BlackRock paper maps AI-agent payments to stablecoins, with Bitcoin as savings layer
“The Machine-Native Economy” argues cards and bank transfers cannot meet machine-speed settlement needs and flags x402 as an early integration surface.
BlackRock has published a research paper arguing autonomous AI agents will need “machine-native” payment rails that can transact instantly without human approvals. The paper proposes a two-tier crypto stack, using stablecoins for spending and settlement and Bitcoin as a store of value, while stressing that live AI-agent payment volume is still small.
Key Takeaways
- BlackRock’s “The Machine-Native Economy” argues cards and bank transfers are a poor fit for autonomous AI agents that need instant, machine-speed payments without human approval.
- The paper separates roles: stablecoins for spending and settlement, and Bitcoin for saving as a store of value between transactions.
- Stablecoin scale is framed as already comparable to major payment networks, with $300B+ circulating supply and “BlackRock-adjusted” volumes of more than $11T in 2025 plus $8.5T in H1 2026.
- Coinbase’s x402, built around HTTP 402 “Payment Required,” is highlighted as a path to per-request agent payments, even as BlackRock notes current agent payment activity remains small and the paper is not a forecast.
BlackRock’s “Machine-Native Economy” Thesis Puts Stablecoins and Bitcoin in Distinct Roles
BlackRock’s research paper “The Machine-Native Economy” puts a specific constraint on the table: payment rails designed for humans do not map cleanly to software acting on its own. The paper’s core claim is that cards and bank transfers cannot serve autonomous AI agents that need to transact instantly, at machine speed, without a person clicking an approval prompt.
Mechanically, the paper is arguing for “machine-native money,” meaning payment instruments that can be held and moved by software with finality and automation as first-class features. BlackRock frames the AI link as a structural driver for digital asset adoption, writing: “Together, these developments position AI as a structural catalyst for digital asset adoption and digital assets as a potential facilitator of the AI economy,” and adding that “this relationship remains underappreciated and could expand the role of digital assets as core infrastructure for an increasingly autonomous digital economy.”
The proposed architecture is deliberately split-brain. Stablecoins, defined here as fiat-pegged crypto tokens used for on-chain settlement, are assigned the spending role. Bitcoin is assigned the savings role, positioned as a store of value an agent can hold between spending events. BlackRock’s own language is broad on the instrument set, stating: “Stablecoins, native cryptoassets, and other on-chain assets can serve as machine-native instruments,” but the paper’s two-tier model is explicit about stablecoins for payments and Bitcoin for saving.
BlackRock also tempers the pitch. It does not call for a crypto-only future for machine money, and it notes that live AI-agent payment volume is still small. The report uses tentative language like “could” and “can,” and includes a disclaimer that it is not a financial forecast.
The Numbers BlackRock Uses to Argue Crypto Rails Are Already at Payment-Network Scale
BlackRock’s adoption case leans heavily on stablecoin scale metrics that are large enough to anchor a mainstream payments comparison. The paper cites stablecoins’ circulating supply as exceeding $300 billion, a level that matters because it implies persistent demand for on-chain dollars as a settlement asset rather than a transient trading tool.
On throughput, BlackRock cites “adjusted transaction volume figures” showing stablecoins processed more than $11 trillion in 2025. The paper frames that as roughly level with Visa’s $11.2 trillion over the same period, while also citing Visa’s total volume of $16.7 trillion. It then adds that stablecoins processed another $8.5 trillion in transaction volume in the first half of 2026.
The catch is in the adjective. The packet describes these as “BlackRock-adjusted” stablecoin volumes, but it does not disclose the methodology, the inclusion criteria, or what is being netted out. Without that, the figures are best read as directional evidence that stablecoin settlement is already large, not as a clean apples-to-apples measure against card-network purchase volume.
That ambiguity matters for traders because the narrative is doing two jobs at once. It is arguing stablecoins are already operating at payment-network scale, and it is using that scale to justify stablecoins as the spending leg of a machine-native stack. If the adjustment method is conservative and replicable, the comparison strengthens. If it is opaque or sensitive to classification choices, the headline numbers can travel further than the underlying measurement.
From AI Model Preferences to x402: The Early Signals BlackRock Points To
To support the idea that machines might naturally separate “spend” from “save,” BlackRock cites a February 2026 study by the Bitcoin Policy Institute as “preliminary support.” The study tested 36 frontier AI models from Anthropic, OpenAI, Google, xAI, and DeepSeek, collecting 9,072 answers in total.
The reported preferences line up with BlackRock’s two-tier framing. In the “store value” choice, models selected Bitcoin 79.1% of the time. In the “spend” choice, models selected stablecoins 53.2% of the time. Traditional bank money was chosen less than 9% of the time across both categories.
As evidence, this is suggestive rather than dispositive. The packet does not include the prompt design, evaluation harness, or whether the models had tool access that could bias them toward on-chain instruments. BlackRock’s own “preliminary” label is doing work here.
Where the paper gets more concrete is infrastructure. BlackRock points to Coinbase’s x402 protocol, which revives HTTP status code 402, “Payment Required,” as a way for software to pay for a resource before it is served. In practice, that is an integration surface for per-request payments, where an AI agent can pay transaction-by-transaction for data or services without a human approving each purchase.
BlackRock is not only theorizing about the category. The packet also notes the firm launched a money market fund earlier in 2026 designed for stablecoin issuers to park reserves, giving BlackRock a direct commercial stake in stablecoin market growth.
Signals to Watch for BlackRock paper pitches AI agent crypto
The first signal is whether x402 moves from a cited concept to a measured payment rail. New integrations, production deployments, or usage metrics tied to HTTP 402 “Payment Required” flows would turn the story from narrative to observable throughput.
The second is whether stablecoin transaction-volume reporting converges on a replicable standard. If other datasets can clarify or reproduce BlackRock’s “adjusted” methodology, the $11 trillion (2025) and $8.5 trillion (H1 2026) figures become more than a persuasive comparison. If the adjustment remains a black box, traders should expect the headline number to be debated whenever it is used as a Visa-scale proxy.
The third is follow-on product or partnership activity from BlackRock tied to stablecoin reserves. The packet’s reference to a money market fund for stablecoin issuers is a reminder that stablecoin growth is not only a thesis for the firm, it is also a distribution opportunity.
The fourth is independent replication of the AI-agent “spend vs save” preference results. Additional third-party studies that corroborate or challenge the February 2026 Bitcoin Policy Institute findings, including the 36-model, 9,072-answer setup and the 79.1% Bitcoin and 53.2% stablecoin splits, would help separate model-training artifacts from durable agent behavior.
Why This Narrative Matters for Traders—and Where It Could Break
The part that matters here is BlackRock separating utility from collateral. Stablecoins are pitched as the spending rail because they settle like dollars, and Bitcoin is pitched as the savings leg because it is framed as value storage outside a single bank balance sheet. That barbell is already how a lot of the market trades the category, and institutional research can make it easier for allocators to talk about stablecoin plumbing and BTC store-of-value in the same breath.
The real test is whether the integration surface becomes real flow. If x402-style per-request payments show measurable uptake and the stablecoin volume methodology becomes legible enough to be repeated, the thesis starts to look structural rather than narrative-driven. If agent payments stay “small” and the adjusted volume figures remain hard to audit, this reads more like an institutional framing exercise than an immediate catalyst for on-chain payment demand.
