The Missing Meter

1. Two facts that cannot both persist

The capex fact

The bust never kills the industry. In the first quarter of 2026, Google, Amazon, Microsoft, and Meta told their investors they intend to spend approximately $725 billion on capital expenditure this year, up roughly 77 percent from a record $410 billion in 2025.[2] Goldman Sachs projects a combined $5.3 trillion from these four companies between fiscal 2025 and fiscal 2030.[3] Investment in information processing equipment and software, about 4 percent of US GDP, was responsible for 92 percent of US GDP growth in the first half of 2025.[4]

The unit-cost fact

At the same time, the unit price of the output is in freefall. Stanford's 2025 AI Index measured the cost of GPT-3.5-equivalent capability falling from $20.00 per million tokens in November 2022 to $0.07 by October 2024, a 280-fold reduction in twenty-three months, with task-level declines ranging from 9× to 900× per year.[1] H100 rental prices fell from a 2023 peak near $8 per hour to $1–$2 by late 2024.[5] In January 2025, a Chinese lab's demonstration that frontier reasoning could be reproduced at a fraction of assumed cost erased roughly $589–$593 billion of Nvidia's market value in a single trading day, the largest one-day loss in stock market history.[6]

The paradox, and what it actually is

Markets are treating these two facts as a paradox, and the paradox has a name: bubble. The historical record suggests something more specific. Peak physical spending and collapsing unit costs, at the same moment, is what the crest of an infrastructure installation phase looks like. It looked exactly like this in 1846, in 1929, and in 2000. What follows the crest is not the death of the industry. It is the arrival of the industry's second act, which in every prior case was larger than the first, and which in every prior case was captured by different companies, operating a different layer, measured by a unit that did not exist during the buildout.

This paper makes four claims. First, the pattern is real: five prior infrastructures followed the same four-phase arc, and the mechanism that drives the arc is visible in AI today. The pattern also has a boundary condition, stated in Section 2, and two of the five cases sit on the far side of it. Second, the strongest objection, that AI capital depreciates too fast for the pattern to hold, is partially correct and must be answered rather than dismissed; the answer changes what to build, not whether the transition occurs. Third, the binding constraint on the efficiency era is now measurement. AI is not the first trillion-dollar infrastructure that lacks a unit for its output; it is the first that has one available, in an adjacent domain, and has not adopted it. Fourth, once such a unit is specified with care, it forces two facts into the open that no current metric can express: the cost of useful work has an interior optimum in offered load, and the industry's standard telemetry cannot resolve the latency bounds its most valuable workloads are sold against.

2. The pattern

Perez's four-phase arc

Every infrastructure revolution got its unit. Inference hasn't. Carlota Perez, studying five technological revolutions since 1771, identified a recurring structure (Technological Revolutions and Financial Capital, 2002).[7] An installation period, financed by speculative capital, builds the physical substrate in a frenzy that always overshoots. A turning point, usually a crash, transfers the substrate from speculators to operators at a discount. A deployment period follows, in which production capital puts the overbuilt substrate to work, and the technology's real economic payoff arrives, decades after the initial breakthrough and years after the bubble that supposedly discredited it.

The frenzy is not a defect in the process. It is the financing mechanism. No rational allocator funds 80 million miles of fiber against demonstrated demand; only a mania does, and the deployment era then inherits the fiber at cents on the dollar. As the venture capitalist Fred Wilson compressed Perez's corollary: nothing important happens without crashes.[8]

Five installments compressed

Five installments of the pattern, compressed:

Two regularities in this table do the argument's work. The bust never killed the industry. Britain's railway shareholders were ruined in 1847; Britain kept the 8,000 miles of track, and the track kept lowering freight costs for a century.[11] Telecom equity holders were wiped out in 2001; the fiber stayed in the ground, bandwidth prices fell up to 90 percent, and YouTube, founded in 2005 on that glut, generated roughly $31 billion for Alphabet in 2023 alone.[20]

Value migrated to the layer that measured and routed, not the layer that owned. Visa issues no cards, extends no credit, and owns no banks; it operates authorization, clearing, and settlement, and in fiscal 2024 it carried 234 billion transactions and roughly $16 trillion of volume at net margins above 50 percent.[15] In each case the physical layer was necessary, competitive, and margin-poor. The measurement-and-coordination layer was singular, standard, and margin-rich.

Exhibits 1 & 2

3. Act I: The bust that builds

The crash destroys capital. It does not destroy capacity. The pattern's first regularity is worth watching in slow motion twice, because the two runs bracket the industrial era and agree on every particular.

Britain, 1845 to 1850. Parliament approved 263 railway acts in 1846 alone, authorizing roughly 9,500 miles of new route.[11] Paid-up railway capital more than tripled inside the decade, from about £30 million to over £100 million by 1849. George Hudson, the Railway King, held the mania's mirror up early: he was paying dividends out of new subscribers' capital, a structure that acquired its modern name only decades later. Monetary tightening and the commercial crisis of 1847 called the loans; shares fell about two thirds by 1850, and only about two thirds of the authorized mileage was ever built. The investors' losses were permanent. So was the track's usefulness.

The United States reran the experiment with fiber, 1996 to 2006. The Telecommunications Act of 1996 opened the field; carriers laid roughly 80 million miles of fiber against demand that did not yet exist; annual capex peaked near $120 billion in 2000, with cumulative investment estimated well beyond $500 billion.[17] The NASDAQ closed at 5,048.62 on March 10, 2000, and fell roughly 78 percent over the following thirty-one months.[18] WorldCom filed the largest bankruptcy in US history at that point. The fiber stayed in the ground. Bandwidth prices fell up to 90 percent. AWS launched in 2006 on the economics the crash created. YouTube, founded in 2005 on that glut, was acquired for $1.65 billion in 2006 and generated revenues that Alphabet's annual YouTube advertising and Premium subscriptions passed $60 billion in 2025.[20]

Capital is destroyed at the turning point. Capacity is not. The write-down is the mechanism by which the next era acquires its inputs below cost.

4. Act II: Value migrates to the meter

The meter did not record the business. The meter created the business. Samuel Insull did not win Chicago by generating more electricity than his competitors; he won by metering demand — a device he licensed after seeing it in Brighton in 1894 — discovering that diverse customers have diverse peak loads, and using the meter to sell the same fixed capital many times over. His unit, the kilowatt-hour, plus his metric, load factor, converted electricity from a per-lamp luxury into a utility, cutting its price 87 percent and growing his customer base 40-fold across two decades.[12]

The box that Malcolm McLean standardized cut cargo handling from $5.86 to $0.16 per ton, a 97 percent collapse, and the TEU became the unit in which the entire logistics industry denominates itself.[16] Dee Hock structured Visa's predecessor so that no member bank could own the network; interchange became the unit; and Visa today carries $16 trillion of annual volume at net margins above 50 percent, owning no cards and no banks.[15]

The pattern's quietest claim is loudest on one axis: in each case the physical layer was necessary, competitive, and margin-poor. The measurement-and-coordination layer was singular, standard, and margin-rich.

5. The depreciation objection

The strongest argument against the pattern is Michael Burry's: the hyperscalers are understating depreciation by roughly $176 billion between 2026 and 2028, which means their earnings are overstated and their capital is not as durable as the railway track or the fiber.[21] The objection is partially correct and must be answered rather than dismissed.

The obvious response is that AI silicon is different from railway iron. An H100 purchased in 2023 faces competitive pressure from Blackwell in 2025 and from whatever follows it; the economic useful life of a GPU is shorter than the accounting useful life that most hyperscalers are currently booking. This is true. But the objection proves too much: if fast depreciation were fatal to the infrastructure pattern, the pattern would not have held for fiber, which became economically obsolete almost immediately after it was laid, and which nonetheless transferred to the deployment era at a discount and enabled YouTube, AWS, and every streaming service. The question is not whether AI silicon depreciates. Secondary H100 prices fall to a fraction of peak within two to three years, based on secondary-market data[64] The question is whether the capacity it represents transfers to the deployment era at a discount, and whether the deployment era's value is captured by the silicon or by the layer above it.

The depreciation dispute is, in fact, the paper's own argument wearing different clothes. The reason no one can state the correct depreciation schedule is that no one can state, in a standard unit, what the depreciating asset actually produces. The meter is missing from the income statement for the same reason it is missing from the invoice. Burry is right that the numbers are wrong. He is wrong that the wrongness is fatal. The answer to Burry changes what to build, not whether the transition occurs.

6. The inheritance mechanism

A data center is a stack of depreciation schedules. Land, shells, substations, transmission interconnects, cooling plant, and fiber depreciate over roughly 14 to 40 years. Silicon depreciates over 3 to 6 years, but can cascade from frontier training to inference, batch, and embedding workloads. Models depreciate in months. The deployment era can inherit the durable half of the stack and any silicon that still meets a lower-tier workload, but it cannot assume that every accelerator retains a resale market.

The public record does not yet provide a cohort-level utilization series for GPUs by age. AWS reports that it has not retired an A100 server and remains sold out of them; Google reports full utilization on seven- and eight-year-old TPUs. Prices and sold-out claims are not utilization rates. The final PDF therefore treats second-life clearing as a prediction to be measured, not an established result.[64]

What the credit market already priced

Every investment-grade GPU financing on the public record is anchored on contracted delivery rather than hardware liquidation value. CoreWeave's $8.5 billion facility draws as hardware is delivered against specific customer contracts, and its later credit agreement refers repeatedly to GPUs without an appraisal, residual-value, or remarketing framework. Lambda's senior secured facility likewise relies on GPU servers and their cash flows, tied to an investment-grade offtaker and fully amortizing by the end of 2030.[90]

The lender's question is a delivery question: whether an operator can demonstrate useful capacity independently of who happens to be buying this year. That is the missing measurement problem the served-token accounting is intended to address.

7. The protocols are arriving

The protocols are arriving in the historically correct order. Payments needed interchange (a standard for who owes whom), then authorization (BASE I, 1973), then settlement (BASE II, 1974). Between late 2024 and early 2026, AI acquired the same three layers in the same order:

• The Model Context Protocol standardizing how agents reach tools and data (open-sourced November 2024, ~97 million monthly SDK downloads, donated to a neutral foundation in December 2025).[30]
• Agent2Agent standardizing how agents reach each other (April 2025, 150-plus partners, IBM folding its competing protocol into it).[31]
• Three competing agent-payment rails — AP2, ACP, and UCP — standardizing how agents settle.[32][33][34]

The donation of these protocols to foundations is not altruism; it is the participants' recognition, learned from TCP/IP's victory over proprietary networking, that a coordination layer is only valuable if it is universal, and only universal if no one owns it. The rails are being laid. What runs on rails is traffic, and traffic requires a tariff, and a tariff requires a unit.

8. The missing meter

The strangest fact about a trillion-dollar industry

AI cannot say what its output costs. As of mid-2026 there exists no standardized, cross-vendor unit of account that expresses dollars per unit of verified useful work, conditioned on a stated service-level objective and a quality floor, governed by a multi-party body, and adopted in production pricing or disclosure.

Compute is bought in GPU-hours. It is consumed as answered queries, completed tasks, resolved tickets, and generated tokens, each under a latency constraint that determines whether the output was worth anything at all: a voice agent's reply that arrives in 200 milliseconds is a product, and the same reply in four seconds is a refund. Between the GPU-hour on the invoice and the useful-work-under-a-deadline on the income statement, there is no standard unit.

Dollars per million tokens is the leading candidate and the industry's de facto price sheet, but a raw token is a kilowatt-hour with no voltage standard: tokens vary in quality (which model, verified against what), in urgency (batch overnight or interactive p99), and in usefulness (a token of correct answer and a token of hallucination invoice identically).

The historical sequence is exact enough to be a specification. Before Insull's meter, electricity was priced per lamp: a flat fee per connected bulb, whatever it consumed — exactly as AI subscriptions today price per seat whatever the seat consumes. Per-lamp pricing made load invisible, so plants ran at 6 percent utilization, so capital costs stayed brutal, so electricity stayed a luxury. The demand meter made consumption visible; visibility revealed that customers peak at different hours; diversity of peaks meant the same turbine could serve the factory by day and the streetcar by night; load factor became the metric that turned utilization into strategy; and the price fell 87 percent while the market grew 40-fold. The meter did not record the business. The meter created the business.

AI's equivalent unit must denominate what the buyer actually buys, and what the buyer buys has three legs: cost, useful output, and a service-level condition. This paper names that unit the served token, abbreviated svt, and specifies it in Appendix A: one output token delivered inside its SLO and above its quality floor.

A futures contract still needs a grade

A liquid compute-futures market would not dispense with a unit. Grain futures reference a graded standard and crude references gravity and sulfur benchmarks. Compute has no such grade, so an early contract can specify its deliverable only by naming hardware or an accelerator-hour, which prices the asset rather than the useful work delivered under a deadline.

In August 2026 the Commodity Futures Trading Commission requested comment on compute derivatives, noting that compute may not yet exhibit the fungibility, standardization, and liquidity usually associated with a commodity derivative. CME Group and Silicon Data then announced H100 and B200 Rental Index Futures that settle against hourly GPU rental cost. The emerging contracts price hardware-time. Prediction 9 asks whether the market later adopts a work- or service-level-conditioned grade.

The prior art, and which leg each one amputates

The industry and the standards bodies have built every leg of the meter separately, which makes the survey of near-misses the strongest evidence that the fused unit is missing rather than impossible. Each candidate below is real, useful, and incomplete in a specific, documentable way.

The obvious objection is that dollars per million tokens is already good enough. It is not, for a precise reason. In 2026, one major lab changed its tokenizer; measured costs on identical prompts moved by double-digit percentages with no change in the underlying work.[59] The kilowatt-hour is defined by physics; it does not change when the utility upgrades its meters. A unit that changes when the vendor changes its implementation is not a unit. It is a price list in disguise.

TPC has price and correctness without the deadline. TPCx-AI has price without the deadline. MLPerf has the deadline and the quality floor without price. Goodput has the deadline without either. FOCUS has the money without the work. OpenTelemetry has the counting without the price, and cannot resolve the tightest deadlines. SPECpower and SCI have the rate form with a different numerator. Every leg is measured somewhere by someone competent, priced nowhere in common, and no congress has been convened.

The inversion: same model, opposite rankings

The same open-weight model served by fifteen providers inverts in cost-effectiveness depending on the workload's latency requirement. The cheap-but-slow provider delivers zero compliant output for the latency-strict workload — its cost per served token is undefined — and wins decisively for the batch workload. Toggle the workload to see the inversion:

The unit computed on production traces

Sections 8.2 and 8.3 are simulated and labeled [SIM]. The final PDF then computes the same quantities on four public production workloads from two operators, totaling 63,824 requests: Microsoft conversation and code traces, plus Moonshot long-context and conversation traces. Request counts, arrival times, and token distributions are [MEASURED, THIRD PARTY]; latency and cost are [SIM] because the public traces do not include a deadline or hardware record.

On Microsoft traffic at a ten-millisecond deadline, the conversation workload reaches its cost minimum at 15 percent utilization and the code workload at 10 percent, versus the synthetic model's 82.7 percent prediction. Two workloads on identical weights and hardware differ in cost of useful work by 39.5x: $1.06 per million served tokens for conversation and $41.86 for code. The final PDF treats the result as workload-specific evidence for an interior economic optimum, not as a uniform claim that production traffic is always burstier than Poisson.

9. What the efficiency era requires

Prescriptively, and in the order the history suggests:

11. The implication, stated in full

If the pattern holds and its boundary condition is satisfied, the most valuable layer in artificial intelligence will not be the models and will not be the chips. It will be the layer that measures, prices, and routes intelligence as work: the interchange, the meter, and the dispatcher, fused. That layer does not exist yet. Its unit does not exist yet for this domain, though it has existed for four decades in another. The protocols beneath it arrived in the last eighteen months, the buyers began demanding it this year, and every prior infrastructure cycle produced exactly one such layer within a decade of its turning point.

The boundary condition of Section 2.1 says what would have to be true for that to happen. The physical layer must stay contestable, which it currently is: fifteen providers serve one open-weight artifact at an 8.6-fold price spread. And the coordination function must be standardized before it can be bundled, which is undecided. Semiconductor fabrication shows what happens when the first condition fails. Cloud computing shows what happens when the second does. AI inference in 2026 has not yet failed either, and the window in which that remains true is the window this paper is about.

Railways got a clearing house. Power got a meter. Freight got a box. Money got interchange. Packets got a protocol. Databases, quietly and without anyone outside the field noticing, got price-per-tpmC in 1988, with an auditor, a disclosure report, and a three-year cost basis, and have been able to answer what a transaction costs ever since.

Railways got a clearing house. Power got a meter. Freight got a box. Money got interchange. Packets got a protocol.

Intelligence gets a meter next. The only open questions are whose, and whether it is neutral before it is bundled.

This draft is circulated for comment. The author thanks Xin Wang for the observation that mature-node fabrication capacity rotates to lower-tier demand rather than merely retiring, and for pointing to regulated compute futures as a complementary top-down instrument.

10. Nine predictions

A thesis that cannot lose is not a thesis. Dated, checkable predictions, with the conditions under which this paper is wrong. Prediction 3 is currently half met.

A draft specification for the unit

Status: draft for public comment. This appendix is deliberately boring. The kilowatt-hour is boring. Price-per-tpmC is boring. That is what made them work.

A.1 Scope and design goals

This appendix specifies a unit of account for machine intelligence delivered as a service, and the measurement rules without which the unit is meaningless. Design goals, in priority order: the unit must denominate what the buyer buys rather than what the seller owns; two parties with the same inputs must compute the same number; every number must carry its provenance; and no single vendor may control the definition. The unit is designed to be computed today from data that already exists, and the worked example in A.9 does so.

A.2 Definitions

A task is a request with a defined acceptable output. A quality floor is an operational acceptance test, declared before measurement: an evaluation threshold, task-completion check, programmatic assertion, human-rating floor, or reference-accuracy bound. Identifying the model artifact is not a quality floor. A service-level objective (SLO) is the latency contract under which output has value, stated as percentile bounds on time-to-first-token (TTFT) and time-per-output-token (TPOT), or their task-level equivalents, over a declared window.Compliant work is output that passes the quality floor and meets every term of the objective. Output that fails either is not discounted work; it is zero work that was paid for. Retried requests count their full spend and only their final compliant output.

A.3 The unit

The cost of useful work over a window is C = S / W, where S is total spend attributable to the workload in the window, including failed and retried requests, and W is compliant work delivered.

W is denominated in compliant tasks. A compliant task is one request with a defined acceptable output that passed its quality floor and met every term of its service-level objective. Units: dollars per compliant task. This is the primary form and the only form comparable across model artifacts, because the number of output tokens a model emits to complete a task is chosen by the seller.

The served token is the sub-case. Where the model artifact is held fixed, with the same weights, version, quantization, and tokenizer, output token counts become a stable measure of delivered quantity and W may be denominated in millions of compliant output tokens. One served token (SVT) is one output token delivered inside its service-level objective and above its quality floor. Units: dollars per million served tokens ($/MSVT). This is the right form for the routing question of where to place a fixed artifact, and it is the question of the worked example in A.9.

The commensurability rule. Token-denominated cost is defined within one artifact class and undefined across artifact classes. Any comparison spanning artifacts must be denominated in compliant tasks. A ranking of dollars per million served tokens across different models is not a conservative approximation; it is the error this specification exists to correct. Under a token denominator a model that pads its answer lowers its apparent cost, while under a task denominator padding raises S and leaves W unchanged. The reference implementation raises an exception rather than returning a cross-artifact token ranking.

A.4 The conditioning tuple

A value of C is non-conforming and undefined unless stated with all six of: (1) workload — the trace or trace-class measured, including arrival pattern and input and output length distributions; (2) model artifact — weights, version, quantization, and tokenizer, explicitly, which also governs A.3's commensurability rule; (3) placement — provider or self-host, silicon, region, batching policy, and offered load; (4) service-level objective — the percentile bounds, stated numerically; (5) quality floor — the acceptance test, stated operationally, with the evaluated fraction; and (6) window — start, end, and sustained load, with figures reported at declared percentiles. The window must be long relative to the system's own response time and stationary within itself. A conforming record reports window length divided by measured p99 request latency, and a split-half comparison of C over the first and second halves. Where the halves differ by more than ten percent, the window is non-stationary, C describes a transient rather than a placement, and the record is marked accordingly. This makes the number reproducible by an independent observer rather than a burst artifact.

A.5 Pre-registration rule

The workload scope and objective are declared before measurement, and results are reported for the declared scope and for the aggregate, both. Scope declared after the fact is selection, not measurement. This rule exists because it is the one every party will be tempted to break, and because the difference between a scoped number and an aggregate number, honestly co-reported, is itself diagnostic information about where a placement fails.

A.6 Provenance taxonomy

Every figure in a conforming record carries exactly one label: [MEASURED], observed on the declared workload in the declared window; [CONFIG], read from configuration rather than observed; [SPEC], taken from a datasheet or published price list; [SIM], produced by a model of the system; or [EST], judgment. Two rules complete the taxonomy. Inheritance: a derived figure carries the weakest label among its inputs. The red line: a comparison between placements may be labeled measured only if every side of it is measured; a measured number divided by a simulated one is a simulation, and presenting it otherwise is the accounting fog reproduced at the level of a single line item. The taxonomy is an auditable five-level simplification of the Type A and Type B distinction in JCGM 100:2008.

A.7 Conforming record

A conforming measurement is publishable as one row: the six-field tuple, S, W, C, the provenance label of each, and the measuring party. Three further fields are required: an interval on C, derived from the Wilson score interval on the compliance rate and propagated through the reciprocal because compliance can sit near zero or one; a record reporting C as a scalar is non-conforming. A sensitivity curve sweeps the objective across a declared range and reports compliance and C at each point. A margin to the objective is the ratio of measured percentile to bound, for time-to-first-token and time-per-output-token. A placement compliant at 0.99 of its bound and one compliant at 0.40 report the same C and are not the same asset.

A.8 Governance

The specification succeeds only if no one owns it. The historical instruction is uniform: interchange worked because Hock's consortium prevented any member from owning the rules; TCP/IP beat better-funded proprietary stacks because it had no owner to distrust; TPC's price-performance metrics are trusted across four decades of adversarial vendor competition because the pricing rules and the audit are the consortium's and not the submitter's; and MCP's donation to a neutral foundation followed the same logic. This draft is offered accordingly, for adoption, amendment, or replacement by any multi-party body that preserves A.3 through A.7. The two bodies best positioned are named in Prediction 2.

A.9 Worked example, entirely from public data

All figures in this example are [SPEC] for prices and [MEASURED, THIRD PARTY] for latency and throughput, from public provider benchmarks accessed in July 2026. The figures are trailing-72-hour medians, so the example is illustrative at p50 rather than a fully conforming p99 record. The same open-weight model (Llama 3.3 Instruct 70B) is served by 15 providers. The public price spread is 8.6×: $0.12 per million blended tokens at the cheapest fp8-quantized artifact to $1.05 at the highest. Throughput spreads from 15.2 tokens per second at the cheapest to 329.6 at the fastest. Expand each step below to follow the tuple from declaration to C.

Same weights, same input, opposite rankings, and the inversion is invisible to both metrics the industry currently prices with. The spread between those two correct answers — a factor of five on identical work — is the routing margin of Section 9, computed from nothing but public numbers. One further detail the tuple forces into the open: the cheapest artifact is fp8-quantized, so the two buyers are not entitled to assume the same quality floor until it is declared and tested, which under A.4 they must state, and under current market convention they are never asked to.

A.10 What this specification excludes, knowingly

Multi-tenant attribution of S is no longer excluded. Two bases are defensible and they bracket the honest answer. Under reserved share, spend is attributed in proportion to reserved capacity, which charges idle reserved capacity to the tenant that reserved it; this is the correct basis under a reservation and the conservative one. Under token share, spend is attributed in proportion to realized compliant output, which ignores idle capacity entirely and is a strict lower bound. A conforming record reports C on both bases and names which it uses. The gap between them is not noise to be eliminated. It is the tenant's own load factor restated in dollars, and it is diagnostic in exactly the way Section 4's diversity factor was.

Each is a reason this appendix is a draft. None is a reason the unit cannot be computed today, because A.9 just computed it.