Author: BlackRock
Compiled by: Golden Finance
Note: On September 23, BlackRock released the research paper "Machine Native Economy: How Digital Assets Connect Intelligence, Business and Computing Power", which explores the growing connection between AI and digital assets, and explains why the widespread application of AI may drive new needs, new uses and new applications in the digital asset economy. The following is the full text:
summary
The extraordinary growth of AI is the most iconic technological theme of this era. At the same time, the rise of digital assets is a parallel technological theme with particularly important implications for financial infrastructure. Historically, these two themes have generally developed on parallel trajectories, but as AI systems increase their ability to interact with financial and economic networks, they are beginning to converge. This article examines this growing relationship and explains why widespread adoption of AI could become an underestimated source of demand, utility and application growth in the digital asset economy.
At the core of this convergence, both AI and digital assets originate from a common foundation: AI stands for machine-native intelligence, and digital assets stand for machine-native currency. With the rise of agentic AI, this fit is particularly critical—agent AI refers to a system that can interact with external tools and infrastructure with limited human intervention, plan and execute multi-step tasks around a set goal; while blockchain provides programmable infrastructure to connect intelligence and economic activities. These capabilities advance AI from content generation to real-world actions, including completing purchases and initiating financial transactions.
Specifically, we explore three key areas of overlap:
The large language model has a similar Tokenization architecture to the blockchain. The large language model divides human language into tokens and numerically encodes them for model interpretation and processing. Blockchain also represents economic value and equity as digital asset tokens, and its design goal is transfer and settlement that can be verified by machines. The functionality of each differs, but both processes convert real-world input into a format that the machine can natively use.
Agency-based commerce requires machine-native payment channels. The rise of agent-based AI and machine-to-machine payments may increase the need for blockchain and other programmable payment infrastructure; stablecoins, native cryptoassets, and other on-chain assets can serve as machine-native tools for payments and settlements on these channels. Existing channels such as ACH and card organization networks already support a considerable degree of automation, but their account opening requirements and settlement economics may make them less suitable for 24/7, very small-amount transactions that require programmable execution. Emerging protocols such as x402 and ACP are being deployed on blockchain networks and advancing in parallel with the adaptation of traditional payment channels to build transaction and settlement layers for more complex agent-based workflows.
Computing power is becoming an emerging and potentially huge market for digital assets. Computing power, the processing power needed to train and run AI systems, is becoming an increasingly important economic resource. Analysts estimate that annualized cloud revenue for hyperscalers could exceed $1 trillion by 2030. As agency capabilities become stronger and last longer, standardized claims to computing power may become an important use case for digital assets in financing and programmable settlement.
Taken together, these developments make AI a structural catalyst for the adoption of digital assets and make digital assets a potential enabler of the AI economy: AI interprets information and guides action, and blockchain provides machine-readable assets and programmable settlement. This relationship remains undervalued and could expand the role of digital assets as the core infrastructure of an increasingly autonomous digital economy.
AI and the tokenization of digital assets transform real-world inputs into machine-native representations
At the architectural level, AI plays a similar role to tokenization in digital assets: converting information and economic rights into discrete, standardized representations that can be processed natively by machines. AI Tokens encode information, while digital asset tokens represent units of value, ownership, or economic interest. In a large language model (LLM), a tokenizer breaks human-readable text into smaller units—usually words, subwords, or characters—and maps them into numerical identifiers. The embedding layer then converts these identifiers into vector representations that the model can use for computation. The model generates numerical token IDs that are decoded back into human-readable text. By continuously converting input text into standardized numerical units, LLM is able to efficiently process large-scale input streams using the highly parallel computing that modern chips excel at.
Blockchain employs functionally comparable, but technically distinct processes for value storage and economic entitlement. Digital asset tokenization is the process of representing financial assets or real-world assets (RWA) such as cash, fund shares, securities or other ownership claims as standardized digital tokens recorded on a distributed ledger. Existing on-chain assets such as stablecoins can be transferred through smart contract transactions. At the transaction level, these assets are expressed through machine-readable transaction data and governance rules. The network verifies that authorization and transactions are legal and valid, and smart contracts or transaction scripts impose permissions and conditions unique to the asset. Together, these mechanisms update the ledger by transferring token balances. Once a transaction is recorded and achieves full finality under the network's consensus rules, it becomes part of the network's canonical ledger state.
In a blockchain workflow, smart contracts execute transaction logic and impose specific rules for financial assets. Anti-money laundering (AML), know-your-customer (KYC) and know-your-agent (KYA) checks are typically performed off-chain, as identity and compliance data can be evaluated off-chain and the verification results passed on-chain to determine transaction eligibility. For tokenized RWA, these controls sit within a broader legal and regulated services framework that relies on authoritative off-chain registration systems while reducing reliance on closed databases and manual reconciliation.
Figure 1: Schematic of the Tokenization process in large language model and blockchain transactions
These parallel workflows—one encoding human context, the other encoding economic equity—will become more important as intelligent systems continue to merge with transaction execution. Both systems use structured, machine-readable representations, which enables LLM-based AI agents to interface with blockchain data more directly than they are faced with many fragmented traditional systems. In turn, greater standardization across asset classes reduces reliance on customized integrations and makes it easier for agents to orchestrate more complex multi-asset workflows. Through programmable interfaces, agents can evaluate balances and rules before executing authorized transactions and verifying settlements, with limited reliance on manual processes.
Recent research from the Bitcoin Policy Institute provides preliminary support for this framework: it reported that in controlled simulations, model outputs generally favored stablecoins for daily payments and Bitcoin for long-term storage of value. These findings reflect simulated model responses rather than actual observed agent behavior, but point to a potential AI-native currency architecture—with stablecoins acting as transaction currencies and Bitcoin acting as a store of value. As explored in the next section, this shared machine-native foundation may enable meaningful proxy payment use cases where some traditional financial channels may be less efficient or economical.
Machine-native transactions require purpose-built proxy payment protocols
As AI agents increase their capabilities and expand their real-world applications, they increasingly require payments and asset infrastructure designed natively for machine-speed commerce. Crypto-native blockchain channels are particularly suitable for high-frequency, sub-1-cent machine-to-machine (M2M) transactions that run 24/7, including API calls, on-demand data, and pay-as-you-go computing power. At the same time, reinvented traditional payment systems will remain important in connecting agents with businesses and consumers run by humans (i.e. business-to-machine B2M, consumer-to-machine C2M).
Stablecoins, native cryptoassets, tokenized real-world assets, and other programmable tools can support transactions and digital ownership (including use as collateral) with the required granularity on a 24/7 basis. More broadly, tokenization can provide standardized, machine-readable representations across asset classes, reduce customized integrations in financial infrastructure, and enable agents to orchestrate increasingly complex workflows more efficiently.
Many existing payment channels are not suitable for agency-based transactions with high transaction volumes and low single amounts. Reasons include:
The account opening, credential issuance and authorization processes may require manual participation;
Merchant processing fees may make transactions of very small amounts uneconomical;
Settlement and Finality Limitations: Most ACH transaction volume settles in one business day or less, card authorization is nearly instantaneous, but merchant settlement and dispute finalization may take longer; and
As the volume of transactions generated by the machine grows, there may be scalability limitations.
The Agent Payments Protocol is built on foundational standards such as MCP and A2A that connect agents to external systems and each other. MCP (Model Context Protocol), launched by Anthropic in November 2024, standardizes how AI applications access external data and workflows. A2A (Agent2Agent), released by Google in April 2025, enables agents to communicate and collaborate across platforms.
Agents using MCP and A2A will use multiple agent payment protocols to complete complex workflows involving payments. x402 is an open payment protocol developed by Coinbase that uses the HTTP 402 "Payment Required" status code to support machine-initiated payments. The protocol is blockchain-agnostic, with stablecoins such as USDC being one of its key early use cases and emerging as a potential standard for high-speed M2M transactions. By providing 24×7, near real-time, verifiable settlement, x402 reduces resource providers’ counterparty risk exposure and delivers requested data or services immediately upon payment confirmation. Because x402 uses digital currencies, including stablecoins held in on-chain wallets, it is able to support high-frequency, small-value transactions without the need for human intervention. When settlement occurs on a permissionless network, higher usage could increase demand for block space and validator services, creating a potential conduit to native crypto assets. The extent of value capture will depend on each network’s fee, staking and gas payment design.
Other emerging standards connect agency-based transactions to existing payment channels and build guardrails for trusted financial execution. The Machine Payment Protocol (MPP), developed by Stripe and Tempo, supports payment for APIs and other HTTP resources, with flexible settlement via stablecoins or traditional payment methods. The Agent Commerce Protocol (ACP), developed by Stripe and OpenAI, enables programmatic checkouts between agents and businesses while allowing sellers to retain their existing commerce and payments infrastructure. Google's Agent Payment Protocol (AP2) uses encrypted authorization credentials and an audit trail to provide evidence of user authorization, while Visa's Trusted Agent Protocol (TAP) helps merchants verify trusted agents and securely receive payment credentials.
Figure 2 shows a simplified agent payment workflow. The human user (1) asks the agent to book flights and hotels within a specified budget. The master AI agent (2) accesses the user's calendar, preferences, and authorized payment information via the MCP connector and then (3) delegates data collection via A2A to a dedicated AI travel subagent. The sub-agent (4) calls the paid air ticket and room price API, and the relevant payment is processed through x402 and settled on the chain. Using the returned information, the host agent (5) completes the reservation through the airline and hotel checkout systems and (6) returns the itinerary and receipt to the user.
Figure 2: Schematic representation of agency-based workflow involving basic protocols and financial protocols
A variety of digital assets may support business intelligence, but stablecoins are likely to lead the way for transactional uses. Stablecoins are digital tokens designed to maintain a stable value relative to a reference currency, most commonly the U.S. dollar. Its price stability provides a reliable measure of value and brings greater predictability in pricing and settlement. Stablecoins are the largest category of tokenized real-world assets, with a circulating market capitalization of over $300 billion as of September 2026. Adjusted stablecoin transaction volume will exceed $11 trillion in 2025, roughly in the same range as Visa and Mastercard’s annual payment volumes. Adjusted stablecoin trading volume is still well below the $93 trillion that will be moved via ACH in 2025; however, its compound annual growth rate (CAGR) from 2020 to 2025 is 80%, compared with about 8.5% for ACH. Increasing clarity on the regulatory environment—including the GENIUS Act in the United States, the EU’s MiCA, Hong Kong’s stablecoin licensing regime, and Singapore’s stablecoin regulatory framework—should support continued adoption and growth of stablecoins.
Figure 3: Transaction volume of stablecoins and major card organization networks
For digital assets, the impact of this growth extends beyond the stablecoins themselves to the blockchain networks on which they are issued and settled. Many major stablecoins are issued on multiple blockchains, allowing market participants to choose between supported settlement venues based on economics and technical fit. These venues range from general-purpose permissionless networks like Ethereum to purpose-built stablecoin networks like Circle’s Arc, where USDC is designed to be a native gas asset. In a permissionless network, native cryptoassets such as ETH underpin consensus, validator rewards, transaction fees and settlement. As stablecoin activity scales, higher demand for block space and network services may support usage-related demand and potential value capture for these assets, depending on each network's fee, staking and gas payment design. Arc also offers another complementary model: higher payments activity could deepen USDC’s utility as both a settlement asset and a means of paying transaction fees.
As autonomous agents proliferate, computing power is becoming an emerging and potentially massive market for digital assets
The operation of AI systems and agents requires a lot of computing power and energy. Investors are primarily focused on the scale of capital expenditures (CapEx) required to build AI infrastructure, with some estimates putting cumulative AI CapEx at more than $5 trillion between 2025 and 2030. But the rise in ongoing operating expenses (OpEx) supporting AI deployments is also worthy of attention. A significant portion of this operating expenditure goes to the AI cloud computing power market, which monetizes the rights to use deployed IT equipment and the power required to operate it. As this market expands, computing power is becoming a unique, sizable, and increasingly investable economic resource that could underpin a new class of digital assets. Taking the main cloud business segments of hyperscale cloud service providers as a rough proxy for market size, consensus market expectations show that the combined revenue of Amazon Cloud Technology (AWS), Microsoft's Intelligent Cloud Division and Google Cloud will be approximately US$1.1 trillion by 2030, equivalent to a compound annual growth rate of 29% from 2025 levels.
As computing power becomes a larger economic input, the need to price and allocate capacity and support financing and hedging should also increase. Historically, large resource markets have developed trading infrastructure to improve liquidity and risk management. We believe computing power may follow a similar path as AI adoption scales. Recent market developments have pointed in this direction, including GPU-backed financing and financing platforms designed around long-term, usage-linked computing revenue. These structures reflect the capital intensity of acquiring leading GPUs and investing in the capacity needed to build leading-edge models. As models improve, demands on both training and inference should continue to grow. As real-world AI use cases expand, inference is expected to become the largest AI workload by 2030 and account for a growing share of data center power needs. The potential user base for inference includes both enterprises and individual consumers, and is much larger and more dispersed than the concentrated group of training market participants.
Figure 4: Global data center power demand by workload
Even as the market is taking shape, there are still important contract design and market structure challenges before standardized computing power products can be scaled up. These challenges include accounting for material differences in chip generation productivity and regional economics (especially where energy costs vary significantly), and establishing workable standards for cash settlements and delivery of contracted capacity. We believe these are important but ultimately solvable design considerations. Basis markets, contracts for difference and other mechanisms used in mature commodity markets provide useful reference for managing heterogeneous assets and localized pricing; while on-chain tokenized markets may enable finer-grained regional and hardware-specific contracts within a shared settlement infrastructure. As these frameworks mature, we expect standardized products—including exchange-traded hashrate futures—to support more transparent price discovery and lead to more effective hedging between providers and users of hashrate capacity. Standardized computing power contracts can also create claims to computing power capacity and related usage rights, which can be represented, transferred, pledged as collateral, and settled through programmable infrastructure. This in turn could expand the participation of institutional investors and make computing power a new opportunity for the broader digital asset ecosystem.
This capability may become especially valuable as agent-based AI adoption scales, with agents increasingly discovering, provisioning, optimizing, and paying for computing power through programmable payment channels like x402. Agents can query the real-time marketplace API to compare available capacity by price, performance, latency, location and hardware expertise, then provision the resources best suited for specific workloads. MCP and A2A can facilitate collaboration between data access and agents, while x402 can support on-demand billing by usage, by model token, or by job. This framework can achieve flexible and just-in-time computing power acquisition with limited manual intervention. Although agent payment activity is still in its early stages, the structural fit between autonomous agents and machine-native payments makes it an area worth paying attention to as the ecosystem develops.
Figure 5: Schematic of the agent-based workflow for optimizing and obtaining on-demand computing power
Stripe's agreement to acquire OpenRouter in August 2026 provided an early strategic signal: model routing and computing power usage optimization are becoming part of the financial infrastructure around AI. OpenRouter distributes workloads to more than 400 models from more than 80 providers based on workload needs and cost-performance tradeoffs, highlighting the economic value of efficiently allocating scarce computing power. This argument is also relevant to the buy-side: given Stripe’s extensive activities in payments, stablecoins, billing, and agency-based commerce, the deal points to a potential convergence between hash power procurement, pay-per-use billing, and programmable settlement. In our view, this may support a future in which agents independently obtain computing power and pay for it through blockchain and other programmable payment channels.
in conclusion
As machines play a larger role in economic activity, AI and blockchain-based digital assets are increasingly converging.
The structured machine-readable representation created through LLM and blockchain tokenization can give AI agents a more direct interface with programmable assets; while protocols such as stablecoins and x402 can support high-frequency, small-amount, and around-the-clock transactions.
At the same time, a standardized and liquid computing power equity market allows agents to acquire, optimize, finance and pay for computing resources as their reasoning needs expand. This ecosystem is still in its early stages, and agency payment activities and computing power market liquidity are still limited.
As AI adoption expands and agent system capabilities increase, digital assets may increasingly become an integral part of the AI economic infrastructure, increasing the overall utility of stablecoins, tokenized RWA, and native cryptoassets that underpin blockchain settlement.