BlackRock published a 30-page research paper on September 22, 2026, titled "The Machine-Native Economy: How digital assets connect intelligence, commerce, and compute." Authored by Will Su, Robert Mitchnick, Jay Jacobs, and William Helm, the paper argues that autonomous AI agents will need progra...
"AI represents machine-native intelligence, while digital assets represent machine-native money." — BlackRock Digital Assets Research, The Machine-Native Economy (September 22, 2026)
BlackRock published a 30-page research paper on September 22, 2026, titled "The Machine-Native Economy: How digital assets connect intelligence, commerce, and compute." Authored by Will Su, Robert Mitchnick, Jay Jacobs, and William Helm, the paper argues that autonomous AI agents will need programmable, 24/7 payment rails — and that stablecoins, not bank wires or card networks, are the natural fit.
The thesis rests on three layers: stablecoins as a spending tool, Bitcoin as a savings instrument, and tokenized compute as a new asset class. BlackRock projects combined cloud revenue from Amazon, Microsoft, and Google reaching $1.1 trillion by 2030, with Goldman Sachs estimating over $5 trillion in cumulative AI capital spending between 2025 and 2030. The paper envisions exchange-traded compute futures emerging alongside this buildout.
The data tells a more cautious story than the headline suggests. TRM Labs analysis of $52.7 million in x402 protocol transactions found that genuine AI agent commercial payments amount to roughly $5,000 to $11,000 per month — a figure that sits several orders of magnitude below the scale BlackRock describes.
BlackRock structures its thesis around three categories of digital asset utility:
Layer 1 — Stablecoins as a spending tool. The paper positions dollar-denominated stablecoins as the default settlement medium for agent-to-agent transactions. Sub-cent fees, programmable authorization, and 24/7 availability are cited as structural advantages over ACH and card networks, which require human confirmation steps and operate within business-hour constraints.
Layer 2 — Bitcoin as a savings tool. Citing a Bitcoin Policy Institute study that tested 36 frontier AI models across 9,072 decision scenarios, BlackRock notes that Bitcoin was selected in 79.1% of store-of-value scenarios, while stablecoins dominated payment scenarios. The paper treats Bitcoin and stablecoins as complementary rather than competing instruments.
Layer 3 — Tokenized compute as a new asset class. This is the paper's most speculative claim. BlackRock writes: "As agents become more capable and persistent, standardized claims on compute capacity could become a significant digital asset use case for financing and programmable settlement." The firm expects exchange-traded compute futures to emerge, enabling transparent price discovery and hedging for both providers and consumers of GPU capacity.
BlackRock's own figures provide the quantitative foundation for its thesis:
| Metric | Value | Period | |--------|-------|--------| | Stablecoin market cap | $300+ billion | September 2026 | | Adjusted transaction volume | $11.2 trillion | Full year 2025 | | Compound annual growth rate | ~80% | 2020–2025 | | ACH network volume | $93 trillion (35.2B txns) | 2025 | | ACH growth rate | ~7.9% annually | Recent trend | | Visa annual volume | $16.7 trillion | 2025 | | Mastercard annual volume | $10.6 trillion | 2025 |
The growth differential is stark. Stablecoins compounded at roughly 80% annually since 2020; ACH grew at 7.9%. At $11.2 trillion in adjusted volume, stablecoins now settle more than Mastercard on an annual basis, according to BlackRock's cited figures. However, adjusted volume figures vary by methodology, and a significant portion of stablecoin volume consists of arbitrage, treasury management, and automated market making rather than end-user commerce.
The paper identifies "agentic AI" — AI systems that plan and execute multi-step tasks autonomously, calling external APIs and services with minimal human oversight — as the primary demand driver. Agents that book flights, purchase data feeds, rent GPU time, or settle invoices need a payment layer that does not require a human to click "confirm" at each step.
BlackRock cites three emerging payment protocols as candidates:
The paper notes these protocols are more complementary than competitive. A production system might use AP2 for spending governance and x402 for settlement, or AP2 for authorization and TAP for settlement through existing Visa relationships.
The gap between BlackRock's thesis and current on-chain reality is substantial. TRM Labs, a blockchain analytics firm, published a detailed analysis of x402 transaction data in September 2026.
Raw figures: x402 processed $52.7 million in total transactions since launching in May 2025. The protocol had 100,000+ registered agents, ran across three chains (Base, Solana, Polygon), and carried Coinbase's full institutional backing.
After filtering: TRM Labs applied three layers of analysis:
Approximately half the transaction volume vanished. Of the remaining $25.62 million, TRM Labs estimated that 0.6% to 7.5% originated from actual AI agents — equating to $154,000 to $1.92 million in cumulative agent payments, or roughly $5,000 to $11,000 per month.
USDC accounted for 99.6% of x402 settlement value. The protocol functions, in TRM Labs' assessment, more as a USDC pipeline than a diverse agent economy.
One analyst quoted by FinTech Weekly described the x402 data as follows: "The infrastructure is ahead of the demand it was built for."
The x402 protocol's daily volume in March 2026 ran at approximately $28,000 across 131,000 transactions, averaging roughly 20 cents per transaction. Artemis, a separate analytics firm, estimated that roughly half of observed x402 activity consisted of self-dealing or wash trading.
Within 90 days of each other in early 2026, every major payment platform launched an AI agent payment protocol:
| Protocol | Operator | Function | Chain/Rail | |----------|----------|----------|------------| | x402 | Coinbase | Stablecoin settlement at API endpoints | Base, Solana, Polygon | | AP2 | Google | Authorization/governance layer | Rail-agnostic | | TAP | Visa | Agent payments via card network | Traditional Visa rails | | Agent Ready | PayPal | Agent commerce enablement | PayPal/Venmo rails |
Google and Coinbase jointly launched the A2A x402 extension, establishing AP2 as a trust layer and x402 as a settlement layer operating in tandem. This collaboration suggests the market may converge on layered architectures rather than winner-take-all outcomes.
As of April 2026, x402 reported 69,000 active agents, 165 million transactions, and $50 million in cumulative volume. CoinGecko's x402-enabled API endpoint charges $0.01 per request, illustrating the micropayment use case the protocol targets.
Coinbase extended its agent infrastructure further on September 22, 2026, opening access to 6,000+ U.S. stocks and ETFs through Coinbase for Agents, enabling autonomous software to handle research, payments, and trade execution within a single account.
The paper's most forward-looking section proposes tokenized claims on computing power as a tradeable asset class. BlackRock's logic:
BlackRock writes: "We expect standardized products, including exchange-traded compute futures, to support more transparent price discovery and more effective hedging for both providers and consumers of compute capacity."
No such product exists today. The paper describes compute-market liquidity as "limited" and acknowledges the concept remains speculative.
The paper's stablecoin-centric vision faces regulatory constraints that it does not fully address:
MiCA in Europe: Euro-denominated stablecoins cannot pay interest under MiCA rules. The ECB discourages reliance on bank deposits backing stablecoins, and its Pontes platform for central-bank-money settlement of tokenized assets offers a competing architecture.
U.S. regulatory uncertainty: The GENIUS Act, which would establish a federal stablecoin framework, has not been finalized. The CLARITY Act, which would have clarified token classification, failed in a 49-50 Senate vote in September 2026. Agent-initiated transactions introduce novel questions about liability, consumer protection, and money transmission licensing.
Agent identity: No existing regulatory framework assigns financial identity to autonomous software agents. Questions about who bears liability when an agent executes an unauthorized transaction remain unresolved.
BlackRock's "Machine-Native Economy" paper presents a coherent long-term thesis: as AI agents proliferate, they will need programmable money, and stablecoins are the most plausible candidate. The firm's three-layer framework — spend (stablecoins), save (Bitcoin), invest (tokenized compute) — provides institutional investors with a mental model for positioning.
The present-tense data, however, does not support the thesis at scale. Genuine agent payment flows measured by TRM Labs are negligible. The infrastructure — x402, AP2, TAP — is built. The agents, at commercially meaningful volumes, have not arrived.
The paper itself acknowledges as much, describing agentic payment activity as "nascent." Whether BlackRock's directional bet proves correct depends on variables outside the digital asset industry: the pace of AI agent deployment, the willingness of enterprises to grant agents financial autonomy, and the regulatory treatment of machine-initiated payments.
The world's largest asset manager has staked a research position. The on-chain evidence says the market has not yet followed.