Resources / References / 02
The Teaser Period: Why the AI Boom Is Hitting a Reset Wall
E² review · 2026-09-25
Claims ledger
Each record separates the claim’s type, its support and the check actually performed. “Reported fact” describes an attributable statement; it does not certify the underlying event. Read the rubric.
Symbol key
Symbols supplement the words; they do not replace them.
Reported source Assumption Calculation Measurement
Source checked Arithmetic reproduced Conflict Check still open
Support uses one to five filled marks, from conjectural to established within scope. This is an ordinal evidence scale, not a probability. A dash means the scale does not apply.
- C01
Subprime defaults were the scheduled consequence of rate resets.
- Type
- Inference
- Support
- Conjectural
- Review
- Contradicted as labeled
A reset can increase payment stress, but it does not make default inevitable. Contemporary Fed testimony described defaults shortly after origination and lender failures before many resets. The article turns one risk mechanism into a sufficient historical explanation.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- Federal Reserve testimony on subprime mortgages — March 22, 2007; early payment defaults, underwriting and lender failures
What would change this assessment: Separate reset effects from underwriting, home prices, refinancing and borrower equity using loan-level evidence.
- C02
The $2.3T cloud backlog represents compute contracts that are not yet billing.
- Type
- Reported fact
- Support
- Conjectural
- Review
- Contradicted as labeled
RPO is unrecognized revenue, not a measure of unbilled AI capacity. Microsoft includes unearned revenue and future invoices. CoreWeave reports 15–25% weighted-average prepayments on active contracts. Oracle reports $75B of prepaid or customer-supplied hardware. These disclosures contradict the blanket description; this review does not independently reconcile the $2.3T aggregate.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- Microsoft 2025 annual report — Revenue recognition; remaining performance obligations; cash flow statement
- CoreWeave 2025 Form 10-K — Customer prepayments; committed contracts; liquidity and revenue recognition
- Oracle FY2026 earnings release — June 10, 2026; RPO growth and prepaid or customer-supplied hardware
What would change this assessment: Provide an entity-by-entity bridge separating AI from other revenue, billed from unbilled balances, and active from uncommenced contracts.
- C03
Oracle’s RPO grew 363% in fiscal 2026.
- Type
- Reported fact
- Support
- Supported
- Review
- Source checked
Oracle reported $638B of RPO, up 363% year over year at fiscal year-end. This verifies the company’s disclosure, not ultimate collection or the proportion that constitutes incremental AI cash demand.
Evidence and revision conditions
- Oracle FY2026 earnings release — June 10, 2026; RPO growth and prepaid or customer-supplied hardware
What would change this assessment: A corrected filing or a reconciliation changing the period or definition.
- C04
A hypothetical $12B contract starts billing after two years of construction.
- Type
- Model assumption
- Support
- Not applicable
- Review
- Source checked
The article explicitly labels this contract hypothetical. It illustrates timing risk. It does not establish typical prepayments, acceptance terms or construction delays across the market.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Replace the illustration with a documented contract or estimate how representative its terms are.
- C05
A 2025–26 signing boom plus 24–36 months mathematically guarantees a 2027–28 commencement boom.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Contradicted as labeled
The stated date ranges permit commencements through 2029. They do not determine where the peak falls. The original AWS agreement also provided immediate access, while Broadcom’s announced deployment extends through 2029.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- AWS–OpenAI announcement — November 3, 2025; $38B over seven years, immediate access, deployment target before end-2026
- OpenAI–Broadcom announcement — October 13, 2025; 10 GW, term sheet, rollout from second half of 2026 through 2029
Depends on C04.
What would change this assessment: Publish dated tranches and their weights; label the peak as a modeled concentration, not a mathematical guarantee.
- C06
Every take-or-pay compute contract is economically a lease and should be treated as senior debt.
- Type
- Inference
- Support
- Conjectural
- Review
- Unresolved interpretation
Fixed purchase commitments can create debt-like risk and deserve liquidity analysis. Lease accounting additionally requires control of an identified asset. An analytical debt adjustment does not establish legal seniority, cancellation rights or a dollar-for-dollar equity deduction without the contract terms.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- FASB review of Topic 842 — PDF page 44; identifying a lease requires control of an identified asset
What would change this assessment: Supply the contract-specific accounting, credit-adjustment and priority analysis; keep these distinct.
- C07
$470B of reported debt plus $1.66T of commitments gives $2.1T of system obligations.
- Type
- Calculated comparison
- Support
- Not applicable
- Review
- Not reproduced
The rounded sum is $2.13T. The published material does not provide a reproducible entity ledger, discount schedule or complete intercompany eliminations for this figure. Adding buyer commitments, seller financing and leases across a supply chain can count linked exposures repeatedly.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C06.
What would change this assessment: Release entity, counterparty, maturity, discount and consolidation schedules. State gross exposure separately from consolidated external obligations.
- C08
Two labs contribute more than half of a $2.3T cloud backlog when their contribution is roughly $1.0T.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Unresolved conflict
$1.0T divided by $2.3T is 43.5%, below half. The article later uses “nearly half,” which is compatible with these rounded inputs. “More than half” needs a different numerator or denominator.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C02.
What would change this assessment: Use a consistent date and scope, or correct the earlier description.
- C09
The largest deal announcements added $636B to four vendors’ market capitalization.
- Type
- Reported measurement
- Support
- Reported
- Review
- Not independently verified
No dated price-and-share-count event table was available in this edition. Even a reproduced change would not isolate the announcement’s causal effect from other news or market returns.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Provide announcement dates, return windows, shares outstanding and a market-adjusted comparison.
- C10
The appendix totals $1,170.4B of OpenAI headline value and $749.0B of modeled 2026–30 spend.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Arithmetic reproduced
The seven parent rows add to both totals. Campus rows are allocations within parent agreements, so they must not be added again. Matching the sum does not authenticate the underlying prices, scope or schedules.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Publish revised row inputs or executed terms; retain a reconciliation between headline value and annual spend.
- C11
The $1.2T OpenAI total consists entirely of signed dollar commitments.
- Type
- Inference
- Support
- Conjectural
- Review
- Contradicted as labeled
The appendix itself identifies estimated dollar values: Broadcom’s $350B and AMD’s $90B are capacity-based estimates. The primary announcements disclose capacity, not those dollar amounts. An announced deployment and an estimated price cannot establish an executed obligation of that size.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- OpenAI–Broadcom announcement — October 13, 2025; 10 GW, term sheet, rollout from second half of 2026 through 2029
- AMD–OpenAI announcement filed with the SEC — October 6, 2025; 6 GW deployment agreement and initial 1 GW in second half of 2026
What would change this assessment: Tag each row as disclosed dollars, reported dollars or capacity-priced estimate; disclose enforceable minimum purchases where available.
- C12
The Broadcom collaboration covers 10 GW with deployment targeted from 2026 through 2029.
- Type
- Reported fact
- Support
- Supported
- Review
- Source checked
The announcement supports the capacity and target period and identifies a signed term sheet. It supplies neither a $350B contract price nor a campus-level invoice schedule.
Evidence and revision conditions
- OpenAI–Broadcom announcement — October 13, 2025; 10 GW, term sheet, rollout from second half of 2026 through 2029
What would change this assessment: An executed agreement or updated deployment disclosure with dollar and timing terms.
- C13
OpenAI’s modeled annual compute costs total $750B across 2026–30.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Arithmetic reproduced
The chart labels are $37B, $132B, $169B, $201B and $211B. They sum to $750B, $1B above the appendix’s rounded $749B. This is an internal reconciliation of the author’s model, not a reconstruction from invoices.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C10.
What would change this assessment: Release unrounded vendor-by-year inputs and explain any bridge from $749B to $750B.
- C14
$712B of two-lab contract notional commences in 2027–28.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Arithmetic reproduced
The chart’s $352B and $360B add to $712B. The author explicitly distinguishes notional from annual cash payments. The date allocation remains modeled; reproducing addition does not establish commencement dates.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Provide contract-level tranche weights and evidence for each date.
- C15
Revenue must compound at 217% annually from 2025 to equal the modeled 2027 compute bill.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Arithmetic reproduced
Using the chart’s $13.1B 2025 revenue base and $132B 2027 cost gives 217.4% annual growth over two years. This reproduces the rounded requirement. Both inputs and the target year remain conditional.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C13.
What would change this assessment: Update either input or the horizon and recompute; independently verify the reported revenue base.
- C16
Modeled 2027 compute costs exceed 200% of management-plan revenue.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Arithmetic reproduced
$132B divided by the approximately $65B management-plan point is 203.1%, consistent with the printed ratio. This is a comparison between forecasts. It does not establish realized costs or revenue.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C13.
What would change this assessment: Supply the original numerical revenue series and dated management forecast; compare later actual results on the same basis.
- C17
The base case leaves $375B of uncovered compute cost in 2026–30.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Unresolved conflict
This does not reconcile with the retrieved revenue chart. Deliberately broad visual bounds imply $210–280B of signed cumulative cost less revenue, or $213–280B summing annual shortfalls only. These are chart-reading bounds, not precise source data. The chart bears a draft/redline label; a version mismatch is possible.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C13.
What would change this assessment: Publish the exact base revenue series, gap definition and figure version, and reconcile the $375B sentence.
- C18
$665B of commitments has a present value of $450–500B.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Not reproduced
The March 2026 private financial statements, maturity schedule and discount rate were not obtained. Many combinations can produce that range. The later $750B plan is a different input; substituting it would not reproduce this claim.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C06.
What would change this assessment: Provide the original commitment schedule, valuation date, discount convention and rate, with evidence for the $665B input.
- C19
OpenAI’s equity is effectively underwater because of its compute commitments.
- Type
- Inference
- Support
- Conjectural
- Review
- Not established
Gross future service payments do not by themselves determine equity value. A valuation also needs cash, other assets and liabilities, the services acquired and future net cash flows. Deducting obligations already included in operating forecasts would double-count their cost. None of this establishes that the equity is fairly priced.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Provide a consistent enterprise-to-equity valuation or liquidity model, including offsets, future cash flows and financing terms.
- C20
Bridgewater’s analysis shows OpenAI exhausting its latest fundraise by early 2027.
- Type
- Reported measurement
- Support
- Reported
- Review
- Not independently verified
The supplied chart crosses below zero in 2027 Q1 under no new capital. Its underlying Bridgewater report and cash-flow workbook were not retrieved. A forecast excluding financing is not an observed cash balance or an unconditional default date.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Publish the dated source, opening cash, operating flows, committed financing and subsequent capital assumptions.
- C21
Anthropic is structurally solvent because modeled compute costs peak near 60% of revenue.
- Type
- Inference
- Support
- Conjectural
- Review
- Not established
The chart prints a 61% peak. That leaves room for other costs but does not measure them, cash timing or financing needs. Two firms with different inputs are a comparison, not a controlled experiment. The narrow conclusion is better modeled compute coverage under the selected assumptions.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Add cash balances, all operating and financing flows, and sensitivity ranges for both firms on consistent terms.
- C22
The consolidated system has $2.4T of notional commencing across 2026–29.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Not reproduced
The chart labels itself consolidated and net of eliminations. The two-lab appendix is not a full reconciliation of the hyperscaler and neocloud layers. Without that ledger, this audit cannot verify the consolidation or the $2.4T total.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Publish gross rows, intercompany eliminations, capacity overlap and remaining external exposures by year.
- C23
Depreciation behaves like take-or-pay once a data center enters service.
- Type
- Inference
- Support
- Conjectural
- Review
- Unresolved interpretation
As a metaphor for fixed accounting expense, the comparison works: low utilization can depress operating profit. The section also explicitly says the asset was already paid for. Depreciation is not a new cash payment, however; its later inclusion among amounts “owed” should not feed a liquidity calculation. That stronger reading remains unresolved.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- Microsoft 2025 annual report — Revenue recognition; remaining performance obligations; cash flow statement
What would change this assessment: Separate accounting profit from cash flow, showing capex, depreciation, interest and principal without double-counting.
- C24
About 80% of a representative leveraged cluster’s monthly costs are fixed.
- Type
- Model assumption
- Support
- Not applicable
- Review
- Not independently verified
The article supplies no representative sample or cost workbook for the share. Fixed versus variable also depends on horizon, utilization and contract terms. Use 80% as a scenario input, not a measured industry constant.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Provide the cluster specification, cost denominator, time horizon and sensitivity to power and financing terms.
- C25
A 30% utilization decline can eliminate a leveraged facility’s operating profit.
- Type
- Calculated output
- Support
- Not applicable
- Review
- Arithmetic reproduced
The possibility is sound. An explicit illustration with revenue 120, fixed cost 80 and variable cost 20 starts at profit 20. Scaling revenue and variable cost down 30% yields a loss of 10. These illustrative units are ours; they do not replicate a measured facility or predict the typical result.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
Depends on C24.
What would change this assessment: For an actual facility, disclose baseline margin, pricing, utilization and which costs change with volume.
- C26
Renegotiation is the most likely response when the bills arrive.
- Type
- Forecast
- Support
- Conjectural
- Review
- Not yet resolvable
The article considers fundraising, renegotiation and subleasing, which are relevant alternatives. It gives no frequencies, comparison model or threshold establishing their ranking. Contract enforceability and creditor priority cannot be inferred from the take-or-pay label alone.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Define the population, period and observable outcome; compare funding, demand growth and renegotiation scenarios.
- C27
Short-term subleases will return capacity just as the 2027–28 deliveries create oversupply.
- Type
- Forecast
- Support
- Conjectural
- Review
- Not yet resolvable
Termination options permit capacity to return; they do not guarantee exercise, synchronized availability or a shortage of replacement tenants. The forecast needs a supply-demand model and a definition of oversupply.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Publish renewal assumptions, net capacity additions, utilization and realized rental prices; specify the forecast window.
- C28
AI-exposed companies constitute about 45% of the S&P 500.
- Type
- Reported measurement
- Support
- Reported
- Review
- Not independently verified
The article does not provide a dated constituent file or a reproducible AI-exposure classification. Index concentration is testable, but “AI-exposed” is an editorial category and does not mean all revenue depends on frontier-lab contracts.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Publish date, constituents, weights and an exposure rule; separate gross index weight from revenue at risk.
- C29
A teaser-period economy cannot produce bad credit data before the reset.
- Type
- Inference
- Support
- Conjectural
- Review
- Contradicted as labeled
The March 2007 testimony recorded early payment defaults and failures already underway. Deferred obligations can make current performance incomplete evidence; they do not make it information-free or rule out warning signals.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
- Federal Reserve testimony on subprime mortgages — March 22, 2007; early payment defaults, underwriting and lender failures
Depends on C01.
What would change this assessment: Replace the universal claim with a test of which early indicators predict later payment stress.
- C30
Demand can grow into supply if counterparties remain funded through the gap.
- Type
- Model assumption
- Support
- Not applicable
- Review
- Source checked
The closing section explicitly admits this bull case and says the wall is not a default forecast. This qualification matters when interpreting the stronger language elsewhere. It still needs measurable demand, cash and financing thresholds.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: State sufficient revenue and liquidity paths, review dates and observations that would weaken the financing-risk thesis.
- C31
The appendix separates headline value from spend and excludes several overlapping or cancellable commitments.
- Type
- Reported fact
- Support
- Supported
- Review
- Source checked
The appendix explains direct-payer silicon treatment, campus allocations and exclusions, including the NVIDIA letter of intent and terminable SpaceX capacity. These are useful safeguards. We verified what the appendix says, not every excluded contract’s terms.
Evidence and revision conditions
- Original article — Retrieved September 25, 2026, including the post-publication appendix; supplied PDF pages 1–39
What would change this assessment: Publish the same exclusions in a dated row-level dataset and link each to its original documentation.
Review history: September 25, 2026 — ledger and arithmetic audit. Original article unchanged; corrections remain open.