READING № 04 · · ~11 MIN
The Machine Economy: The Fourth Reading
The fourth monthly reading of the Machine Economy Index. The index rose in October on a rebound in the agent registry, while the payment count fell back and one payer made more than half the month's payment value.
Published · Updated
What this is
This is the fourth monthly reading of the Machine Economy Index (MEI), as of 31 October 2026. The index combines 14 metrics in four components (Payment, Physical, Legal, Macro), weighted equally and joined by a geometric mean, under which a strong component only partly compensates for a weak one. Every input is fetched from a primary source on a fixed schedule, and every bound, weight and rule is published.
The index rose in October, from 29.3 to 30.2. One input did it: activity on ERC-8004, the Ethereum registry where AI agents record an on-chain identity, more than tripled and on its own added 1.04 points, more than the whole rise. Everything else together took off a little.
On the payment rail, which we read through x402, the open standard for machine payments over the web, October finished what September started. The transaction count fell to 3.4 million as the payments of the two addresses that had carried it left the 30-day window. The money went the other way: one address on Base paid more than half of the month's value, in 16 payments. And our count of x402's payers and payees, which we had held still since 24 September, was released on 24 October with thousands of single-use payers inside it; the figure's note records them until the last of them leaves the window, by 13 November.
Where things stand
The Machine Economy Index reads 30.2 out of 100 (31 October 2026), up from 29.3 on 30 September.
| Component | Oct | Sep | What it measures |
|---|---|---|---|
| Payment | 36 | 32 | Machine-to-machine settlement — on-chain volume, transaction counts, identity |
| Physical | 50 | 51 | Compute and storage capacity, and how much is being used |
| Legal | 11 | 11 | Whether the world's legal systems have caught up. GDP-weighted. |
| Macro | 43 | 42 | Adoption — developer and enterprise |
Each change in the table has passed our direction test. Before we say that the index or a component rose or fell, we recompute the change under 10,000 random variations of the settings our method had to choose, and check that its direction holds in at least 90% of them; for Legal, we check it under all three ways of weighting its categories that we publish. From 30 September to 31 October, the index's rise and the moves of Payment, Physical and Macro held in every variation. Legal is unchanged at 11, the same published value at both ends, so we make no directional claim for it. Our live pages apply the same test to the last 30 days before they show a direction (the rule called R1 on our methodology page), and it passes for 31 October too: 29.1 on 1 October against 30.2.
If you watched our live pages on 20 October, you saw the index read 31.2 and then, that evening, 29.6. The first figure was on version 1.2 and the second on version 1.3, which went live that day: it dates each legal finding from the day the law starts to apply, and restates the index from launch on that basis, as our September report explained. The two figures are on different bases, so we draw no direction from them, and every other figure in this report is on the 1.3 basis.
| Input | 30 Sep | 31 Oct | Effect on the index |
|---|---|---|---|
| ERC-8004 registry activity (capped, 30 days) | 279 | 1,036 | +1.04 |
| x402 transactions (30 days)† | 16.5M | 3.4M | −0.44 |
| x402 agent roles (30 days)† | 37,020 | 45,941 | +0.16 |
| x402 on-chain volume (30 days)† | $1.00M | $1.50M | +0.15 |
| Python agent-framework downloads (30 days) | 64.9M | 80.1M | +0.11 |
| MCP SDK downloads (30 days) | 220.0M | 258.3M | +0.08 |
| US firms using AI (Census Bureau) | 23.8% | 24.6% | +0.04 |
| Filecoin raw storage capacity | 1,368 PiB | 1,231 PiB | −0.05 |
| Akash compute spend (30 days) | $284,221 | $194,892 | −0.05 |
| Filecoin storage utilisation | 40.2% | 39.5% | −0.02 |
| Akash leases (30 days) | 26,996 | 25,639 | −0.01 |
| Legal (the LRRS) | 11 | 11 | 0 |
| Nvidia data-centre revenue; EU firms using AI; M2M subscriptions | unchanged | unchanged | 0 |
† On 30 September these values were frozen, held or paused, as our September report explained, so part of each effect is the catch-up to current values.
Each effect is the index on 31 October minus the index with only that input put back at its 30 September value. The effects add up to +1.02, and the index rose by 0.99; the difference is the geometric mean's interaction between inputs.
The balance gap, the plain average of the four components (35.1) minus the geometric mean (30.2), is 4.9: it shows how much the imbalance among the components holds the index back. Legal remains by far the weakest component, so it is still the input whose gains would lift the index most, point for point.
Payment — the count falls back, the money concentrates
The Payment component reads 36, up from 32. The identity registry, in the next section, added 4.7 points to it; the three x402 figures (transactions, on-chain volume and agent roles) together took off 0.7.
The count after the two payers
Our September report showed that most of the x402 transaction count had come from two addresses, each paying a single recipient, and that both stopped in mid-September. October shows the rest. As their payments left the 30-day window, the count fell from the 16.5 million we stored on 30 September (a frozen value: x402scan, the public indexer our x402 figures come from, had stopped updating; the count really stood at about 13.7 million) to 12.1 million on 1 October and 2.7 million on 8 October. It ended the month at 3.4 million. That fall took 0.44 points off the index.
What remains is a different rail. In the 30 days to 31 October, the largest payer made 12% of the transactions. The average payment was about 44 cents, but more than half of that comes from one payer's 16 payments, described below; without them, it was about 20 cents.
One payer, more than half the money
The value settled went the other way. In the 30 days to 31 October, $1.50 million settled through x402, against about $977,000 in the 30 days to 30 September. One address on Base paid $829,001 of it: 55%, in 16 payments averaging about $51,800, through one facilitator, the service that checks and settles x402 payments for the two sides. Without it, the month's value fell from about $847,000 to about $668,000.
- SETTLED
- $1.50M
- FROM ONE ADDRESS
- 55%
- PAYMENTS FROM IT
- 16
We publish volume uncapped: a dollar settled is a dollar, however it is split. Concentration is what our daily diagnostics are for: figures we publish every day on how x402's numbers are made up, which don't enter the score. One of their review triggers, X1, opens a review of the volume measure if one payer holds at least half of the 30-day volume on 14 consecutive days; the daily readings that count towards it began in November. The address's last payment was on 26 October, so unless it pays again, all of its payments leave the 30-day window by 25 November.
Agent roles: the hold ends, with a farm inside
Agent roles, the third x402 figure, adds up x402scan's payers and payees over 30 days. On 23 and 24 September one address paid a cent to each of 24,383 new addresses, and each became a payee. We held the figure at its 23 September value, 37,020, while that spray was in the 30-day window. On 24 October it had left, the fetch stored 49,219 and the hold ended. The figure read 45,941 on 31 October.
What it counted by then included a farm. Between 21 September and 14 October, 12,455 new addresses each paid one service one to three times, one address after another, and paid no one else. They are counted in agent roles until their payments leave the 30-day window, between 21 October and 13 November, and the figure carries a note saying so. A second payer sent one cent to each of 1,172 addresses, 30 to 55 a day, until 10 October. Each grew too slowly to move the figure by half in a day, so neither would have tripped our value-jump safeguard, which refuses to store a daily figure that moves by more than half at once. The spray was a sudden step; these were slow slopes.
Our methodology ruling of October, Ruling #19, replaces agent roles with a count of distinct paying addresses, in a new version due no earlier than 14 November. It goes live on a day when none of the distortions we have recorded, such as the spray and the farm, is in its 30-day window, or after 30 days with the effect of each one disclosed.
What was built on the rails
Around these numbers, the rails kept being built, much of it by the card networks:
- Mastercard opened early access to Agent Pay for Machines, built on x402 (20 October). An agent with a task and a spending limit answers "402 Payment Required" with a signed voucher, and the provider redeems the vouchers later, off-chain. Its pilot partners include Cloudflare, Coinbase and Base.
- Visa, BILL and Billtrust said AI agents now pay business invoices end to end, live, acting only on approved invoices (21 October).
- Stripe introduced treasury tools for agents (21 October): through its server for the Model Context Protocol, the open standard for connecting agents to tools, an agent can check balances, pay invoices, move funds and convert currencies, with the startup setting what needs a human's approval.
- Circle added agent payments to its CPN Managed Payments service: x402 or the Machine Payments Protocol handles how the agent pays, and Circle handles the seller's side, including payout in local currency (19 October).
- Sierra announced the Personal Agent Protocol, developed with Meta (6 October): an open standard built on OAuth for how a person's agent deals with businesses, with partners including Shopify, Stripe and Walmart, and a first specification due later in October.
Security got a hard look. The security firm V12 disclosed 25 vulnerabilities in x402's software kits, three of them critical (made public on 22 October, according to the Cloud Security Alliance); they were reported to the x402 team on 12 August, and the critical ones were fixed by 20 August. And a paper on agent wallets (I-HITL, 26 October) found that, across ten frontier models, an agent that can read the user's email will often raise its spending cap on its own.
Identity — the registry fills up again
Activity on the ERC-8004 agent registry rose from 279 to 1,036. We publish it with a cap of 100 events on each registering address and on each content origin, the website or IPFS location that hosts an agent's description file. Raw events rose from 1,261 to 4,145, and distinct addresses from 111 to 924.
The raw activity was concentrated, and the caps did their work. Two contracts made 69% of the window's raw events: one sent all of its 1,634 events from a single account, and the other's 1,239 came from 430 accounts. Each counts for at most 100. The rest of the rise came from new registrants: more than 800 of them, most registering one or a few times. The window's agent files were hosted at only 71 distinct origins, and the cap on each origin took the count from 1,374 to 1,036.
- RAW EVENTS
- 4,145
- PUBLISHED (CAPPED)
- 1,036
- FROM TWO CONTRACTS
- 69%
Identity also moved off-chain. Proof put x401 into production (15 October): an open HTTP protocol with which a site can ask an agent whom it acts for, and what that person approved, backed by identity proofing to a NIST standard and built to work with x402.
Physical — the captive chips head for market
The Physical component reads 50, down from 51. Nvidia's data-centre revenue stays at $89.02 billion until its next quarterly results. The move came from decentralised compute and storage: Akash's compute spend over 30 days fell 31%, to about $195,000, and its leases 5%, to 25,639; Filecoin's raw storage capacity fell 10%, to about 1,231 pebibytes, and its utilisation from 40.2% to 39.5%. Filecoin's Solstice upgrade (19 October) changed how its block rewards are paid; nothing we have read links it to the fall in capacity.
The month's news went to the caveat this component has carried since July, that captive silicon is outside it:
- Google will offer its TPUs on Oracle Cloud Infrastructure from 2027 (27 October).
- Amazon intends to offer its Trainium chips outside AWS in late 2027, with its chips business growing at triple-digit rates (30 October).
- Microsoft's Maia 200 now serves its production workloads, including Copilot, and Maia 300 comes with multi-gigawatt deployments from 2027 (12 October).
- Nvidia said it will invest $3 billion in Meta's equity offering, and Meta plans to expand its use of Nvidia's systems (26 October).
Two of the largest buyers' own chips, Google's TPUs and Amazon's Trainium, are set to be offered outside their owners' clouds from 2027. If that market grows, Nvidia's revenue will cover less of the compute being built; our methodology reviews the figure if Nvidia's share of the merchant market, the chips sold to others, falls below 60%.
Legal — unchanged, and a standard arrives
The Legal component reads 11, unchanged on the 1.3 basis: no finding in the index changed its status in October. Since then, a new legal survey, live from 5 November, has put Legal at 16; the November report will cover it.
The EU published its first harmonised standard under the AI Act on 19 October: EN 18286:2026, on quality management for providers of high-risk AI. A provider that follows it is presumed to meet Article 17, the Act's requirement to run a quality-management system. It changes no score, because none of our five legal categories covers product standards, but it is the first harmonised standard the Act's high-risk rules will rely on.
Macro — the downloads rise
The Macro component reads 43, up from 42. Downloads of our basket of five Python agent frameworks rose from 64.9 million over 30 days to 80.1 million, two-thirds of the rise from langgraph; the basket stays well below its level before the one-day step of 25 August that our September report described. Downloads of the SDK for the Model Context Protocol, the open standard for connecting AI agents to tools and data, rose from 220.0 million to 258.3 million.
The share of US firms using AI rose from 23.8% to 24.6% in the Census Bureau's business survey, and the EU figure held at about 20%. The UK's statistics office put its own figure at 30% of businesses (8 October). The Census Bureau has also started calling AI "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)", after a US executive order of 29 September told federal agencies to use that name in their own documents. The Census says the definition is unchanged, and so is our series: the answer label our adapter reads still said "AI current" on 2 November.
What we cannot see
Our standing blind spots are unchanged: closed-rail machine payments (corporate cards, cloud billing, metered APIs), off-chain payments such as those over Bitcoin's Lightning network (L402), captive silicon, energy, embodied machines, and digital-asset property law. October gave the off-chain one a new case: x402 payments settled off-chain. Mastercard's Agent Pay for Machines redeems its vouchers off-chain, so payments made that way will not show in the on-chain data our x402 figures read.
What we are watching
From 14 November 2026, at the earliest. The first x402 version under Ruling #19, which counts agent participation as distinct paying addresses. Its version note will set out the hold, the farm and how the old and new series are joined.
25 November 2026. Unless it pays again, the last of the 55% payer's payments leaves the 30-day window. Until then, trigger X1 reads its share every day.
2 December 2026. The EU AI Act's duty to mark synthetic content in a machine-readable format reaches AI systems placed on the market before 2 August 2026.
2027. Google's TPUs on Oracle's cloud and Amazon's Trainium outside AWS: captive chips offered beyond their owners' clouds.
How to check us
Everything in this report is derived from data you can pull yourself, or from primary sources you can read.
- The live index and every component, updated daily: /mei
- Every metric, its value, its source, its notes, and when it was last fetched: /data
- The full methodology, with the review triggers: /methodology
- The daily x402 diagnostics, with their history: /api/v1/diagnostics
- Version 1.3, its rules and the restated history: /methodology/versions/1.3
- A public API, free and unauthenticated: /api
The findings that rest on measurement come from the index and its stored history. The payer-level x402 figures come from x402scan's public per-payer data, read on 7 November 2026: we rebuilt each 30-day window from its daily records, so they can differ by a few percent from the totals x402scan showed on days its view was current (in our checks, up to about 3% for counts and 7% for amounts). The registry's contract-level figures come from our own read of Ethereum mainnet on 7 November 2026. The findings that rest on outside events come from primary sources: company releases, the Commission's and statistics offices' pages, the security firm's disclosure and the paper's abstract. Developments we could not confirm against a primary source are left out.
If we have made a mistake, we would like to know. The index is not the argument. The index is the instrument. Everything above is what it currently reads.
The Machine Economy Index is published by MachineEconomy.ai. This report covers October 2026 and was published on 8 November 2026. Its figures are as of 31 October 2026, on methodology version 1.3; anything it mentions from after that date carries its own date. The index is recomputed daily; the live value will differ.
Note added on 2026-11-11. The first x402 version, which counts agent participation as distinct paying addresses, will go live no earlier than 12 December 2026, not 14 November as written above. Our methodology ruling of 11 November, Ruling #26, fixes its days. The version is ready on the day its finished code runs in production alongside the current one, before it replaces it, and we publish that date on the day. It goes live on the first day after that on which both the old and the new count are current and the check of the day before finds no recorded concentration in either 30-day window. If no such day comes within 30 days, it goes live from the 31st day, and its version note states how much of each count the recorded concentrations held. A recorded concentration is at least 1,000 addresses that, in one 30-day window, paid only one payee or were paid by only one payer. It is a count: it says nothing about who runs the addresses or why, and it changes no value. The spray and the farm described above are two of them. Seven more were recorded on 11 November, and the last of them stays in the 30-day window until 11 December.