Written by: Xiaobing

The world's largest asset management company BlackRock released a research report this week titled "Machine Native Economy: How Digital Assets Connect Intelligence, Commerce, and Computing Power," proposing a succinct and bold equation: AI is "machine-native intelligence," digital assets are "machine-native money," and blockchain is the programmable infrastructure that connects the two.

This report was written by the BlackRock Digital Assets Research team, led by Will Su and Robert Mitchnick, the latter of whom is also in charge of the BlackRock iShares product line.

The core argument of the report is: as more software agents (AI Agents) begin autonomously purchasing data, calling APIs, and renting computing power, they require a payment track that does not depend on human clicks to "confirm." Traditional banking systems and credit card frameworks cannot serve software without identification, while stablecoins and blockchain can neatly fill this gap.

In simpler terms: robots need to spend money, but banks won’t open accounts for robots, so the robots have to use cryptocurrency.

This narrative is not new in the crypto industry, but the endorsement from an institution of BlackRock's scale carries different significance.

The key question is: how far is the path from narrative to revenue? Which chains and tokens are truly on this path, and how many are just bystanders along the way?

The report outlines three layers

BlackRock's pathway consists of three layers: stablecoins as "tools for spending," Bitcoin as "tools for saving," and tokenized computing power as a "new asset class."

Stablecoins: The only layer with real traffic. As of September 2026, the total market capitalization of stablecoins exceeds $300 billion, with adjusted trading volumes exceeding $11 trillion for the entire year of 2025, roughly equivalent to Visa's transaction processing volume during the same period. USDT leads with about $189.5 billion, followed by USDC at about $77.3 billion. This is operational infrastructure, not a dashed line on a roadmap.

Visa's stablecoin settlement project has an annualized processing volume of $7 billion, covering nine chains. Mastercard has also included eight chains, including Solana, Arbitrum, Base, Ethereum, and Polygon, in its stablecoin settlement network. SoFi has just become the first U.S. national bank to launch on the Mastercard stablecoin track, with 70% of its SoFiUSD supply deployed on Solana.

Bitcoin: The narrative is beautiful, but the evidence is weakening. BlackRock cited a study from the Bitcoin Policy Institute earlier this year that tested 36 AI models in simulated scenarios, finding that they chose Bitcoin as a store of value in 79.1% of cases and stablecoins for payment in 53.2%. This data seems very favorable to Bitcoin, but the latest round of testing (early September, with 9 updated models) showed a marked shift: the proportion choosing stablecoins for payment rose to 77%, while those choosing Bitcoin for payment dropped from 36% to 20%. Only Grok 4.6 listed Bitcoin as the preferred payment tool. BlackRock's report did not even mention the Bitcoin Lightning Network or the L402 protocol, opting instead for Coinbase-led x402, Stripe's MPP, Google's AP2, and Visa's TAP.

From another angle: BlackRock assigned Bitcoin the role of a "savings account," but in the actual payment phase of the agent economy, Bitcoin has become marginalized.

Tokenized computing power: The greatest imagination, the furthest from realization. The most eye-catching forward-looking judgment in the report is: "As agents become more persistent and powerful, standardized claims on compute capacity may become an important use case for digital assets in financing and programmable settlement." BlackRock estimates that by 2030, the combined revenue of AWS, Microsoft Azure, and Google Cloud will reach approximately $1.1 trillion. If computing power becomes a tradable standardized commodity like oil or wheat, the market space for financial derivatives will obviously be larger.

But this remains a possibility on paper. No tokenized computing power market currently possesses the level of standardization required for commodity futures.

$5,000: The real monthly volume of agent payments

BlackRock acknowledges in the report that agent payments are still in the early stages, but just how early is this "early"? A study released by blockchain intelligence company TRM Labs on September 9 provided a cool-headed answer.

Since its launch in May 2025, the Coinbase-led x402 protocol has recorded 205 million settlement transactions, totaling approximately $52.7 million. On the surface, this seems like an exciting number. However, TRM Labs applied three layers of filtering: removing self-payments where the payer and seller addresses are the same, filtering out abnormal flows concentrated among a few payers, and excluding sellers with fewer than 10 buyers. After filtering, about half of the transaction volume evaporated, leaving approximately $25.62 million that can be classified as reasonable buy-sell transactions.

Of this $25.62 million, only 0.6% to 7.5% of the transactions seem to come from real AI agents. In absolute numbers, that translates to $154,000 to $1.92 million. In terms of the monthly operating rate for 2026, the actual commercial payments from agents are between $5,000 and $11,000 per month.

Each month $5,000. This is the actual transaction volume from agents produced by a protocol driven by 100,000 registered agents, operating across three chains, and fully supported by Coinbase.

TRM Labs identified its primary signal for recognizing AI activity as transaction amounts: transactions averaging less than $1 are more likely to originate from agent micro-payments. However, the researchers themselves noted that "dedicated agents" designed for single tasks may not fit this model, so real agent activity may be underestimated. Even so, with the gap between thousands and hundreds of thousands of dollars, this disparity is large enough to raise doubts about the existence of the entire market.

USDC accounted for 99.6% of the settlement value on x402. This validates BlackRock's assertion that "stablecoins are the preferred payment tool of agents," but also indicates that x402 is currently more like a pipeline for USDC, rather than a diverse agent economy.

Who is really making money: The truth about revenue at the settlement layer, computing layer, and data layer

BlackRock's report describes a layered system: the underlying settlement network (L1/L2) charges gas fees per transaction; the middle layer of computing and data markets earns commissions from supply-demand matching; and the upper layer of stablecoin issuers profits from interest on reserve assets. This framework is correct, but the scale of revenue at each layer varies greatly.

Settlement layer: Base reaps the biggest early benefits. 67% of x402's transaction volume runs on Coinbase's L2 network, Base, while the rest is distributed between Solana and Polygon. However, the gas revenue from agent payments for these chains is negligible. Based on an average transaction fee of $0.30 and a gas fee of approximately $0.001, the total on-chain fees from $52.7 million would be only several tens of thousands of dollars. A meaningful signal for L1/L2 is yet to be seen until the scale reaches several billion transactions per month. Currently, the money these chains earn from regular stablecoin transfers and DeFi activities far exceeds that from agent payments.

Solana is the most active chain in stablecoin settlements. In April 2026, it accounted for 32.6% of the adjusted weekly stablecoin transfer volume, surpassing Ethereum's 27.8%. In February alone, the stablecoin transaction volume on Solana reached $650 billion. But this volume mainly comes from DeFi and cross-border payments, not from the agent economy. Counting Solana’s stablecoin processing volume as agent payment revenue is a logical leap.

Computing layer: Revenue is growing, but attribution is vague. The decentralized computing market is the most direct reflection of BlackRock's "tokenized computing power" narrative. The real revenue data of leading projects is as follows:

Aethir is currently the highest-earning DePIN computing project, generating $127.8 million in total revenue for the year 2025, with an annual recurring revenue reaching $166 million in Q3, covering 94 countries and over 440K GPU containers. This revenue comes from AI training, inference, and cloud gaming needs from over 150 enterprise clients.

Akash's data tells a more complex story. Akash officially reported that Q1 2026 marked its first full period of computing expenditure exceeding $5 million; however, Messari’s independent tracking shows that the on-chain rental revenue in the same quarter was only $253,000, down 45% quarter-over-quarter, and annualized at about $1 million. The difference between these two figures is roughly fivefold, which may stem from different accounting standards for AkashML inference business revenue. The inference throughput of AkashML indeed grew from about 5 billion tokens per day in May to over 10 billion tokens per day in July, but the AKT token (approximately $0.51, down about 94% from its 2021 peak) did not capture this growth.

io.net reported approximately $20 million in annualized on-chain revenue, with over 10,000 active nodes and 56 enterprise clients, about 72% cheaper than AWS.

Render Network accumulated over 77.5 million frames of rendering, 5,600 nodes, and a 279% year-on-year increase in token burn, but the revenue data at the protocol level has not been disclosed publicly.

In summary, the DePIN computing segment tracked by DeFiLlama had an annualized revenue of about $180 million to $220 million in Q1 2026, with Aethir accounting for the bulk. However, the total circulating market value of around $10 billion in the entire DePIN ecosystem generated only $72 million in on-chain revenue in the fiscal year 2025. This is not an undervalued track; it is a track with a huge gap between revenue and valuation.

The more critical question is: the revenue growth of the computing market is currently driven by the natural spillover of enterprise AI demand (cloud vendors' capacity shortages, leading clients to turn to decentralized alternatives), which is almost unrelated to agent payments. The scenario where agents autonomously procure computing power and settle with stablecoins, as described by BlackRock, has not yet occurred on a scalable level on any of these platforms.

Data layer: The most blank segment. BlackRock's report lists "data purchasing" as one of the core activities of the agent economy, but there are virtually no worthwhile revenue cases in the on-chain data marketplace. CoinGecko's paid API is one of the early deployments of x402, but the scale of such services is too small to support an independent asset class narrative.

The competitive landscape of agent payment protocols: Crypto is not the only player

The five agent payment protocols listed in BlackRock's report reflect a fact that the crypto industry selectively overlooks: the competition among agent payment protocols is far more complex than merely "on-chain vs off-chain."

x402 is a stablecoin payment protocol built by Coinbase around the HTTP 402 status code, where the agent initiates the request, and the server returns the price and address, allowing the agent to settle on-chain with USDC without requiring an account. This is the crypto-native path.

MPP (Machine Payments Protocol) was jointly developed by Stripe and Tempo, launched in March 2026, characterized by its support for both stablecoin and fiat currency settlements. Anthropic, OpenAI, Shopify, Etsy, Visa, and Mastercard are all signatories. Stripe's strategy is clear: to lock in the standards with its distribution advantage before x402 seizes the high ground of standards.

ACP (Agentic Commerce Protocol) was jointly developed by OpenAI and Stripe and standardized the checkout process within the AI surface. AP2 is an authorization layer protocol launched by Google, addressing the question of "who authorized this transaction." TAP is Visa's trusted agent protocol, addressing the identity verification issue of "is this agent a legitimate buyer."

These protocols can be combined to form a layered stack: an agent purchase may simultaneously use AP2 for authorization, ACP for checkout, and x402 or MPP for settlement. Crypto payments (x402) occupy only part of the settlement layer in this stack. If Stripe’s MPP allows agents to complete the same micro-payments in fiat currency, then the exclusivity of on-chain settlements disappears.

For chains and tokens, competitors are not just other chains, but also the entire traditional payment system's agent-based upgrade.

Where is the real alpha?

Returning to the initial question: what chains and tokens can agent payments bring revenue to?

If we define "bringing revenue" as "already bringing verifiable revenue from agent activities today," the answer is sobering: almost none. The real agent payment flow of $5,000 to $11,000 per month, dispersed across Base, Solana, and Polygon, has no impact on the revenue reports of any chain.

If we extend the timeframe to include “potential beneficiaries in the next 3-5 years,” the order of beneficiaries is roughly:

Stablecoin issuers (Circle, Tether) are the most certain winners. Regardless of which chain agents trade on or which protocol they use, USDC and USDT are the default payment mediums. However, USDC and USDT themselves are not investable tokens; Circle’s revenue comes from reserve interest, not from on-chain fees.

Low-fee settlement chains (Solana, Base) have a structural advantage in competition. The typical transaction amount for agents is below $1, and a gas fee of $0.001 is the entry barrier. The main Ethereum network is naturally excluded from micro-payment scenarios. However, the agent payment revenue for Solana and Base now rounds to zero.

Decentralized computing networks (Aethir, Akash, Render, io.net) have real revenue, but this revenue comes from enterprise AI demand, not from agent autonomous procurement. The "tokenization of computing power" scenario described by BlackRock, whereby agents buy and sell standardized computing power like trading futures, does not exist on any platform currently. The AKT token price has dropped by 94% from its peak, and the market has not assigned a valuation premium for Akash's growth in usage.

The value capture mechanisms for protocol-specific tokens are generally weak. Most DePIN project tokens do not have direct distribution rights to revenue, the amount of tokens burned is negligible relative to market capitalization, and token inflation far exceeds consumption demand. Render's token burn increased by 279% year-on-year, but by the end of 2025, the cumulative amount burned was only about 0.2% of market value.