory, it is irrelevance: Microsoft can pay a publisher through a private contract and never touch a public ledger, and nothing in the EU AI Act forces it to. What changes that is scarcity. If a single frontier lab signs a data deal with onchain royalty settlement attached, the CLARITY Act’s taxonomy gives that structure a defensible legal home and every competitor has to answer it. Watch Story Protocol’s royalty vault volume through the end of 2026. Under $10 million in annualized payouts means crypto lost the accounting layer to the incumbents, and the whole category gets repriced again.


Free Daily Newsletter

The Daily Brief

What's moving crypto, AI and markets, explained in 5 minutes. Every weekday morning.

Free weekday newsletter  ·  No spam, ever  ·  Unsubscribe anytime

Share This Article
Follow:
Mohana Priya is a staff writer at The Central Bulletin, where she covers DeFi, Ethereum, Web3 infrastructure, and the fast moving world of AI. She has a talent for taking the most intricate protocol mechanics and regulatory shifts and turning them into journalism any reader can follow and trust. Research comes first in everything she writes, and her work consistently surfaces the detail others miss. For Mohana, the goal is never just to report what happened, but to explain why it matters.