Blockchain protocols designed to handle and process AI tasks could be poised for fast growth in multi-trillion-dollar markets forming around the costs of data computation, according to a new research report by the crypto financial firm Galaxy.

That might be due to their ease of use by AI agents, since many of the on-chain projects are "permissionless, open-source" and "programmatic" in nature, writes Lucas Tcheyan, a vice president of research at Galaxy. The idea is that agents, which are really just AI-equipped computer programs, could navigate over the Internet to platforms where they can process tasks and pay for them, without needing human involvement or necessarily setting up accounts.

These characteristics fit "the form most naturally suited to the class of agents likely to drive most of the demand for on-chain capital market inference," Tcheyan wrote.

Investors, analysts and researchers are paying close attention to whether blockchain projects could take market share from traditional finance and tech companies in the fast-growing AI industry, and these emerging "inference capital markets" could serve as a fresh proving ground.

The term represents the infrastructure used to route AI queries to computers that are best-equipped to handle them, or that cost the least. Platforms like OpenRouter let users choose from hundreds of AI computer models, and traditional financial firms are building trading systems for hedging and speculating on the cost of computation.

For example, indexes tracking the price of computer-chip rentals are now published daily on Bloomberg terminals, a widely used financial data provider across Wall Street, Tcheyan noted.

But crypto projects such as Venice and Morpheus could be able to muscle their way in, and the opportunity is now because it's still early, he wrote: "A multi-trillion-dollar asset class is being assembled in real time, and the market structure for compute as a financial asset – indices, futures, credit, tokenized capacity – barely exists yet."

Goldman Sachs and AI futures

The report details the extent to which big money is already moving in traditional markets, especially for financing the massive spree of data-center building.

Goldman Sachs, a major Wall Street bank, pegs AI infrastructure spending at $7.6 trillion between 2026 and 2031. ICE, the New York Stock Exchange's parent, plans to list GPU futures against an index from Ornn, which raised $33 million in an a16z-led round in June. (GPU stands for graphics processing unit, a type of computer chip that's increasingly used to process AI queries.)

CME Group, the Chicago-based commodities exchange operator, is working with Silicon Data on its own futures contracts.

"Compute as a commodity could soon be a reality," Tcheyan wrote.

Notably, over the past three months, on-chain players have served just 0.5% to 1% of AI jobs processed by OpenRouter, a popular app for steering tasks to the most appropriate or cost-effective artificial-intelligence computer program, according to the analyst.

"While this demonstrates some initial traction outside of the crypto-native community, it is still a small portion of total usage and demonstrates these providers cannot yet compete with established centralized offerings, whether due to lack of distribution, relative cost or other factors," Tcheyan cautioned.

On-chain inference-market players

Here are the on-chain projects Galaxy noted were vying for a slice of the new inference markets:

  • Venice, Erik Voorhees's privacy-focused platform, is the report's centerpiece. It raised $65 million in June, has burned 42% of its VVV token, and sells DIEM, a claim on $1 a day of inference in perpetuity. It's held "largely by speculators," with less than half its inference used weekly.

  • Morpheus tokenizes inference access alongside Venice. It gets one line, no analysis.

  • Pearl is a Bitcoin fork that swaps SHA-256 for matrix multiplication, so mining and inference become the same act. It adds 0.5% to 10% overhead, and its Together.ai endpoint runs 25% below Together's rate. Its launch drew a mining rush "with little sign any of it was serving real inference."

  • Ambient, a Solana fork, "standardizes the entire network on one large open-weight model and builds its consensus around verifying that model's output," as described by Tcheyan.

  • Chutes and AkashML let anyone plug in a GPU and monetize it. Galaxy says the feature is "largely replicable by centralized providers."

  • USD.AI lends stablecoin deposits against neocloud GPUs at 70% to 80% loan-to-value, paying stakers around 8%. "It works precisely because it does not mint a token," the analyst wrote.

(Disclosure noted in the report: Galaxy Digital holds positions in DIEM and VVV, and Ornn is a Galaxy portfolio company.)

(HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: The first draft of this story was written on Claude Desktop using a custom skill and prompt. I then used Distro Publisher, our MCP server, to file a draft directly to DeAI News on DistroVerse. I used the DistroVerse editing interface to make some edits (in this case almost a complete rewrite), and I personally checked all the facts prior to publication.)