What looks bubbly
Broad-US CAPE is 41.1×, versus 44.2× at the dot-com extreme. Builder capex grew 84% while sales grew 19%.
ASSESSMENT · 2026-10-14
Bubble warning signs. No defensible crash clock.
Valuations are high and investment is outrunning sales. Demand is growing too. The question is whether future cash profits justify the prices and financial obligations being taken on today.
What pushes against a bust
These are established cash-generating firms. Five builders generate $707B operating cash against $586B investment. Paid AI spending is rising; technology adoption is real.
What could turn pressure into a crash
Disappointing sales or margins, idle capacity and refinancing stress. Aggregate cash coverage buys time; it does not protect each borrower or justify every stock price. Usefulness does not guarantee investor returns.
The historical cases below are selected crashes, not a probability model. They show similarities and differences; estimating crash odds also requires non-crash booms and out-of-sample validation. CAPE is broad-market context; company accounts cover all businesses.
01 · PRICE VS EARNINGS
What are investors paying for AI-exposed earnings?
How much price for $1 of annual earnings? Whole-company trailing P/E, September 4, 2026 close. These selected companies expose different parts of AI’s economics; they are not a complete sector or the Mag7.
Scenario, not an earnings forecast. Gray: trailing P/E baseline (≈ means independently calculated). Green: multiple under your earnings assumption. Common scale: 0–60×.
ASML ASML ↗≈53.6×
Chipmaking equipment · US GAAP. $1714.88 ordinary-share price ÷ €27.55 summed quarterly diluted earnings, converted at September 4 ECB reference FX ($1.1622/€), not closing FX. Reported results · FX ↗
TSMC TSM ↗≈30.9×
Foundry · TIFRS. $428.91 ADR price ÷ $13.86 summed quarterly diluted ADR earnings. Company quarterly currency translations; one ADR represents five shares, already included. Reported results ↗
Nvidia NVDA ↗29.1×
Investment gains were 13.3% of pre-tax profit: $30.6B of $229.7B, trailing year to July 26. These are not chip sales. An isolated after-tax EPS adjustment is unverified, so no gains-excluded P/E is shown. Annual + interim results ↗
Microsoft MSFT ↗27.8×
28.9× without OpenAI investment effects. Microsoft’s FY2026 adjustment lowers EPS from $17.95 to $17.28 at the same share price. It still includes other investment gains, including Anthropic. Company reconciliation ↗
Oracle ORCL ↗27.2×
Meta META ↗23.2×
Micron MU ↗22.9×
Amazon AMZN ↗20.8×
Alphabet GOOGL ↗17.0×
The denominator matters. Investment revaluations can boost reported earnings without more customer demand. The Microsoft comparison uses its disclosed after-tax adjustment; Nvidia’s pre-tax share is not the same measure. Both differ from the hypothetical green bars above. If sustainable earnings halve and price stays fixed, P/E doubles.
Stock Analysis supplies prices and unmarked multiples; those quotes were checked against displayed price/EPS. Marked estimates use the disclosed earnings and currency calculations. Provider discrepancies are preserved rather than silently relabeled. Earnings include non-AI operations. The scenario changes earnings only; it does not predict a price or crash. This is one year of earnings, unlike ten-year CAPE. No company-average “AI P/E” is calculated.
TSMC and ASML use different accounting and currency conventions; these approximate comparisons are not currency-neutral. Private labs have no verified comparable public-stock P/E here. Forward multiples are omitted because their forecast horizon and accounting basis have not been verified. Audit and definitions ↗
02 · SPENDING VS CASH
How hard is the buildout stretching its funders?
%
Micron · memory cycle 98.6 % · FY2024AI builders (5 firms) 47.4 % · 2024-12-31Microsoft + Intel · incumbents 26.2 % · FY1998Cisco · equipment supplier 15.0 % · FY1998Global Crossing · fiber operator No observation near this date
In the operating-cash view, above 100% means equipment spending exceeds operating cash, before dividends and debt repayments. Company ratios, with Global Crossing on a continuing-operations basis; annual dots versus AI trailing-year snapshots. AI starts January 2023, dot-com FY1996, Micron FY2022.
Cash is the dividing line. Global Crossing had $3.66B of sales in 2001 but used $0.91B continuing operating cash before $2.64B capex; bankruptcy followed. Micron’s trailing year to May 2026 generated $51.4B operating cash against $25.3B equipment spending—stronger funding, not proof that demand will last.
Coverage: one of the ten crash episodes has matched cash accounts here, plus today’s builders and Micron’s cycle. Global Crossing’s negative-cash and bankruptcy-era ratios are omitted, not zero. Customer commitments are not earned cash.
Periods, accounting differences and customer commitments
Same ratio, different cohorts. AI year 0 = January 2023; dot-com year 0 = FY1996. AI uses dated whole-company TTM snapshots; dot-com uses separately plotted annual supplier, operator and incumbent accounts. Micron starts in FY2022 and adds a trailing-year snapshot through May 28, 2026; it is a memory-cycle comparison, not an eleventh crash. Annual dots are not interpolated. Above 100% of operating cash needs reserves or financing, before dividends and debt repayments.
Revenue did not guarantee survival. Global Crossing spent $3.37B in 2000 against $0.55B continuing operating cash (613%). In 2001 it reported $3.66B sales but used $0.91B operating cash, before $2.64B capex. It filed for bankruptcy in January 2002, before many bonds’ original 2004–2009 maturities. Long-dated debt did not eliminate liquidity risk.
Global Crossing’s 2001 negative-cash ratio is undefined; 2002 bankruptcy-era cash coverage is omitted, not a recovery signal. Continuing operations exclude discontinued businesses and separately presented reorganization cash. Restated filing ↗. Cisco’s FY2002 $2.64B capex includes ~$1.9B buying formerly leased property. Cisco report ↗. Cisco is a supplier, Global Crossing an operator: their revenues are never added as independent final demand.
Matched investment/cash accounts exist here for 1 of 10 historical episodes. Other selected episodes have no verified series; they are not drawn. Accounts, periods and missing data ↗
Memory demand can reverse. Micron’s FY2023 equipment spending was $7.7B against $1.6B operating cash; the trailing year to May 2026 shows $25.3B against $51.4B. More cash now supports investment, but does not establish lasting demand. Operating margin moved from −37.0% to +65.6%; these are whole-company results, not AI-only profit. Gross cash equipment purchases exclude separate equipment-contract financing payments and are shown before grants and asset-sale receipts. 2023 · 2025 · latest statements ↗
Customer commitments are not cash already earned. Micron reported $422M of contract liabilities, mainly customer deposits, at May 28, alongside approximately $5B of minimum contracted revenue still to be earned. The separate $22B expected deposits and related financial commitments (about $18B in cash deposits) includes later agreements; it is not cash received or sales. The filing does not isolate deposits’ contribution to operating cash. Contract and cash-flow disclosures ↗
Oracle: cash buys time; the buildout still needs funding
As of May 31, 2026: $31.3B cash versus $7.2B principal due in FY2027. That covers scheduled principal alone. FY2026 cash capex exceeded operating cash by $23.7B; continuing such a gap consumes reserves or needs financing.
The $638B backlog is future revenue with costs to serve, not cash available for repayment. New leases commit another $260B; their annual payment schedule is undisclosed.
Debt maturities, backlog timing and lease obligations
| Period | Debt principal due | Expected backlog revenue (not cash receipts) |
|---|---|---|
| FY2027 | $7.2B | ~$76.6B |
| FY2028–29 | $15.6B | ~$216.9B |
| FY2030–31 | $17.0B | ~$216.9B |
| After FY2031 | $90.3B | ~$127.6B |
Contracted sales are not a repayment fund. The $638B backlog has costs to serve and uncertain cash timing. New leases, largely for data centers, also commit $260B over long terms; Oracle does not disclose their annual payment schedule in this filing.
Principal excludes interest, leases and other uses; this is not a liquidity forecast. Operating lease payments already affect operating cash, so subtracting them again would double-count. New leases generally start FY2027–29, lasting 15–19 years. June 22 filing ↗
03 · SUPPLY SPENDING → LATER REVENUE
Did sales follow the investment?
Today’s builders: $586B annual cash investment alongside $1853B whole-company annual revenue. This is not AI-only demand or project payback. Compare the historical paths below.
More sales can mean higher prices, not proportionally more usage. Micron’s memory revenue grew much faster than the quantity shipped:
| Memory | Sales growth | Average price growth | Bit-shipment growth |
|---|---|---|---|
| DRAM | +211% | ≈+140% | ≈+30% |
| NAND | +183% | ≈+130% | Low-20% range increase |
These are nine-month product-category figures, not the trailing-year totals plotted below. Price and volume effects combine multiplicatively; the reported rounded changes do not reconcile exactly. Memory bits shipped are not AI use or customer productivity. Company explanation ↗
% of initial revenue
Micron · memory cycle · spending 27.3 % of initial revenue · FY2024Micron · memory cycle · revenue 81.6 % of initial revenue · FY2024AI builders (5 firms) · spending 18.0 % of initial revenue · 2024-12-31AI builders (5 firms) · revenue 122.3 % of initial revenue · 2024-12-31Microsoft + Intel · incumbents · spending 14.3 % of initial revenue · FY1998Microsoft + Intel · incumbents · revenue 138.1 % of initial revenue · FY1998Cisco · equipment supplier · spending 5.1 % of initial revenue · FY1998Cisco · equipment supplier · revenue 100.0 % of initial revenue · FY1998Global Crossing · fiber operator · spending No observation near this dateGlobal Crossing · fiber operator · revenue No observation near this date
Filled = capital spending; dashed lines / hollow dots = revenue. Both lines use the same denominator within each cohort, so their levels remain comparable. Cohort baselines are their first available revenue observation, not an identical calendar year. With a lag, revenue is shifted back to the spending date it is compared with. Only dates with both observed earlier spending and later revenue are shown; no unmatched earlier investment is implied.
Micron sales include non-AI memory and are upstream of customer revenue: do not add them as new final demand. AI: five builders across all businesses; dot-com cohorts remain separate. Cisco ends in July; Global Crossing in December; Microsoft/Intel have different fiscal ends. Sales include existing businesses; AI-attributable revenue is not separately identified. Later revenue does not prove that earlier capex caused it, and revenue is not cash available for payback. TTM snapshots overlap and must not be summed. AI lags shift dated TTM snapshots; dot-com lags compare fiscal-year labels, not identical calendar periods. Micron annual lags match fiscal years despite 52/53-week date shifts; its latest TTM has no matched 12/24-month predecessor here and is omitted at those lags. Neither traces project cash flows. Micron starts September 1, 2022. Year 0: AI January 2023; dot-com FY1996. AI accounts ↗ · Historical statements ↗.
First-observation bases: Micron · memory cycle: FY2022, $30.76B · AI builders (5 firms): 2023-03-31, $1165.48B · Microsoft + Intel · incumbents: FY1996, $29.52B · Cisco · equipment supplier: FY1998, $8.49B · Global Crossing · fiber operator: FY2000, $3.50B. Global Crossing starts in FY2000 because earlier comparable restated rows are unavailable here. Its 2002 observation is after bankruptcy, with prepetition interest suspended; it is not normal profitability or a known-at-the-time forecast.
Why not ten supply-and-revenue lines?
Gold and Bitcoin are assets without corporate revenue. Housing is shown through residential investment and housing-service flows, not house resale turnover; imputed owner rent is not cash revenue. Matched investment, income and cohort definitions are missing for US railroads 1873, 1929, Japan and the China episodes. UK railways provide an additional infrastructure comparison. Inventing or substituting unrelated accounts would create a misleading ten-case comparison.
Revenue is growing. Is it covering costs?
Operating profit per $100 of revenue. Same calculation and axis for each business; choose reported periods. Below zero means sales did not cover operating costs.
SpaceX AI segment · Grok, X and infrastructure · quarterly-49.1%
Q2 2026 statement ↗ · $2.56B revenue; −$1.26B operating loss.
Cisco · Equipment supplier · annual FY ends July+17.1%
FY2000 statement ↗ · $18.93B revenue; +$3.23B operating profit.
Global Crossing · Fiber operator · annual FY ends December-37.6%
FY2000 statement ↗ · $3.50B revenue; −$1.32B operating loss.
-50%Vertical rule = zero profit+50%
One linear scale, resized for the selected observations. Quarterly and annual ratios share units, but cover different durations and business mixes. Historical statements include later restatements; this is not a prediction backtest.
Cisco’s FY2001 reversal: inventory and restructuring charges accompanied the bust. Later recovery also benefited from inventory reserves and the end of goodwill amortization.
Global Crossing’s FY2001 loss: includes $17.18B of asset impairment. Lower depreciation after that write-down flatters FY2002; bankruptcy-era improvement does not establish healthy demand.
Operating income includes reported charges and depreciation, before interest and tax. It is not free cash flow. These are selected businesses, not sector averages; neither a positive margin nor resemblance to a failed operator establishes crash odds.
AI coverage: Infrastructure services contributed $1.6B of year-over-year sales growth in SpaceX’s Q2 2026, not solely end-user tokens. Comparable OpenAI / Anthropic operating profit and free cash flow remain unverified here.
The economic reading
Closest useful analogy: infrastructure built ahead of uncertain returns. Railways and dot-com show that lasting customer value can coexist with investor losses. Today’s large incumbents have operating cash, but spending is absorbing more of it. The decisive missing link is AI-attributable cash profit—not token counts, resale turnover or total cloud-company revenue.
More paid use, durable margins and higher utilization would support the spending. Debt that matures before those cash flows arrive, or weak pricing despite heavy use, would strengthen the bust case. A price chart alone cannot settle either question.
AI valuation concentration · Mag7, adjacent stocks and earnings quality
How much of the S&P 500 is Mag7 or AI-adjacent?
Mag7: 31.2% of market value, 25.7% of expected earnings
Yardeni / LSEG, September 4, 2026. The same-chart shares imply Mag7 trades at about 21% more per dollar of forecast earnings than the index overall. Forecast earnings may disappoint; this is not a share of crash risk.
Mag7 market-value share31.2%
Mag7 forward-earnings share25.7%
Yardeni chart, Figure 2 ↗. Premium calculation: 31.2 ÷ 25.7 − 1. Analyst forward earnings, not realized cash; both Alphabet share classes included.
Ownership exposure beyond Mag7
Share of SPY portfolio value, September 3, 2026. An S&P 500 ETF proxy—not percentages of the index’s P/E.
Magnificent Seven34.0%
Selected AI-adjacent companies8.9%
Other holdings and cash57.1%
AI-adjacent earnings share is still unverified. The SPY holdings snapshot and Yardeni series have different dates and construction; do not combine their weights. A complete attribution needs consistent signed earnings for every constituent. Individual P/Es cannot be added or averaged into a “percent of P/E.”
A high-multiple example—and an earnings trap
Palantir: 149.1× trailing earnings at September 4 close. Selected example, not an all-index “worst” ranking. High P/E can reflect growth expectations or temporarily low earnings; a loss-making company has no meaningful positive P/E.
FactSet’s August 28 analysis highlights large investment gains in Alphabet and Amazon earnings. A lower reported P/E can therefore reflect asset revaluations rather than customer cash. Removing those gains correctly requires tax and earnings-basis reconciliation.
Group membership and calculation
Magnificent Seven: NVDA, AAPL, MSFT, AMZN, GOOGL, GOOG, META, TSLA.
Selected AI-adjacent companies: AVGO, MU, AMD, INTC, PLTR, LRCX, AMAT, ORCL, KLAC, ANET, DELL, SMCI.
Disjoint editorial groups; selected AI-adjacent firms are not an exhaustive or pure-AI universe. Alphabet’s two share classes belong to one company. ASML and TSMC are outside this index universe. Remaining fund weight stays in other/cash rather than being normalized away.
With consistent constituent earnings yields, group earnings contribution is Σ(weight × earnings/price); group P/E is group weight divided by that contribution. Include signed losses. Same-date constituent earnings and index methodology remain required before publishing a decomposition.
State Street full holdings workbook ↗Demand and financing detail · labs, paid spending, payback assumptions and Oracle
Supply gets built. Who pays for it?
Sources & boundaries ↗SUPPLY · BUILDERS BUY CAPACITY
Chips, data centers and power
$586B trailing cash capex by Microsoft, Alphabet, Amazon, Meta and Oracle. They purchase equipment and infrastructure; labs also contract cloud capacity.
All company businesses, not AI-only. Supplier sales are the other side of these purchases—not extra final demand.
MIDDLE · SELLING ACCESS
Cloud → labs and applications
Cloud operators sell compute to OpenAI, Anthropic, xAI, developers and businesses. Labs sell subscriptions and tokens; application firms package them into services.
OpenAI >$40B; Anthropic ~$65B annualized sales pace. August reports, not profit or earned full-year revenue. Different gross/net accounting prevents a comparable sum.
FINAL DEMAND · CUSTOMER BUDGETS
Businesses and paid consumers
Software and API budgets buy work: coding, support, analysis and content. Consumers buy subscriptions. Google and Meta also aim to improve advertising revenue through internal AI.
Productivity is a possible customer benefit, not supplier revenue. Benefits must sustain willingness to pay after competition and operating costs.
Payments flow back from customers through the chain. Firms can occupy multiple layers. Equity investment, borrowing and customer purchases are separate flows; never add all layers’ revenue together.
Paid demand: breadth versus dollars
Ramp businesses paying each vendor · July 2026
Anthropic43.5%
OpenAI39.7%
xAI4%
Ramp, published August 12 ↗. Selected business-payment sample; overlapping vendors, not revenue market share or measured productivity. A payment counts adoption; free usage is missed. Sample limits ↗
Spending depth: $11.95 per employee per month
Ramp’s reported median, July 2026: 2.6× July 2025’s $4.55. Spending per employee is growing in this sample; it does not establish supplier profitability.
2026-07 · $11.95 per employee per month
Ramp chart and data ↗ · published August 12; checked September 7. Includes subscriptions, coding agents, API tokens and GPU cloud spending. Not all US businesses, a same-firm retention measure or pure end-customer demand. Do not add it to lab or cloud revenue. A broader sample spending more would strengthen the demand case; weaker spending or margins would weaken payback.
For spending scale, Exponential View estimates $110B trailing end-customer AI sales and >$175B annualized in June after supply-chain deduplication. It is an analyst estimate, with a different boundary from global builder capex. It cannot establish a measured payback period.
What must the buildout earn to pay back?
Hypothetical investment cohort. Change the assumptions to see the required return.
Steady state: 5 overlapping equal vintages, each earning for 5 years. Not the first year's revenue requirement. No ramp-up or resale value. Assumed margin includes tax, working capital and upkeep, excludes the initial investment; do not deduct that investment twice. These are required cash flows, not observed AI returns.
Oracle borrowed while investing—and still built its cash reserve
$36.9B net borrowing versus $32.0B operating cash. FY2026, ended May 31. Cash investment exceeded operating cash by $23.7B, but financing and other flows also increased cash on hand.
Cash sources · $82.4B
Operating cash flow$32.0B
Investment sales less purchases$3.8B
Net borrowing cash flow$36.9B
Short-term capex financing$3.3B
Common and convertible preferred proceeds$6.4B
Cash uses · $62.0B
Cash capital investment$55.7B
Dividends and share repurchases$6.0B
Other financing$0.3B
Cash on hand: $10.8B → $31.3B. Opening cash + $82.4B sources − $62.0B uses + $0.096B currency effect. Bars share a $0–60B scale. Amounts are rounded; the underlying statement reconciles exactly.
Whole-company cash flows, not an AI funding allocation. Borrowing is net of debt repayments and commercial-paper flows; short-term capex financing is shown separately. Equity includes common and mandatory-convertible preferred proceeds. Operating cash already contains $4.6B of customer financing prepayments. Do not add those again or call the cash shortfall a debt-funded percentage of AI capex. June 10 cash-flow statement ↗ · dated research, checked September 7.
Financing is not demand: Nvidia’s >$500B third-party capital target is preliminary, not deployed funding. Alphabet’s $49.6B June stock/convertible proceeds are realized financing, not customer sales. Disclosures and scope ↗
Funding stress: useful technology can still miss its debt payments
Three questions drawn from Noah Smith’s December 2025 analysis: Does AI create value? Does cash arrive before debt is due? Who keeps the margin? This is our application of his published arguments, not his review or endorsement.
1. Demand is not producer profit
Ramp shows paying customers. Competition, lower token prices and upkeep can still leave insufficient cash to repay investors. Smith’s April 2026 update is more optimistic about coding use cases; it does not settle every company’s returns.
2. Look beyond the pooled cash surplus
Oracle FY2026: $32.0B operating cash − $55.7B cash capex = −$23.7B free cash flow. Operating cash covered 57.4% of cash investment; the 42.6% shortfall is not a measured debt-funded share.
Borrowing proceeds $46.1B; debt repayments $6.9B plus net commercial-paper repayment $2.3B. These company-wide flows also fund other uses. Actual cash-flow statement ↗
3. Follow creditors, prepayments and guarantees
Oracle’s $4.6B customer financing prepayments already enter operating cash. Counting them again as extra capex funding would double-count. Meta’s Hyperion venture also combines outside capital, leases and a conditional residual-value guarantee; ownership share alone does not describe risk.
FCF = operating cash − cash capex. FCF is what remains after investment, so “capex financed by FCF” can obscure the accounting. A true debt/equity/internal-cash funding split needs a matched sources-and-uses statement, not capex minus OCF alone. Financing and research evidence ↗
Who is exposed? Company cash coverage, chip supply and economic scale
Microsoft ↗63¢ capex / $1 operating cash12 months to 2026-06-30
Alphabet ↗71¢ capex / $1 operating cash12 months to 2026-06-30
Amazon ↗107¢ capex / $1 operating cash12 months to 2026-06-30
Meta ↗69¢ capex / $1 operating cash12 months to 2026-06-30
Oracle ↗161¢ capex / $1 operating cash12 months to 2026-08-31
Whole-company cash accounts. Above $1 requires reserves or external funding before distributions. Nvidia supplies equipment; its revenue does not establish customer returns.
Economic scale: 1.80%. These global builders’ capital spending is that large relative to annualized US nominal GDP (Q2 2026). This is a size comparison—not domestic AI investment or a contribution to GDP. BEA / FRED ↗
Follow the money upstream
Supplier sales confirm the buildout is happening. They do not prove the eventual AI customer earns a return. Do not add these sales to lab or cloud revenue.
TSMC · making the chips
$40.2B quarterly sales, up 33.7% in USD year over year (Q2 2026). Advanced nodes made up 77% of wafer revenue.
Recognized foundry sales across AI and non-AI applications. Watch whether capacity and customer orders keep pace with downstream usage.
Quarterly report ↗ASML · making the tools
€9.3B quarterly sales (Q2 2026). Management plans roughly 30% more low-NA EUV and immersion DUV capacity in 2027.
Sales are observed; capacity is a plan. More equipment capacity can ease bottlenecks while increasing overcapacity risk.
Results & capacity plan ↗Micron · memory
$13.8B cloud-memory sales, versus $3.4B a year earlier (fiscal Q3, ended May 2026).
Broader than AI or high-bandwidth memory. Price increases and product mix can drive revenue; this is not a measure of physical demand alone.
Quarterly results ↗Paid access is growing; profitability is still a separate question
ChatGPT business users
5M → >9M paying users
August 2025 to February 2026 disclosure snapshots. Purchased business access, not measured daily use.
Microsoft 365 Copilot
15M → >30M paid seats
January to July 2026 earnings-call disclosures. Licensed seats, not active users; no comparable absolute 2025 baseline verified.
Google processed tokens
>480T → >3.2 quadrillion / month
May 2025 to May 2026 disclosures. Mixed paid, free and internal workloads; changing token mix prevents a constant-quality demand comparison.
Company-reported, irregular snapshots reviewed September 7, 2026. Do not sum overlapping populations or multiply seats/tokens by list prices to infer revenue. These measures do not establish retention, cash margins, paid GPU occupancy or infrastructure payback. Falling usage per paid seat or weaker renewals would weaken the demand thesis; sustained paid expansion with better margins would strengthen it.
Data freshness · some feeds retained older data · editorial snapshots dated separately
Automatic sources refresh on visits when due. A successful check does not mean a new reporting period. The response may show cached data while a refresh runs.
Microsoft accounts2026-06-30Latest source verification: 2026-10-14 15:49 UTC
Alphabet accounts2026-06-30Latest source verification: 2026-10-14 15:49 UTC
Amazon accounts2026-06-30Latest source verification: 2026-10-14 15:49 UTC
Meta accounts2026-06-30Latest source verification: 2026-10-14 15:49 UTC
Oracle accounts2026-08-31Latest source verification: 2026-10-14 15:49 UTC
Treasury rates2026-10-09Latest source verification: 2026-10-14 15:49 UTC
Nvidia accounts2026-07-26Latest source verification: 2026-10-14 15:49 UTC
Accounts refresh last attempted 2026-10-14 15:49 UTC. Next eligible accounts check: 2026-10-14 21:49 UTC. These are eligibility times, not guaranteed scheduled ingestion.
- Market refresh unavailable: HTTP 520. Retaining dated estimate.
Editorial snapshots: CAPE observations through August 2026; stock P/E prices at September 4 close; lab revenue reports through August; supplier accounts through their labeled quarters; GDP Q2 2026. These do not update from a page visit. The hourly research task is paused; these snapshots require a separate source review.
Historical reconstructions and forecast models remain versioned research. Nasdaq research feed status ↗ reports its own last successful check and failures; it is not used as a headline crash prediction. Prospective forecast record →