Methodology

SemiFlow tracks semiconductor supply chain health by extracting financial metrics from public filings and classifying each company's supply-demand condition. This page explains the data sources, scoring model, and analytical framework.

Data Sources

All financial data comes from public regulatory filings and market data feeds. No proprietary data, estimates, or analyst consensus figures are used.

SEC EDGAR (14 US-listed companies)

Quarterly 10-Q and annual 10-K filings pulled via the EDGAR XBRL CompanyFacts API. Revenue, cost of revenue, gross profit, inventory, capital expenditures, and shares outstanding are extracted from standardized US-GAAP tags. Filing dates and period end dates come directly from SEC submission metadata.

Yahoo Finance (4 companies)

TSMC, ASML, Samsung Electronics, and SK Hynix are sourced via Yahoo Finance quarterly financials. TSMC reports in TWD and is converted to USD using quarterly exchange rates. ASML uses its NASDAQ-listed ticker which reports in USD. Korean-listed companies (Samsung, SK Hynix) report in KRW; exchange rates are fetched quarterly to convert to USD equivalents.

Filing Types

10-Q (US quarterly), 10-K (US annual), 6-K (foreign interim reports like ARM Holdings). Yahoo Finance companies (TSMC, ASML, Samsung, SK Hynix) use quarterly earnings data directly. The pipeline targets quarterly data for all 18 companies.

The pipeline runs weekly. During earnings season (roughly weeks 4 through 7 of each calendar quarter), new filings arrive over a 2-3 week window. Between earnings seasons, the data is unchanged.

Epoch AI (Compute Demand Tab)

Physical chip deployment estimates from Epoch AI, licensed under CC-BY. Covers AI accelerator shipments by vendor (NVIDIA, Google, AMD, Amazon, Huawei) with quarterly granularity. All figures are industry estimates, not audited financials. The Compute tab clearly labels this data as "ESTIMATE DATA" to distinguish it from GAAP-sourced metrics on the Flow and Data tabs.

The Compute tab's cross-reference panel combines Epoch's physical shipment estimates with SemiFlow's GAAP financial data to detect divergences between physical unit volume and reported revenue. This analysis is only shown for companies where Epoch directly tracks shipments. Upstream companies (foundries, equipment, memory) are excluded to avoid misleading implied proxies.

Supply Chain Tiers

The 18 tracked companies span 5 tiers of the semiconductor supply chain. The tier structure reflects how signals propagate through the ecosystem: design tool demand leads chip production, which drives equipment orders. Watching all tiers simultaneously reveals supply-demand imbalances that single-company analysis would miss.

EDA Tools Chip design software. Rising bookings signal more chips entering the pipeline. SNPS, CDNS

Designers Fabless and IDM companies designing GPUs, AI accelerators, networking silicon. NVDA, AVGO, AMD, QCOM, ARM, MRVL, ON, INTC

Foundry Manufacture the chips. Utilization and node mix reveal bottlenecks. TSM

Memory DRAM, HBM, NAND. Pricing cycles and inventory write-downs track the cycle. MU, SSNG, SKHX

Equipment Lithography, deposition, etch. Order backlogs reflect future capacity investment. ASML, AMAT, LRCX, KLAC

Intel is classified in the Designers tier despite operating fabs. The majority of Intel's revenue comes from its product divisions (Client Computing, Data Center, Network & Edge), not from Intel Foundry Services (IFS), which has minimal external revenue. Its consolidated financials reflect chip design economics more than foundry economics. CapEx intensity will be higher than fabless peers due to its IDM structure.

Samsung Electronics (SSNG) is placed in the Memory tier as a proxy for its Device Solutions semiconductor division. However, Samsung's publicly available financials are consolidated across all business units including mobile, display, and appliances. This dilutes Memory tier averages. SemiFlow notes this limitation and plans to incorporate segment-level data as it becomes available.

Financial Metrics

Each company card displays metrics chosen for their supply chain signaling value, not just headline financial performance.

Gross Margin

Reflects pricing power and cost structure. Rising margins suggest demand exceeding supply. Falling margins signal overcapacity or pricing pressure.

Gross Margin QoQ

Quarter-over-quarter change in percentage points. The direction matters more than the absolute level, since each tier has structurally different margin ranges.

Inventory Days

Days of inventory on hand, calculated as (Inventory ÷ COGS) × days in the reporting period. This uses single-quarter COGS, which may differ from company self-reported figures that sometimes use annualized or trailing-average COGS. Applying one consistent formula across all 18 companies is more valuable for cross-company comparison than matching each company's own reported number. Below 90 suggests lean conditions. Above 150 signals potential oversupply.

Revenue QoQ / YoY

Sequential and year-over-year revenue growth. Distinguishes between structural demand acceleration and normal seasonal patterns.

CapEx

Capital expenditures in absolute terms. High CapEx signals capacity expansion commitments. CapEx intensity (CapEx / Revenue) shows how aggressively a company is investing relative to its size.

Sparkline Trend

Up to 8 quarters of historical gross margin or inventory days, displayed as bar charts. Shows the trajectory, not just the current snapshot.

EDA companies (Synopsys, Cadence) typically do not carry physical inventory, so inventory days is shown as N/A. ARM Holdings, as a pure IP licensing business, also has no inventory.

Supply State Classification

Each company is classified as Tight, Balanced, or Glut based on a two-axis scoring model. The model evaluates tightness signals and glut signals independently, each scored 0 to 5.

Tight

Balanced

Glut

Signal	Axis
Inventory days below normal	Tightness
Inventory days falling QoQ	Tightness
Gross margin rising QoQ	Tightness
Backlog elevated or growing	Tightness
Inventory days above normal	Glut
Inventory days rising QoQ	Glut
Gross margin falling QoQ	Glut
Backlog shrinking or soft	Glut

Classification uses the relative gap between the two scores. A company with strong tightness signals and no glut signals is classified Tight with high confidence. Mixed signals result in a Balanced classification. Specific score thresholds and cutoff logic are proprietary.

Compute Demand Analysis

The Compute tab (Pro) provides a second analytical lens by overlaying physical chip deployment data from Epoch AI onto the financial data from EDGAR. This enables three types of analysis not available from financial filings alone.

Direction Alignment

Compares whether physical shipment trends and GAAP revenue are moving in the same direction. Agreement confirms the signal. Disagreement flags potential ASP shifts, deferred revenue, or data lags.

Magnitude Analysis

Even when direction agrees, extreme gaps between volume growth and revenue growth reveal pricing dynamics. Volume outpacing revenue suggests discounting. Revenue outpacing volume suggests pricing power.

Operational Classification

Combines shipment volume direction with inventory days movement to classify the operational state (e.g., demand exceeding supply, strategic pre-build, or inventory accumulation risk).

Ecosystem Attach

For selected vendors, tracks how total revenue per chip shipped changes over time. Rising values indicate expanding ecosystem revenue (networking, software, systems) per unit of silicon deployed.

Epoch AI data uses H100-equivalent as a normalization unit, converting all chip architectures to a common compute baseline based on peak INT8 throughput. This allows cross-vendor comparison but is an approximation, not a precise benchmark. All Compute tab metrics carry an "ESTIMATE DATA" label and should not be treated as GAAP-grade figures.

Specific classification thresholds and scoring weights used in the cross-reference engine are proprietary to SemiFlow and are not disclosed.

Currency Conversion

Three companies report in non-USD currencies: TSMC (TWD), Samsung (KRW), and SK Hynix (KRW). ASML uses its NASDAQ listing which reports in USD. For aggregated views (tier totals, market cap), the pipeline converts foreign currencies to USD using quarterly average exchange rates fetched from Yahoo Finance. Each company's card shows revenue and inventory in the reporting currency of its data source.

Leading Indicators

Quarterly SEC filings arrive 30 to 45 days after the quarter ends. To fill that gap, SemiFlow tracks three monthly data sources that provide real-time demand and trade signals between earnings seasons.

TSMC Monthly Revenue

TSMC publishes unaudited consolidated revenue approximately 10 days after each month ends. Reported in NT$ (Taiwan dollars) and converted to approximate USD using the prevailing quarterly FX rate. Because TSMC manufactures chips for the majority of the tracked Design tier companies, its monthly revenue is the single best high-frequency signal for overall semiconductor demand. Sourced from TSMC Investor Relations.

South Korea Semiconductor Exports

Korea Customs Service publishes monthly semiconductor export data, with preliminary first-20-days figures often available before month-end. South Korea accounts for roughly 20% of global semiconductor exports, driven by Samsung and SK Hynix. Rising exports signal strengthening memory demand; declining exports often lead inventory corrections by one to two quarters. Relevant to the Memory tier (MU, SKHX, SSNG). Sourced from Korea Customs Service.

Taiwan Export Orders (Electronics & ICT)

Taiwan's Ministry of Finance publishes monthly export order data around the 20th of the following month. The electronics and ICT categories capture orders placed with Taiwanese manufacturers, including TSMC and its packaging partners. Because orders lead shipments by one to three months, this series provides earlier visibility than TSMC's own revenue figures. Relevant to the Design tier and Foundry tier. Sourced from Taiwan MOEA.

Leading indicator data is currently entered manually from official government and corporate sources, typically requiring about 15 minutes of updates once a month. The free tier shows the latest values. The paid tier includes historical trends, year-over-year comparisons, and trend direction.

Hyperscaler Capital Expenditure

The paid tier includes an aggregate view of capital expenditures from five major cloud and AI infrastructure companies: Microsoft, Alphabet (Google), Amazon, Meta, and Oracle. These companies are the largest buyers of AI accelerators, networking equipment, and custom silicon, making their combined capex the most direct demand signal for the upstream semiconductor supply chain.

Capex figures are extracted from SEC 10-Q and 10-K filings via EDGAR XBRL, the same pipeline used for semiconductor companies. Because these five companies have different fiscal year ends (Microsoft in June, Oracle in May, others in December), the dashboard aligns all figures to calendar quarters so they can be compared and summed. A quarter is only displayed when at least four of the five companies have reported, to avoid misleading totals from incomplete coverage.

Limitations

SemiFlow uses GAAP figures exclusively. Non-GAAP adjustments (stock-based compensation exclusions, restructuring charges, acquisition-related amortization) are not applied. This means gross margins may differ from the figures companies highlight in their earnings press releases, which typically emphasize non-GAAP metrics.

Quarterly filings arrive 30 to 45 days after the quarter ends. During this lag window, the dashboard shows prior-quarter data. Yahoo Finance sourced companies may have a similar lag before quarterly data is fully populated.

Inventory days are calculated using single-quarter COGS applied uniformly across all companies. This may produce figures that differ slightly from company self-reported inventory days, which sometimes use annualized or trailing-average cost denominators. The consistency of applying one formula across the full coverage set is prioritized over matching individual company disclosures.

The scoring model is designed for five input signals (inventory level, inventory trend, margin trend, backlog status, and utilization rate). In practice, backlog and utilization data are rarely available in structured filings. Most companies are scored on three of the five possible signals. The classification threshold accounts for this by requiring a lower minimum score (2.5 out of 5) to trigger a tight or glut state.

The scoring model does not incorporate qualitative factors such as earnings call commentary, guidance revisions, product launch timing, or geopolitical risk. It is a quantitative screen based entirely on disclosed financial data.

Epoch AI updates its chip sales dataset periodically, not on a fixed schedule. Some vendors may have incomplete data for the most recent quarter. When this occurs, the Compute tab displays "Awaiting data" rather than showing potentially misleading zero values. Share percentages in partial quarters reflect only reported vendors and are marked accordingly.

See the supply chain in action.

View the Dashboard