Digital Assets

Blockchain and AI: where the two technologies actually meet

Artificial intelligence and blockchain are routinely mentioned in the same breath, and just as routinely conflated. This note separates the places where the two technologies genuinely solve problems for each other from the places where the connection is mostly narrative — and sets out how we think about the difference.

Manuel E. De Luque MuntanerCEO & Founder

31 July 20269 min read

Key takeaways

The convergence is real in a small number of specific places — provenance and verification, machine-native payments, and markets for compute and data — and thin almost everywhere else.
Generative models make digital content cheap to fabricate, which raises the value of cryptographic proof of origin and integrity. This is the most durable link between the two fields.
Autonomous software agents need to transact without a human in the loop, and programmable, always-on settlement rails are a natural fit — though the regulatory and operational questions are genuinely unresolved.
A large share of what is marketed as “AI plus blockchain” is branding attached to a token rather than infrastructure solving a coordination problem; the two deserve to be separated before any capital is committed.
For allocators, thematic exposure of this kind has tended to behave as an expression of risk appetite rather than as diversification — which is why our own strategies stay in the most liquid parts of the market and use machine learning inside the research process rather than as a theme to buy.

Two technologies with opposite strengths

It is worth being precise about what each technology is good at, because the complementarity is not where most commentary places it.

Modern AI is probabilistic, data-hungry and, in practice, highly centralised: the frontier is set by a small number of organisations with access to very large compute clusters. It produces output that is fluent, cheap and — crucially — difficult to attribute or verify after the fact.

A blockchain is close to the opposite. It is deterministic, replicated and expensive per unit of computation. What it is genuinely good at is producing a shared, tamper-evident record that multiple parties who do not trust each other can agree on, without appointing one of them as the referee.

Stated that way, the useful connection becomes clearer. It is not that blockchains will run AI models — for the most part they cannot, economically. It is that one technology is increasingly good at generating content and actions at scale, and the other is a machine for establishing what is true about a record.

Where the connection is substantive

A handful of areas stand up to scrutiny. What they share is that each involves a party needing to trust something it cannot directly observe — which is the problem a tamper-evident record is built to address.

Provenance and attestation — as synthetic media becomes indistinguishable from captured media, the question shifts from “does this look real?” to “can this be traced to a source that stands behind it?” Cryptographic signing, content hashes and tamper-evident logs are a credible answer, and the emerging provenance standards in the media and device industries lean on exactly these primitives.
Machine-native settlement — software agents that book, purchase or subscribe on a user's behalf need to pay for things without a human approving each transaction. Programmable, always-on, low-minimum settlement is a better fit for that pattern than card rails designed around human authorisation, with significant unresolved questions about authority, liability and controls.
Markets for compute and data — coordinating supply of a scarce resource across many independent providers is a real coordination problem, and token-based incentives are one way to bootstrap such a market. Whether these markets can compete on cost, latency and reliability with large cloud providers is a separate and still open question.
Identity and proof of personhood — if the marginal cost of generating plausible human behaviour falls to near zero, systems that need to know they are dealing with a distinct person acquire a hard problem. Verifiable credentials that prove an attribute without disclosing the underlying data are one of the more serious lines of work here.
Verifiable computation — proving that a specific model produced a specific output, without revealing the model, is conceptually powerful for auditability. It remains early, and the computational overhead is currently substantial relative to the benefit in most commercial settings.

Where it is mostly narrative

A balanced reading has to give as much weight to the weak claims as to the strong ones. Several of the most widely repeated connections do not survive much examination.

Branding attached to a token — a great many projects have added an AI descriptor without a corresponding change in what the system does. The test is whether the technology requires decentralisation to work, or whether an ordinary database and a company would do the job better.
On-chain training and inference — running meaningful models directly on a public ledger is not economically sensible today, and the gap is one of orders of magnitude rather than a matter of tuning.
“AI-managed” strategies as a marketing claim — the phrase is used to describe everything from a rules-based script to genuine statistical research. It carries no information on its own; what matters is the process, the controls and the evidence behind it.
Diversification that isn't — thematic baskets built around a narrative have tended to move together, and with the broader risk-appetite cycle. Exposure of this kind should be understood as concentrated and directional, not as a hedge.

The other direction: AI applied to blockchain data

The less discussed half of the relationship is the more immediately practical one. Public ledgers are an unusually good dataset: complete, timestamped, machine-readable and available to everyone at the same time. Very few markets offer anything comparable.

That makes them well suited to statistical and machine-learning techniques — for monitoring flows and concentration, flagging anomalous behaviour, understanding market microstructure and liquidity, and supporting counterparty and operational due diligence. This is where the two fields already meet in day-to-day institutional practice, quietly and without a token attached.

What this means for allocators

The sensible posture is neither dismissal nor enthusiasm. Some of what is being built here addresses real problems that are about to get harder, and provenance in particular looks structural rather than cyclical. Much of the rest is a narrative in search of a use case.

The practical discipline is to separate the two questions that usually get merged: is this technology likely to matter, and is this instrument a sound way to express that view? A theme can be entirely correct and still be a poor investment if the exposure is illiquid, concentrated, or priced for an outcome that has not happened. Our own approach is to keep exposure in the most liquid, best-understood parts of the market and to apply these techniques inside our research process, where they are testable — rather than treating them as a story to buy.

Block Asset Management

How Block Asset Management helps

Technology narratives arrive in this asset class faster than evidence does. Our role is to hold the two apart.

Evidence before narrative

We assess technology claims on what a system demonstrably does, not on how it is described. A compelling story is not an input to an allocation decision.

Liquidity-first exposure

Our systematic strategies concentrate on the most liquid segments of the digital asset market rather than thematic baskets, so positions can be sized, monitored and exited with discipline.

Machine learning inside the process

We use statistical and machine-learning methods in research and risk monitoring, subject to validation and human oversight — an approach we set out in more detail in our note on AI and machine learning in quantitative research.

Due diligence that tests the claim

When we assess external managers, a stated technology edge is something to be evidenced — process, data, controls and people — not accepted at face value.

Governed access

Professional and qualified investors get exposure through defined risk limits, continuous monitoring and a controlled, auditable process.

The connection between blockchain and artificial intelligence is real, but it is narrower and more specific than the volume of commentary suggests. It lies in verification, settlement and coordination — the places where output needs to be trusted, machines need to transact, or independent parties need to agree on a record without a referee.

If your organisation is weighing how these developments bear on a digital asset allocation, our investor relations team would be glad to discuss our approach. Professional and qualified investors can also register for access to our detailed strategy materials.

Important information

This material is provided for information purposes only and is intended for professional and qualified investors. It is commentary on technology and market developments and is not legal, tax or investment advice, nor an offer, solicitation or recommendation of any strategy or financial instrument. References to technologies or market segments are illustrative and do not constitute a view on any specific asset. Digital assets are volatile and involve significant risk, including the possible loss of the entire amount invested. Past performance is not a reliable indicator of future results.

Subscribe to our investor updates

Request Information Register for access