Analytics Methodology & Data Policy

Semiconductor Cycle Economics uses peak-to-trough drawdown to compare each memory-cycle issuer against its own history. Cycle rows measure operating-margin swing in percentage points, revenue decline from peak to trough, quarters from peak to trough, and recovery speed from trough to the next selected peak.

SanDisk is treated with a three-part data basis: Form 10 carve-out annual statements for FY2022-FY2024, the first post-separation Form 10-K for FY2025, and current standalone public-company quarterlies after the February 21, 2025 separation.

WDC flash/SanDisk predecessor segment history uses pre-spinoff Western Digital 10-K filings as originally published for FY2024 and earlier, not later filings that reclassify the separated flash/SanDisk business as discontinued operations.

AI Power Translation measures disclosure precision: how specifically an issuer connects AI or data-center power demand to orders, backlog, revenue, or margin evidence in public filings and company-filed earnings materials.

The disclosure scale has four levels. Not mentioned means the reviewed period does not mention AI or data centers. Qualitative means AI or data centers are mentioned without a numeric order, backlog, revenue, or segment metric. Partially quantified means a numeric metric is tied to AI/data-center exposure, or broad company/segment order and backlog metrics are disclosed while AI/data-center demand is identified as a driver, but the issuer does not fully split AI/data-center economics. Fully quantified split means the issuer separately quantifies AI/data-center contribution to the relevant financial or backlog metric.

This project does not use valuation multiples or a single margin/backlog ratio as the core metric because power-equipment, grid, and facility-infrastructure economics move through long-cycle orders, backlog, capacity reservations, and project timing before they appear cleanly in revenue or margin. The first question is therefore whether the issuer has made the AI/data-center link observable at all.

Backlog and order growth are not attributed to AI by default. GE Vernova demand is labeled as broader power, electrification, grid, industrial, or services demand unless the filing or company-filed release directly ties the metric to data centers or AI. A missing AI/data-center disclosure is kept as a finding, not filled with inference.

GE Vernova separated from GE on April 2, 2024. FY2022 and FY2023 rows are treated as pre-spinoff combined/carve-out history, FY2024 is a transition-year consolidated and combined row, and FY2025 is the first full standalone public-company year in this dataset. Those bases are labeled in source notes and profile tables rather than smoothed into one uninterrupted standalone trend.

Vertiv is reused as an existing standalone issuer profile. Its data-center and AI-infrastructure language is often more direct than GE Vernova's, but its filing metrics remain issuer-level and do not isolate pure AI revenue, pure liquid-cooling economics, or data-center-only margin.

Metrics are collected from public issuer filings, primarily Form 10-K, Form 20-F, annual reports, or equivalent issuer disclosures.

Each row in the financial metrics dataset keeps sourceUrl, sourceType, sourceNote, currency, unit, fiscalYear, and periodEnded where available. Source notes document whether values are reported directly in the filing or computed from comparable line items.

Figures are presented as extracted and structured for analysis; this documentation does not claim perfect completeness or freedom from transcription error.

For the current analytics dataset, source mapping records, validation scripts, and chart reproduction files where available, see the Evidence & reproducibility package.

Operating Margin and CapEx Intensity are chosen because together they address one specific question: which companies convert AI infrastructure demand into operating profit, and at what direct capital cost.

Operating Margin measures how efficiently a company turns revenue into operating profit at its current asset base. CapEx Intensity measures what proportion of revenue must be reinvested in direct capital expenditure to sustain or expand that position. The combination maps a business model’s structural economic position—not just its financial performance at a single point in time.

Gross Margin captures product-level pricing power but does not reflect operating cost structure differences across business models. ROIC requires a full capital base reconstruction that is difficult to apply consistently across issuers with different accounting treatments, currencies, and acquisition histories. R&D Intensity is a third relevant dimension and is tracked separately in each Evidence Profile—it supplements but does not replace the capital burden axis in the core comparison.

These two metrics are not claimed to be exhaustive. They are chosen for comparability across the current eight-company issuer sample and to directly address the profit capture vs capital burden framing this module uses.

• Revenue: reported top-line sales, net sales, or revenue for the fiscal period, as labeled in the issuer filing.
• Gross profit: reported gross profit when disclosed; otherwise computed from comparable filing line items (for example, net sales minus cost of sales).
• Operating income: operating income or operating profit (or closest comparable filing subtotal) for the period.
• R&D expense: reported research and development expense, or engineering / research and development costs when separately disclosed.
• CapEx: capital expenditure or the closest comparable cash-flow line item; stored as a positive magnitude in the dataset.
• Revenue growth: year-over-year change in revenue for the same issuer and fiscal-year convention.
• Margins and intensities: ratios such as gross margin, operating margin, R&D intensity, and CapEx intensity are calculated from stored filing figures (for example, operating margin as operating income divided by revenue, when both are present).

The Semiconductor Cycle Economics project uses a different analytical shape from the latest-year AI Infrastructure comparison and the Cloud CapEx translation ratio. It compares each company against its own history across selected cycle peaks and troughs.

• Peak-to-trough drawdown: operating margin at a selected cycle peak minus operating margin at the following trough, expressed in percentage points. The same row also stores the revenue decline from peak fiscal year to trough fiscal year.
• Cycle boundary rule: peaks and troughs are selected from fiscal-year operating-margin local highs and following local lows. The current cycle is labeled in-progress when a new peak has not been established. Latest-quarter actuals may be shown after direct filing-table review, but they are labeled as quarters and are not annualized.
• Recovery speed: quarters from trough to the next selected peak when the cycle is resolved. For current-cycle rows, recovery speed is shown only as elapsed quarters from trough to the latest comparable annual period unless the row is explicitly labeled as a latest-quarter margin point.
• CapEx: capital expenditure uses purchases of property and equipment from the cash-flow statement, stored as a positive amount, matching the Cloud CapEx project convention.
• Inventory days: approximate inventory divided by revenue per day for the fiscal year. It is a directional cycle-pressure indicator, not a full inventory accounting reconstruction.

SanDisk is treated with a three-part data basis. First, the Form 10 package supplies audited combined carve-out annual statements only for FY2022, FY2023, and FY2024. Second, the FY2025 row comes from SanDisk's first post-separation Form 10-K and is labeled separately because the filing includes periods before February 21, 2025 that were derived from WDC combined records. Third, current-cycle quarterly rows after the separation are SanDisk standalone public-company filings. Historical flash rows before the separation are labeled as Western Digital Flash predecessor data and are not presented as standalone SanDisk Corp financials.

WDC flash/SanDisk predecessor segment history must use pre-spinoff Western Digital 10-K filings as originally published for FY2024 and earlier. Later WDC filings reclassify the separated flash/SanDisk business as discontinued operations, so those later restated presentations are not used to reconstruct pre-spinoff flash history. Original FY2020 and FY2021 WDC 10-Ks disclosed only Flash-based product revenue; Flash segment gross profit, operating income, and CapEx are unavailable on that original-filing basis. FY2022-FY2024 original WDC 10-Ks disclose Flash revenue and gross margin but still do not disclose Flash operating income or Flash segment CapEx.

SK Hynix and Samsung memory/division data are excluded from the core dataset because this analytics section uses SEC-filer comparability. SK Hynix's official newsroom can still be cited as context where it frames current results against the 2018 semiconductor super boom; that context is not a dataset row or chart point.

The review protocol is designed to reduce extraction and interpretation errors in filing-based issuer-level metrics. It is a consistency-check process, not an audit opinion and not an assertion that every stored figure is free from error. The module remains filing-based issuer-level analytics.

This is not an audit opinion.

The goal is to lower risk from transcription mistakes, unit mismatches, formula drift, fiscal-year mislabeling, and line-item selection errors—not to imply third-party assurance or audit-style attestation.

This validation summary documents how each issuer row is checked against filing sources, unit and currency rules, formula logic, and known interpretation risks. It is not an audit opinion and does not guarantee every stored figure is error-free.

All metrics are sourced from issuer 10-K and 20-F filings.

Raw values are recorded before ratio calculation to allow formula verification.

Known limits: consolidated metrics cannot isolate AI-specific revenue.

What this project clarified: how to read an income statement, and why the same AI exposure can mean very different things financially.

Revenue and other monetary amounts are displayed in each issuer's reporting currency.

No foreign-exchange conversion is applied unless a specific output explicitly states otherwise.

Fiscal year labels follow issuer reporting conventions.

Cross-company comparison in the active Profit Capture vs Capital Burden view relies primarily on ratios, not on ranking absolute revenue across currencies. The current comparison layer includes eight issuer-level companies.

Project fact: TSMC reports in TWD, ASML in EUR, and most U.S. issuers in USD—reflecting filing currency, not a consolidated FX policy.

• Evidence profile: a single-company, filing-based, issuer-level evidence page with preserved source trail and metric context.
• Comparison view: a selected subset of issuers compared on a shared metric set under explicit boundaries.
• A company can have a published evidence profile without being included in every CM Terminal Analytics comparison view.

The AI Infrastructure Economics project currently has eight published company evidence profiles. The active Profit Capture vs Capital Burden comparison is an eight-profile view across those issuer-level profiles.

The current AI Infrastructure Economics project uses a role-based anchor sample, not a complete investable universe or a complete AI infrastructure company universe. Each issuer is selected as a filing-accessible anchor for one economic role in the AI infrastructure value chain. The active comparison remains issuer-level.

The set is chosen for economic role coverage, not as an investment coverage list. The sample is fixed to issuers with filing-accessible, comparable issuer-level treatment under the current module rules.

This sample is designed to build an issuer-level economic map, not to claim complete coverage of the AI infrastructure universe.

AI exposure confidence is not a company-quality score, investment rating, or prediction of future returns. It measures how directly issuer-level disclosures connect reported financial metrics to AI infrastructure economics.

The confidence label does not mean higher or lower business quality. It only indicates how cleanly current filings allow the module to connect issuer-level metrics to AI infrastructure economics.

Current financial metrics are issuer-level unless a specific field or page explicitly states otherwise.

Issuer-level metrics describe business-model economics; they do not automatically isolate AI-specific revenue unless the issuer reports such a segment or line item.

Segment-level metrics are not the default interpretation layer in the current module.

AI exposure descriptions elsewhere in the hub are qualitative bridges from public filings and business descriptions—they are not precise AI revenue estimates unless the issuer discloses comparable figures.

Status labels used in this module:

1. Collected — data has been extracted from issuer filings and stored in the project dataset.
2. Reviewed — raw values, formulas, source links, and filing notes have been checked against source documents within the project's review process.
3. Comparison-ready — a reviewed issuer row is included in an active comparison view under the current comparison rules.

Status labels are module status terms, not external audit opinions.

• CM Terminal Analytics is descriptive business analytics.
• It does not provide investment advice.
• It does not recommend buying or selling securities.
• It does not estimate target prices or forward returns unless an output is explicitly labeled as a modeling exercise.
• It should not be read as a substitute for professional financial advice.

This module's evidence is bounded by the disclosure-vs-narrative method described in Research strategy; it does not claim parity with non-public information.

• Evidence library: eight published company profiles.
• Current comparison: eight-profile Profit Capture vs Capital Burden view.
• Out of scope in the current module: segment-level bridge tables, an AI CapEx transmission map (see Cloud CapEx Translation for the demand-origin side of that question), and additional infrastructure issuers beyond the current eight-profile set.

The Cloud CapEx Translation module is a separate analytics project that asks whether hyperscaler capital expenditure is translating into cloud revenue — or accumulating as future depreciation. It is not part of the AI Infrastructure Economics project; it has its own scope, sample, and filing provenance.

• Sample: Microsoft (Intelligent Cloud), Amazon (AWS), Alphabet (Google Cloud), and Meta Platforms (no external cloud segment — contrast case).
• Annual series: FY2021–FY2025 from Form 10-K filings. Interim signals updated from 10-Q filings when material new data is available.
• Update cadence: annual data updated upon 10-K filing (typically February–August). Interim signals refreshed on a best-effort basis from quarterly 10-Q filings.
• Known distortion: Microsoft restructured its segment reporting in FY2025, restating prior-year Intelligent Cloud figures. All years use the restated basis for comparability.
• Out of scope: issuer-level profit-capture comparisons for hyperscalers (AI Infrastructure Economics project), and hyperscaler supply-chain transmission to vendors (also AI Infrastructure Economics).

This is descriptive business analytics based on public filings. It does not provide investment advice or valuation targets.