The Console

Every instrument on one desk. The needle row is the live reading; pick a switch or walk the rack — boards read the engine on cadence, and every figure traces to the data spine. Nothing here is a frozen snapshot.

source: ai_fragility engine · data as of 2026-08-08

Composite fragility across the 68 firms · F-score

FILED

Composite · 2026 Q2

49

Moderate, and holding — four quarters of composite stress across the 68, none of it priced.

Compare firms · L1–L5
loading the universe…

Pick firms to read their engine scores side by side — composite + all six indicators, live from watchlist.json.

The composite read over the 68-firm build-out scorecard — six indicators (I1–I6), weighted and aggregated into one number. Higher is more fragile. This is the engine behind the Bubble Watch band; here you see all six inputs, the universe breakdown, and the firms driving the pressure.

The index universe is the build-out core — Layers 1–4 (40 scored firms); the 25 Layer-5 adopters are tracked as comparators and excluded from the composite. Each gauge states its own coverage as scored/68. Where an indicator is unscored for a firm, the composite renormalises over the weights present — a missing indicator is neither a zero nor a silent drop.

Composite index · 0–100
49.4
Moderate
Universe · 68 firms · 5 layers
15 active
19 moderate
18 watch
16 inactive

Six Indicators

I1weight 20%
Depreciation integrity
39 mean9 elevated · 48/68 scored
I2weight 20%
Capex vs demand
62 mean33 elevated · 49/68 scored
I3weight 15%
Insider selling
41 mean11 elevated · 51/68 scored
I4weight 20%
Circular financing
49 mean18 elevated · 49/68 scored
I5weight 10%
Energy / returns
40 mean5 elevated · 49/68 scored
I6weight 15%
Organic demand
57 mean22 elevated · 49/68 scored

By Layer

The layer decomposition is being rebuilt to match the published map of the complex. The index itself is unchanged.

Most Elevated Firms

FirmLayerComposite FElevatedConvergence
XAI xAIL3
77.4
4 / 6active
TSLA TeslaL5
72.2
5 / 6moderate
ORCL OracleL2
69.2
4 / 6active
CRWV CoreWeaveL2
67.0
4 / 6active
SMCI Super MicroL1
66.3
4 / 6active
NVDA NVIDIAL1
65.0
4 / 6active
MSFT MicrosoftL2
63.1
4 / 6active
AVGO BroadcomL1
62.2
3 / 6active
OPENAI OpenAIL3
60.6
3 / 6active
AMD AMDL1
59.6
3 / 6active
GOOGL AlphabetL2
58.6
4 / 6moderate
ANTHROPIC AnthropicL3
58.0
3 / 6active

AI Fragility Index: composite of I1–I6 over 68-firm universe, layers 1–4 weighted (I1 20% · I2 20% · I3 15% · I4 20% · I5 10% · I6 15%). Indicator gauges = universe mean from indicators.json; composite = snapshot bubble_index. Engine: ai_fragility. As of 2026-08-01.

Open the full instrument →
← → or J / K walk the rack

The market's story minus the filings · D(t)

FILED

D(t) · 2026 Q2 · series high

+4.06

The market term spiked while ground-truth fell — the widest gap in the series.

D(t) · divergence+4.06σM(t) minus G(t) · 2026Q2 · series high
M(t) · market narrative+2.83standardized narrative read from calls and releases
G(t) · ground truth-1.23the filings themselves — revenue, capex, margins

What's driving the gap

AI-layoff share
+1.49
Insider selling
+1.24
Capex vs cash
+0.94

Hover the chart to scrub the quarters. Recomputed live by the ai_fragility engine each build, stamped with the as-of date.

Open the full instrument →
← → or J / K walk the rack

Committed compute ÷ supplier equity funded in cash · Recycling ratio

MIXED

Funded-cash basis

15.5×

On a funded-cash basis the build-out is its own biggest customer.

Tierratioclassas of
Funded cash — primary equity15.5×FILED2026-08-08
Filed — SEC-disclosed instruments5.2×FILED2026-07-02
Filed + reported — incl. secondary3.6×MIXED2026-07-02
PV-adjusted — funded supplier equity, 10%/yr13.0×ESTIMATE2026-07-02

Committed compute ÷ supplier equity funded in cash. Only the funded-cash tier is engine-live; each tier carries its OWN as-of — freshness and provenance are never conflated.

Open the full instrument →
← → or J / K walk the rack

Does the build-out close its own loop · The Circuit

MIXED

Financing graph

Does the build-out close its own loop? The circuit reads stretching — harder than last quarter.

The question the build-out cannot comfortably answer: does it close its own loop? Here is Indicator I4 in full — the financing graph, the loop coverage, the mark-to-model gap, and the loop concentration. Real numbers from the engine's circular-financing ledger. Not a forecast; the ledger as it stands.

Four Loop Gauges

Loop Coverage
Compute funded by circular equity
6.5%
% of committed compute funded by loop equity
$34.8B circular equity ÷ $539.5B compute committed (the core labs' own firm + reported compute commitments — OpenAI · Anthropic · xAI) — the inverse of the Recycling Ratio's 15.5×
Loop Concentration
Concentration
96%
% of committed in top cluster
8 detected cycles · 6 roundtrips in the graph
Mark-to-Model Gap
Mark-to-Model
$18.2B
capital valued at projected (not realised) revenue
Assets on balance sheets whose value assumes AI revenue projections hold
Graph Topology
Loop Nodes / Edges
12 / 33
tracked entities / financing edges
6 roundtrip paths detected in the financing graph
estimate · authored line

Pressure above 10%. There is no accounting standard for this measure. The line is anchored by analogy on ASC 280-10-50-42, which requires a public company to disclose any single customer representing 10% or more of revenues — the standard-setters’ judgment about when concentration in a demand relationship becomes material. Our measure is not that measure; the analogy is ours, and it is the reasoning we would defend. At 6.5%, vendor-funded demand sits below the level at which accounting standards treat single-source concentration as material.

Loop state · from the engine
How much of “the market” is the same money going around?
The financing graph, edge by edge — funded vs committed, at marked-up value. Straight from the engine’s Indicator-4 ledger; the recycling ratio is this table divided out.

The Circular-Financing Ledger

Every amount-bearing financing edge — all three types (invests · buys_compute · supplies), largest first — with firmness and the SEC accession where available. This is the whole ledger, not a top-N sample: the rows below sum to the total committed. The coverage gauge reads a subset of these rows — the core labs' own compute commitments — so every headline on this page traces back to the table. (Mark-to-model sits in its own gauge; it is a valuation adjustment, not a financing flow.)

From To Type $B Firmness Accession
OpenAIMicrosoftbuys_compute$250.0Bfirm0001193125-25-256321 ↗
OpenAIAmazonbuys_compute$138.0Bfirm0001018724-26-000014 ↗
NvidiaOpenAIinvests$100.0Bsoft0001045810-25-000230 ↗
AnthropicAmazonbuys_compute$100.0BreportedANTHROPIC sheet — REPORTED (CNBC Apr 20 2026; 1M+ Trainium chips; 5 GW …
GoogleAnthropicinvests$43.0BreportedANTHROPIC sheet — REPORTED; GOOGL sheet confirms investment exists but …
AmazonOpenAIinvests$35.0Bfirm0001018724-26-000014 ↗
NvidiaOpenAIinvests$30.0Bsoft0001045810-26-000021 ↗
AnthropicMicrosoftbuys_compute$30.0BreportedANTHROPIC sheet — REPORTED (CNBC Nov 2025)
NvidiaxAIsupplies$18.0BreportedXAI sheet — REPORTED
AnthropicxAIbuys_compute$15.0BreportedXAI sheet — REPORTED (announced 2026-05-20); Anthropic is xAI Colossus …
AmazonOpenAIinvests$15.0Bfirm0001018724-26-000014 ↗
MetaCoreWeavebuys_compute$14.2Bfirm0001769628-25-000050
MicrosoftOpenAIinvests$13.0Bfirm0001193125-25-256321 ↗
GooglexAIbuys_compute$11.0BreportedXAI sheet — REPORTED (announced 2026-06-05; regulatory filing abhs.in)
NvidiaAnthropicinvests$10.0Bfirm0001045810-25-000230 ↗
AmazonAnthropicinvests$8.0Bfirm0001018724-26-000014 ↗
OpenAICoreWeavebuys_compute$6.5BfirmCRWV sheet; CoreWeave FY2025 10-K (MSA entered May 2025) — PRIMARY
NvidiaCoreWeavebuys_compute$6.3Bfirm0001769628-25-000047
MicrosoftAnthropicinvests$5.0BreportedANTHROPIC sheet — REPORTED (CNBC Nov 2025)
NvidiaCoreWeaveinvests$3.7BsoftNVDA sheet; CoreWeave S-1 filed 2025-03-03 — PRIMARY (>5%); Q1 2026 13F…
TeslaxAIinvests$2.0Bfirm0001628280-26-026673 ↗
NvidiaCoreWeavebuys_compute$0.3BfirmNVDA sheet; CoreWeave S-1 filed 2025-03-03 — PRIMARY (Nvidia paid CoreW…
Total — all financing edges, every type · firm-committed$854.0Bfirm: $498.3B

The Circuit: Indicator I4 (circular financing) from the ai_fragility engine. Loop coverage = circular vendor equity ÷ compute committed (funded basis) — the share of the build-out's compute funded by its own loop; the inverse of the Recycling Ratio instrument's compute-per-equity reading. Financing edges: all 22 financing edges, every type (invests · buys_compute · supplies) · $854.0B total committed · $498.3B firm. Mark-to-model: capital valued at projected AI revenue. As of 2026-08-01.

Open the full instrument →
← → or J / K walk the rack

What the five biggest buyers say they'll spend in 2026 · Capex Watch

MIXED

The spend · 2026 calendar year

$695–725B

Guided $495–525B (Microsoft · Alphabet · Meta) and an estimated ~$200B for Amazon. Oracle's filed $55.7B is shown for contrast, not in the headline.

Firm2026 spendclassbasis
Amazon$200BESTIMATEno company full-year 2026 guidance exists — desk estimate. Method: FY2025 filed $131.8B (10-K 0001018724-26-000004) → TTM-to-Q1 $151.0B → Q1-2026 filed $44.2B ×4 = $176.8B floor + the filed sequential ramp ≈ $200B (10-Q 0001018724-26-000014)
Microsoft$190BGUIDANCECFO A. Hood, FY26 Q3 call, 2026-04-29 — stated calendar-2026
Alphabet$180–190BGUIDANCEQ1-2026 release, 2026-04-29 (raised for Intersect)
Meta$125–145BGUIDANCEpress release 2026-04-29, verbatim
Oracle$55.7BFILED10-K FYE 2026-05-31, accession 0001193125-26-277521

Headline $695–725B = the four hyperscalers; Oracle’s filed figure is contrast, not headline. No growth rate is paired with the 2026 figure — mixed evidence classes do not average.

Open the full instrument →
← → or J / K walk the rack

The full universe, scored from filings · Company Screener

FILED

Five layers

68

The full five-layer universe, scored from filings. The screener works these numbers.

Instruments

Quick access to any firm, arranged by its layer in The Complex. Click a chip to open the firm’s dossier. For the scored board — the six fragility indicators by company, by layer — open The Complex.

◿ as of 2026-07-19 · 52 firms in the chain · 9 unscored · 25 outside · one membership rule
Open The Complex — the scored screener →
What does this company SELL into the AI build-out? A company that sells nothing into it is not in The Complex.

Position in the chain is determined by what a company supplies, not by what it is called or how AI-adjacent it feels. This is the only test a reader can check us on, and it is applied identically to every name considered.

The chain · who sells what to whom
L1Compute & Infrastructure14

Sells the machines — silicon, fab equipment, servers, networking.

L2Hyperscalers & Cloud7

Sells capacity — operates data centres and rents compute at scale, including neoclouds.

L3Model Labs3

Sells or serves frontier models — trains foundation models as its primary business.

L4AI Software & Applications17

Sells software built on models — applications, tooling and platforms above the model layer.

L5Energy11

Sells power, or the systems that generate, deliver and manage it — the physical constraint under the whole build-out. Energy and power only.

Bloom Energyunscored · +238.7%Constellation Energyunscored · -24.8%Eatonunscored · +30.4%Entergyunscored · +24%GE Vernovaunscored · +64.8%NextEra EnergyNRG Energyunscored · -9.3%nVent Electricunscored · +66.7%Quanta Servicesunscored · +64.5%VertivVistraunscored · +4.1%
activemoderatewatchinactive · a solid chip opens the firm; a dashed chip is an unscored energy name (no page yet)
Outside The Complex · the market comparison set (25)

These names remain in companies.json and keep their /instruments/markets/<ticker>/ pages. They are the market comparison set — referenced from Markets, not a layer of The Complex.

Open the full instrument →
← → or J / K walk the rack

Calls pre-committed before the print · Earnings Desk

GUIDANCE
FirmPre-committed watchEarnings window
xAI XAIPrivate — does not report earnings
Tesla TSLADepreciation integrity, Capex vs demand, Insider selling, Circular financing, Organic demand2026-07-22 · AMC · reported
Oracle ORCLDepreciation integrity, Capex vs demand, Circular financing, Organic demand2026-09-10 · AMC · estimated
CoreWeave CRWVDepreciation integrity, Capex vs demand, Circular financing, Organic demandNot tracked
Super Micro SMCICapex vs demand, Insider selling, Circular financing, Organic demand2026-08-04 · AMC · estimated
NVIDIA NVDADepreciation integrity, Capex vs demand, Circular financing, Organic demandNot tracked
Microsoft MSFTDepreciation integrity, Capex vs demand, Circular financing, Organic demand2026-07-29 · AMC · reported
Broadcom AVGOCapex vs demand, Circular financing, Organic demandNot tracked
OpenAI OPENAIPrivate — does not report earnings
AMD AMDCapex vs demand, Circular financing, Organic demand2026-08-04 · AMC · confirmed
Alphabet GOOGLDepreciation integrity, Capex vs demand, Circular financing, Organic demand2026-07-22 · AMC · reported
Anthropic ANTHROPICPrivate — does not report earnings
BigBear.ai BBAICapex vs demand, Circular financing, Organic demandNot tracked
Micron MUCapex vs demand, Organic demandNot tracked
Intel INTCDepreciation integrity, Capex vs demand, Circular financingNot tracked
Caterpillar CATDepreciation integrity, Insider selling, Circular financingNot tracked
C3.ai AICapex vs demand, Insider selling, Organic demandNot tracked
Snowflake SNOWCapex vs demand, Organic demandNot tracked
MongoDB MDBCapex vs demand, Insider selling, Organic demandNot tracked
SoundHound AI SOUNCapex vs demand, Insider selling, Circular financingNot tracked
Marvell MRVLCapex vs demand, Organic demandNot tracked
Upstart UPSTCircular financing, Organic demandNot tracked
Adobe ADBECapex vs demand, Insider selling, Energy / returnsNot tracked
Dell DELLCapex vs demand, Insider selling, Organic demandNot tracked
Vertiv VRTCapex vs demand, Organic demandNot tracked
Amazon AMZNCapex vs demand, Circular financing2026-07-30 · AMC · reported
Atlassian TEAMCapex vs demandNot tracked
Salesforce CRMCapex vs demand, Organic demandNot tracked
Qualcomm QCOMCapex vs demand, Organic demandNot tracked
Meta METADepreciation integrity, Capex vs demandNot tracked
NextEra Energy NEEEnergy / returnsNot tracked
Arm ARMCapex vs demandNot tracked
Datadog DDOGInsider sellingNot tracked
TSMC TSMCapex vs demandNot tracked
Cloudflare NETCapex vs demandNot tracked
UiPath PATHInsider selling, Organic demandNot tracked
ServiceNow NOWNot tracked
Disney DISNot tracked
Lam Research LRCXCapex vs demandNot tracked
Eli Lilly LLYNot tracked
Cisco CSCONot tracked
ASML ASMLCapex vs demandNot tracked
Netflix NFLXNot tracked
Palantir PLTRInsider selling2026-08-04 · AMC · estimated
Intuit INTUNot tracked
CrowdStrike CRWDNot tracked
Palo Alto Networks PANWNot tracked
IBM IBMNot tracked
Apple AAPL2026-07-30 · AMC · reported
GE Aerospace GENot tracked
Deere DENot tracked
Accenture ACNNot tracked
Walmart WMTNot tracked
Costco COSTNot tracked
Coca-Cola KONot tracked
Procter & Gamble PGNot tracked
McDonald's MCDNot tracked
Home Depot HDNot tracked
UnitedHealth UNHNot tracked
JPMorgan Chase JPMNot tracked
Visa VNot tracked
Mastercard MANot tracked
Exxon Mobil XOMNot tracked
Boeing BANot tracked
FedEx FDXNot tracked
Nike NKENot tracked
T-Mobile TMUSNot tracked
Comcast CMCSANot tracked
All 68 pre-committed calls · as of 2026-08-01
Open the full instrument →
← → or J / K walk the rack

The cuts, floor-led · Layoffs

MIXED

Roster as of 2026-07-13 · 26 days

The AI claim is the floor, 68,400 — not the 160,588 tracker total. Attribution scored per firm.

AI-attributable · floor68,400defensible — AI-driven × high confidence. The claim.
Generous read · ceiling129,300AI-driven or -partial, all confidences
Tracked total160,588every tracked cut — NOT the AI figure

Attribution is scored per company from primary filings and company statements — cut count × attribution × confidence — not from press aggregates. Where no primary exists for a figure, we say so on the record. 69 companies; pending-analysis and none-reported firms are stated absences, not zeros. Open the full board →

Open the full instrument →
← → or J / K walk the rack

Score a portfolio in the browser · Exposure

ESTIMATE

In-browser

Score a portfolio against the fragility universe, in the browser.

Open the full instrument →
← → or J / K walk the rack

Scrub the market term against ground truth · Divergence Explorer

FILED
Live

The Divergence

data as of 2026-08-08

How to read it: D(t) = M(t) − G(t), in standard deviations. A positive, widening D means the story is running ahead of the receipts — recomputed live by the engine, stamped with the as-of date.

Permalink · the full Explorer →
← → or J / K walk the rack

Four scaling vectors on one log scale · Scaling Timeline

MIXED

Log scale

Four scaling vectors on one log scale — three race upward, one collapses.

Instruments

The vectors of AI scaling — anchored, sourced, on a log scale.

◿ SOURCED ANCHORS · Epoch AI · OpenAI · SemiAnalysis

The interactive timeline is being migrated to the house chart standard. Its data is loaded and sourced below; the plotted, scrubbable chart returns with the renderer. Nothing here is estimated — every anchor traces to a public milestone.

The headline read
~5×/yrTraining compute
~every 5 monthsCompute doubling
~1000×Context window, 2020–2024
100MW → 2GWCluster power
~10×/yrInference cost
The vectors · where each is anchored
VectorSpanSource
Compute (training FLOP)2012–2025 · 8 anchorsEpoch AI training-compute series
Context window (tokens)2019–2026 · 6 anchorsOpenAI / Google model documentation
Cluster power (MW)2018–2026 · 6 anchorsSemiAnalysis / datacenters.com — xAI Colossus
Inference cost ($/Mtok)2021–2024 · 4 anchorsa16z LLMflation
Algorithmic efficiency2020–2026 · 4 anchorsEpoch AI algorithmic progress in language models
Energy efficiency (FLOP/watt)2019–2026 · 5 anchorsEpoch AI ML hardware energy efficiency
Open the full instrument →
← → or J / K walk the rack

AI against dot-com, aligned at boom start · Cycle Memory

ESTIMATE

Shape, not prophecy

AI against dot-com, aligned at boom start. Shape, not prophecy.

Cycle Memory · the same chart, honestly aligned

One index, both eras, no peak-picking. The NASDAQ Composite indexed to 100 at each era’s boom start: Feb 1996 for the telecom build-out, Nov 2022 for the AI build-out. The x-axis is months since that anchor. The two lines land on the same scale. Draw your own conclusions.

AI era now
AI index now
Telecom at same month
Telecom peak month
Awaiting data · live ingest not yet wired The AI-era line currently uses T3-authored anchors from FRED NASDAQCOM public data. Automated monthly ingest is not yet wired. The telecom-era anchors are fixed public record from FRED NASDAQCOM. Both series are indexed to 100 at boom start; no values are hardcoded in this page.
What this does and does not say The overlay shows shape, not prophecy. Two cycles is a sample of two. This instrument cannot tell you when or whether the AI cycle peaks, only where it sits relative to one prior analog. A chart that looks similar is not a forecast; it is a question made visible.

The structural analog

In 2000, Lucent carried ~$15B of customer financing against ~$300M operating cash flow — dark fiber overcapacity. Today ~$540B committed AI compute rests on ~$35B filed supplier equity (~15.5×). See recycling-ratio and methods instruments.

Related instruments

THE CATCH · CYCLE MEMORY · FRED NASDAQCOM both eras · shape, not prophecy. Telecom boom start: Feb 1996 (Telecom Act signing). AI boom start: Nov 2022 (ChatGPT launch). T3-authored series — both eras indexed to 100 at boom start.

Open the full instrument →
← → or J / K walk the rack

Adoption vs demonstrated productivity · The Lag

FILED

By vertical

Deployed everywhere, visible almost nowhere — the payoff scoreboard by industry.

The payoff scoreboard reads industry adoption against demonstrated productivity, from the verticals’ own filings — deployed everywhere, visible almost nowhere. Each bar is a vertical; the tally in the needle row is how many are paying off today. The chart is live from the engine; the Industries board carries the per-vertical receipts.

The per-vertical receipts · Industries →
← → or J / K walk the rack

How the desk measures D(t) · The Instrument

METHOD

Method · self-index

How the desk measures D(t) — the method, stated in advance.

We instrument ourselves to index the future. Capex Watch reads the financing runway from the outside. This reads the productivity lag from the inside — by measuring, in real time, how much cognitive work AI absorbs inside our own research desk. The result is D(t): the divergence between market narrative M(t) and filings ground-truth G(t).

D(t) · divergence+4.06M(t) minus G(t) — narrative ahead of the filings · 2026Q2
M(t) · market narrative+2.83standardized narrative read from calls and releases
G(t) · ground truth-1.23the filings themselves — revenue, capex, margins
F_env

The Plumbing

environmental signal

Infrastructure friction, tooling latency, context-switching cost — the invisible tax on every cognitive operation. Logged as tool-call latency distributions and model-switch frequency.

F_cap

The Brain

capability signal · human moat

How much of the analytical lift is genuinely AI-generated versus human-directed synthesis — output lineage tagged in session logs: AI-drafted, editor-revised, fully original.

F_dir

The Steering Wheel

direction signal · human moat

Editorial agency — how often the human overrides, redirects, or discards AI output. High F_dir means the desk still steers. Override rate, prompt-revision frequency, final-pass edit distance.

← → or J / K walk the rack

The Index's printable summary brief · Fragility Brief

FILED

Printable · reproducible

The Index's printable summary brief — six indicators, reproducible.

The dataset

Filing-sourced and reproducible

Six filing-sourced indicator tables sit underneath this brief, each computed from the source filings: an accounting table of useful-life changes, a capex-versus-demand table, the insider Form 4 record, the financing graph of the compute complex, the disclosed energy commitments, and a ground-truth deterioration series. The brief is built to be reproducible — every figure derives from those filing-sourced tables.

Download the tables · CSV, primary-source notes inline

depreciation.csvcapex_demand.csvinsider.csvground_truth.csvsoxx_daily.csvfinancing_edges.csvGitHub — data + reproducible models ↗

Each table carries its 10-K / Form 4 accession numbers inline, and shows blanks (NOT_SOURCED) rather than imputing. The circular-financing edge ledger (indicator 04) is published above — revised 2026-07-02 with the Amazon–OpenAI equity legs and a funded_usd column; the energy indicator (05) rests on qualitative disclosures and is discussed in its section. Cite as: The Desk, “AI Fragility” dataset (2026), /brief.

One discipline runs through all of it: where a value cannot be sourced cleanly from a filing, it is shown blank rather than imputed. The point is to read the cycle in the numbers the companies publish themselves, not in estimates layered on top of them.

Indicator 01 · Depreciation integrity

Are the assets aging faster than the books admit?

Has a firm extended the useful life of its depreciable assets — converting paper income without a matching dollar of cash? A life shortened scores zero, regardless of size.

The first indicator asks a narrow accounting question with a wide reach. When a firm extends the useful life of its servers, the same hardware cost is spread over more years, annual depreciation falls, and reported operating income rises — on paper alone, with no extra cash, no new customer.

# §2.1 — paper benefit from a useful-life extension
delta_dna = ppe_depreciable × ( 1/life_old − 1/life_new )
# hard rule: a life shortened scores 0, regardless of size

The direction of travel is uniform: every firm that touched a useful life lengthened it, and four did so while running the largest AI-capex programs on record. Amazon is the control — it moved the same lever the other way, six years to five, and absorbed a $1.4B charge against income, which is why it scores zero here despite carrying the heaviest depreciation line ($41.86B) in the set.

The signal is not the size of depreciation; it is the choice to make it smaller while everyone's assets are aging faster.

Indicator 02 · Capex vs demand gap

Is the spending outrunning the demand?

Is AI capital spending outpacing the revenue that would justify it? The break-even hurdle is set generously, so the firm gets credit for all segment revenue, not just AI lines.

# §2.2 — required revenue per $1 of capex per year
factor = ( CoC + 1/L ) / m = ( 0.10 + 1/6 ) / 0.30 = 0.889
# fail when FY2025 segment revenue < capex × 0.889

One firm fails the break-even test on full segment revenue: Alphabet, where Google Cloud's $58.7B sits $22.6B below the $81.3B the capex requires — a 28% shortfall. Capex is also growing roughly 2–4× faster than the revenue lines it funds across the cohort, even where the level test still clears.

FirmCapex / revenue growth
Meta3.95×
Amazon3.25×
Alphabet2.07×

At the system level the aggregate gap widens from $78B to $90B over four quarters. Spending is being committed ahead of the demand — and the test is built to flatter the firms, not to indict them.

Indicator 03 · Insider selling intensity

What are the people who know most actually doing?

Two kinds of insider selling look identical on a tape and mean opposite things. Pre-scheduled 10b5-1 plan sales score low; the signal is discretionary selling — a sale an officer chose to make, in a window when they held material non-public information, with no 10b5-1 footnote on the Form 4.

The three compute leaders divide cleanly. The discretionary cluster — not the headline dollar — is what scores, which is why the largest sellers by dollar (both on 10b5-1 plans) are discounted while smaller discretionary clusters rate higher.

FirmDiscretionary10b5-1 planLargest single seller
NVDA$0.93B$1.57BDir. Mark Stevens $802M discretionary
AVGO$0.50BCo-founder Samueli $749M plan
AMD$0.02B$0.29BCEO Su plan

NVDA's $0.93B discretionary is led by director Mark Stevens at $802M with no detected plan, against $1.57B run through confirmed 10b5-1 plans — including CEO Huang's $1.05B, under 1% of his stake. AVGO's $0.50B discretionary is spread across the entire C-suite — CEO Tan, the CLO, the CFO, and two more officers, none with a detected plan. AMD is the quiet one.

Discretionary selling is not a one-quarter event. The universe-level Form 4 total rises every quarter across the window — from $0.85B in 2025Q3 to $1.10B in 2026Q2, a 29% increase — while the same names were guiding investors toward accelerating AI demand.

Indicator 04 · Circular financing

Is the money going in a circle?

The structure is a loop: an investor funds a lab, the lab commits to buy compute from the investor's cloud, that cloud revenue underwrites the investor's capex, and the capex buys the investor's own chips through the lab it funded.

The financing graph of the AI-compute complex is a directed multigraph over twelve principals and four edge types — invests · buys_compute · supplies · marks_up. The recycling ratio measures the loop's leverage: compute committed out of the core labs (OpenAI, Anthropic, xAI) divided by equity put in, across three provenance tiers.

The same dollar of disclosed equity supports roughly 15.5× committed compute on a funded-cash basis (revised 2026-07-02 from 26× — Amazon's Q1 2026 $15B funded OpenAI stake widened the equity base), easing to ~3.6× only when every reported secondary round is admitted as equity. Present-valued at 10% over each commitment's disclosed horizon, the funded-cash ratio is about 13× — nearer 11× if the undated Microsoft commitment is discounted over a typical cloud term. Provenance, not arithmetic, moves the number; stock or flow, discounted or not, the loop turns far above any arm's-length benchmark.

Recycling ratio by equity tier — funded supplier equity → filed → +reported → PV-adjusted.

Two destinations carry the loop: of the labs' committed compute — the same $540B universe as the ratio — Microsoft and Amazon receive 96% (98% on the filing-grade subset). Mark-to-model gains booked on those same customer stakes total $18.2B (Microsoft +$5.9B — primarily the OpenAI recapitalization dilution gain — Amazon +$12.3B) — earnings recognized on the appreciation of the firms one funds. Eight directed cycles run through the cash-flow subgraph, and the largest single commitment — Nvidia's $6.3B backstop to CoreWeave — surfaced only in a September 2025 8-K (accession 0001769628), absent from the March 2025 IPO prospectus that first sold the relationship.

Indicator 05 · Energy & diminishing returns

Are physical limits starting to bind?

Are power, cooling, and chip economics beginning to cap capability gains? This is the thinnest-data indicator in the framework and carries the lowest weight (0.10) — we will not present estimate as measurement.

The firm-level cost-per-capability curve is largely proprietary, so this indicator does not try to measure it. What the filings do record, unambiguously, is the scale of power being committed — the appearance of gigawatt-scale capacity figures inside the same compute-purchase agreements that drive the circular-financing loop. The build stops being denominated in dollars and starts being denominated in power.

Power commitmentCapacityProvenance
OpenAI → AMD6 GWFiling 8-K EX-99.1, 2025-10-06
Anthropic → Amazon5 GWMedia not yet filed
Anthropic → Google>1 GWMedia not yet filed

Three edges carry an explicit gigawatt figure — 12 GW in aggregate — but exactly one is filing-sourced. By the methodology's own rule, that single filing item is the floor under any elevated read: the indicator is directionally supportive, not independently load-bearing, and is flagged as such. The cost-per-capability curve that would let it stand on its own is deferred to Phase 2.

Indicator 06 · Organic end-user demand

Is the revenue real, or recycled?

Does reported AI revenue reflect genuine paid adoption by independent end-users — or is it recycled through the same ecosystem that funds the build, or rebranded from existing product lines?

The test is anchored on the MIT NANDA finding that roughly 95% of enterprise GenAI pilots show no measurable P&L impact (Fortune, August 2025). Headline growth in the 30–50%+ band scores well only when paired with demonstrated paid retention and pilot-to-production conversion above 50%; growth sourced from ecosystem participants scores worse, not better. The indicator scores the source of the growth, not its rate.

Revenue growth alone clears the headline band for most of the complex — CoreWeave at 168%, Broadcom at 64%, four firms clustered at 32–36%. CoreWeave is the limiting case: 67% of its FY2025 revenue is a single counterparty — Microsoft, "Customer A" in its 10-K — with the remainder committed by OpenAI, Meta, and Nvidia. Every named buyer is an investor in, or a lab funded by, the same circular structure.

That is growth from ecosystem participants rather than demonstrated independent end-user retention — the band the rubric reserves for recycled demand, and exactly what the NANDA anchor predicts: an "AI revenue" label growing fastest where the demand is most recycled, not where paid conversion is most proven.

The synthesis · Divergence gauge

The tape versus the filings

D(t) = M(t) − G(t) sets a market signal against a ground-truth signal. The market term M(t) is the equal-weight mean of three full-window z-scored components of SOXX price behaviour — 63-day momentum, price-to-trend overextension, and 20-day annualized instability. The ground-truth term G(t) is the negative mean of three deterioration z-scores — AI-layoff share, discretionary insider selling, and the capex gap. The gauge widens when momentum and overextension climb while the fundamentals erode.

Toggle between the composite (M, G, D) and the three ground-truth signals underneath G(t). Source: SOXX + ground-truth series.

Through 2025Q1 the two signals track close and D(t) sits below zero — price had not yet detached from fundamentals. In 2026Q2 the gap inverts hard: M(t) jumps to +2.83 as SOXX closes at 639.45 (63-day momentum +88.0%, instability +0.74 annualized) while G(t) falls to −1.23, dragged by the AI-layoff share and discretionary insider selling both reaching their window highs.

D(t) widens from −1.80 to +4.06, a +5.86 swing — the strongest move in this four-quarter series so far (n=4: descriptive, not a long-run signal).

Method & limitations

What would prove this wrong

This brief is built to be reproducible: every figure derives only from filing-sourced inputs. Each indicator is computed only from filing-sourced inputs; where a value cannot be sourced cleanly it is shown blank rather than imputed.

Two Phase-1 simplifications are stated plainly. The divergence gauge standardizes its components over the full window — it is descriptive, not real-time: it carries look-ahead bias and is not a tradeable signal, and an expanding-window version is deferred. It also weights its three market components equally; empirical calibration is future work. Indicator 05 (energy) rests on the thinnest data in the set and is weighted accordingly — directionally supportive, not independently load-bearing.

The falsifier is built in: if the ground-truth signal turns back up — demand converting, the capex gap closing, insider selling normalizing — the divergence closes and the boom earns its price. We publish the number either way.

Printable brief →
← → or J / K walk the rack

Every figure sourced or labeled · Methods

METHOD

The reference

Every figure sourced or labeled — the single reference for how we measure.

How the desk measures the build-out — every method and source stated in advance. The public track record — the calls, the restatements, the receipts — is on the Receipts page.

Methods, Sources & Data

How we measure the build-out

Everything the desk publishes rolls up to two instruments and one universe of companies. This page is the single reference for both — the rubric and its weights, the divergence formula, the company set, every named source, the data-freshness dates, how we label a figure, and what would prove us wrong. It's meant to be linked from every metric on the site.

Compiled from the published methodology · data as of 2026-08-01 · the reproducible models and tables are published on the desk’s Resource Hub

Overview

Two instruments, read together

The desk measures one question — the race between the productivity lag and the financing runway — on two instruments. The Fragility Index is a structural read: the equal-weighted average of six filing-sourced indicators across the companies that are the build-out. The divergence gauge is a timing read: a market signal set against a ground-truth signal, quarter by quarter. The Index says how fragile the structure is; the gauge says whether the market's price has detached from the fundamentals. They sit side by side because either alone is half the picture.

The rubric

The six fragility indicators

Each company is scored 0–100 on six indicators (higher = more fragile). The weights are fixed and sum to 1.00. Two indicators have a published closed-form formula; the other four are computed from filings but described qualitatively rather than as a single equation.

IndicatorWhat it asksWeight
01 · Depreciation integrityAre the assets aging faster than the books admit? A life shortened scores zero, regardless of size.0.20
02 · Capex vs demand gapIs AI capital spending outpacing the revenue that would justify it? The break-even hurdle is set generously.0.20
03 · Insider selling intensityPre-scheduled 10b5-1 sales score low; the signal is discretionary selling in a window holding material non-public information.0.15
04 · Circular financingIs the money going in a loop — investor funds a lab, lab buys the investor's compute, that revenue underwrites the capex?0.20
05 · Energy & diminishing returnsAre power, cooling, and chip economics starting to cap capability gains? The thinnest-data indicator — lowest weight.0.10
06 · Organic end-user demandIs reported AI revenue genuine paid adoption by independent end-users — or recycled, or rebranded from existing lines?0.15

The two published formulas

# 01 — paper benefit from a useful-life extension
delta_dna = ppe_depreciable × ( 1/life_old − 1/life_new )
# hard rule: a life shortened scores 0, regardless of size
# 02 — required revenue per $1 of capex per year
factor = ( CoC + 1/L ) / m = ( 0.10 + 1/6 ) / 0.30 = 0.889
# fail when FY2025 segment revenue < capex × 0.889

Honesty note

Indicators 03–06 are computed only from filing-sourced inputs but do not publish a single closed-form equation; they are read as standing analyst judgments on the 0–100 scale, reviewed each quarter. Indicator scores are internal judgment, read for convergence; figures are sourced or labelled.

Instrument one

The Fragility Index (0–100)

A company's composite is the weighted mean of its six indicator scores (missing scores are dropped and the weights renormalized). The Index is the equal-weighted average of those composites across the build-out core (Layers 1–4: compute & infrastructure, hyperscalers & cloud, model labs, AI software). The broader-market comparators in Layer 5 are excluded — they're the control group, not the build-out. It's reproducible from the published per-company scores.

composite(c) = Σ wᵢ·scoreᵢ / Σ wᵢ   (over present scores)
Index = mean( composite(c) )  for c in Layers 1–4

Band thresholds

ReadingBand
≥ 75Severe
60–74Elevated
45–59Moderate
30–44Contained
< 30Low

Current reading: 49 (Moderate), 2026 Q2. The Index is recomputed from the published company scores; see the live Fragility Index.

Instrument two

The divergence gauge — D(t)

The gauge sets a market signal against a ground-truth signal:

D(t) = M(t) − G(t)

M(t), the market term, is the equal-weight mean of three full-window z-scored components of SOXX price behaviour — 63-day momentum, price-to-trend overextension, and 20-day annualized instability. G(t), the ground-truth term, is the negative mean of three deterioration z-scores — AI-layoff share, discretionary insider selling, and the capex gap. The gauge widens when momentum and overextension climb while the fundamentals erode.

The four-quarter series

QuarterM(t)G(t)D(t)
2025 Q3−0.820.98−1.80
2025 Q4−0.730.43−1.16
2026 Q1−0.50−0.18−0.32
2026 Q22.83−1.23+4.06

Stated limitation

The gauge standardizes its components over the full window — it is descriptive, not real-time: it carries look-ahead bias and is not a tradeable signal (an expanding-window version is deferred). The +4.06 reading is the strongest move in a four-quarter series (n=4: descriptive, not a long-run signal). It weights its three market components equally; empirical calibration is future work.

The set

The company universe — 43, 68, and the rings

Two counts appear across the site, and they measure different things:

SetWhat it isNames
Full boardAll five layers of the build-out — the Markets tape.68
Build-out coreLayers 1–4 (compute & infrastructure, hyperscalers & cloud, model labs, AI software). This is what the Fragility Index scores.43
ComparatorsLayer 5 — the broader-market control group, excluded from the Index.25

So 68 = the full five-layer set; 43 = the build-out core (Layers 1–4); the remaining 25 are the Layer-5 comparators that act as a control group. The Fragility Index is computed on the 43, not the 68.

Separately, Explained maps the build-out as three rings — a core of 43, a supply chain of 39 that feeds it, and a demand ring of 21 industries that must pay it back. The ring counts are a different partition from the five-layer board; the enumerated name lists for the 39-name supply chain and 21-name demand ring are not yet published alongside the core.

Provenance

Data, sources & reproducibility

Every figure derives from filing-sourced inputs; where a value cannot be sourced cleanly from a filing, it is shown blank rather than imputed. Each table carries its 10-K / Form 4 accession numbers inline. The indicator pipeline is computed in Python, and the underlying tables are published open.

Named sources

SEC 10-K filings — accounting & capex tables (e.g. Amazon FY2025 10-K Note 1, accn 0001018724-26-000004). sourced
SEC Form 4 — the insider-selling record (discretionary vs 10b5-1). sourced
SEC 8-K — financing structure (e.g. Nvidia→CoreWeave backstop, 8-K accn 0001769628; OpenAI→AMD 6 GW, 8-K EX-99.1, 2025-10-06). sourced
SOXX daily — iShares Semiconductor ETF price series, the market term for M(t) (soxx_daily.csv). sourced
MIT NANDA / Fortune (Aug 2025) — ~95% of enterprise GenAI pilots show no measurable P&L impact. attributed
Kansas City Fed / BLS — AI-attributable TFP (~+0.07pp/yr). attributed
Michael Burry (Scion) — ~$176B understated depreciation 2026–2028; carried as his allegation, not an audited figure. attributed

Open data & reproducible models

the desk’s Resource Hub (/research/resources/) — data + reproducible models.
CSV tables: depreciation · capex_demand · insider · ground_truth · soxx_daily.

Freshness — when each dataset was last generated

DatasetFeedsGenerated
chart-data.jsondivergence gauge, recycling ratio2026-08-01
circuit-vitals.jsonCircuit vitals (adoption, recycling, players)2026-07-20
history.jsonFragility & divergence history2026-07-05
payoff-data.jsonindustry payoff coverage2026-06-30
circuit-reports.jsonthe weekly Circuit reportWeek of June 30, 2026

Discipline

How we label a figure

Credibility is the only thing the desk sells, so every number carries its provenance. Three tags run through the site:

sourcedattributed / estimateeditorial read

Sourced figures trace to a primary filing with its accession number. Attributed figures are named as such — a short-seller's allegation, a survey, a Fed estimate — never dressed up as measurement. Editorial reads (the regime call, the loop framing) are labelled as The Desk's interpretation of the measures, not a measurement. Unverified figures are flagged or removed. A few standing examples:

· The regime call is The Desk's editorial read of the metrics, not a measurement.
· The recycling ratio moves with provenance, not arithmetic: ~15.5× on a funded-cash basis, ~13× present-valued at 10%, easing to ~3.6× when every reported secondary round is admitted as equity (revised 2026-07-02 — the Amazon–OpenAI equity legs entered the ledger).
· The convergence of indicators is weighted as corroboration, not four independent votes — they share a common driver (capex ahead of monetization).
· The self-measurement on the Instrument is n=1, illustrative — directional, not precision.

The exits

Falsifiers & revisions

The falsifier is built in: if the ground-truth signal turns back up — demand converting, the capex gap closing, insider selling normalizing — the divergence closes and the boom earns its price. We publish the number either way. The full ledger of what would prove us wrong, each with its current reading and status, is the Falsifier Watch — an append-only page that also records every revision we've made, dated. The most recent revisions reconciled the Amazon depreciation figure (the canary: a ~$1.4B actual-2025 depreciation step-up per the FY2025 10-K, plus a separate ~$920M one-time Q4 2024 charge per the FY2024 10-K), relabelled the divergence gauge as a short-series directional read, and corrected the reproducibility wording (Python pipeline, tables published at /data).

Read next: the live Fragility Index · Capex Watch · the founding data brief · Falsifier Watch.

Permalink · the canonical receipts →
← → or J / K walk the rack

The exits, published in advance · Falsifiers

LEDGER

Revision ledger

The exits, published in advance. What would prove the desk wrong.

Nine exits, published in advance. Status is a human ruling with an append-only revision trail; bound figures are live from the engine. 1 TRIGGERED. As of 2026-07-20.

The claimdirectioncurrent readingstatus
Organic demand is realKill condition: no-ROI falls below 60% for two consecutive quartersdisconfirmsMIT-NANDA ~95% no-ROIWatching2026-07-20
Capacity is used, not just builtKill condition: utilization >70% sustaineddisconfirmscluster utilization ~40–60% (reported) — the zone of maximum ambiguityWatching2026-07-20
AI lifts measured productivityKill condition: TFP turns durably positive at scaledisconfirmsAI-attributable TFP ~+0.07pp/yr (KC Fed / BLS) — not yet in the aggregatesNot triggered2026-07-20
Insiders believe their own storyKill condition: a cluster of open-market insider buying appearsdisconfirmsdiscretionary insider buying ≈ $0 — one buyer (Alphabet) in the AI coreWatching2026-07-20
The loop funds itself with real cashKill condition: ratio compresses toward 1.0× (genuine end-demand)disconfirmsfunded-cash recycling {recycling}× (live: 15.5×)Watching2026-07-20
AI revenue growth overtakes capex growth for several quartersKill condition: AI-revenue growth > capex growth, sustaineddisconfirmscapex still materially outpacing AI-cloud revenueNot triggered2026-07-20
Training silicon repurposes cleanly for inference at scale, extending real asset lifeKill condition: clean disclosed evidence of at-scale repurposingdisconfirmsasserted by hyperscalers; no clean disclosed evidence yetWatching2026-07-20
A hyperscaler shortens useful life citing AI obsolescenceKill condition: a filed useful-life REDUCTION naming AI/ML obsolescenceconfirmsAmazon cut servers 6→5yr (eff. 2025-01-01), ~$920M accelerated depreciation + ~$1.4B run-rate — the canaryTRIGGERED2025-02-07 · 1 revision
The insider tape prints a fifth consecutive up-quarterKill condition: Q3 2026 discretionary selling > $1.10B with zero AI-core open-market buys outside Alphabetconfirmsfour consecutive rises: $0.85B → $1.10B (Q3 2025 → Q2 2026)Watching2026-07-20
Permalink · the canonical page →
← → or J / K walk the rack