Semiconductor foundry model
Wafer capacity, utilisation, yield and fab depreciation. Build a foundry from its fabs: capacity by node added a year after the spending, utilisation through a cycle, revenue as wafers times price per wafer, yield as a cost, fixed and variable cost of revenue, a vintage-by-vintage depreciation waterfall, and free cash flow valued in a DCF, then read what a downturn year does to gross margin.
By Surojit Chakraverti, founder of L3VLUP and an investor running a long-short healthcare and technology equities strategy.
On this subject: semiconductor foundry model.Updated 4 November 2026
Who builds it, and for whatThe model a semiconductor equity research analyst keeps on a foundry and rebuilds every quarter from the wafer shipments, utilisation, node mix and capital spending guidance the company reports, and that a technology banker or a hardware investor builds before taking a view on a fab, a capacity deal or the cycle. The questions it answers are the ones the sector argues about: how far margin falls when utilisation does, how much of the price rise is mix, and whether this year’s capital spending earns its depreciation.
Every sheet below, in your browser, with its outputs and checks
Starter workbook free with an account. Worked model: one a month with Plus, every one with Pro. Downloads
| A | B | C | D | E | F | G | |
|---|---|---|---|---|---|---|---|
| 1 | Assumptions | ||||||
| 2 | $m unless stated. Company inputs in column C; node inputs in the three node columns. Wafers are 12-inch equivalents. All figures are invented. | ||||||
| 4 | Input | Unit | Advanced | Mainstream | Mature | Notes | |
| 5 | Scenario and timing | ||||||
| 6 | Scenario | # | 1 | 1 = base, 2 = bear, 3 = bull. Selects the utilisation swing, price changes and capacity spending on the Scenarios sheet. | |||
| 7 | Time from capital spending to wafers out | years | 1 | Tools bought in a year are installed and qualified, then produce and are depreciated from the next. Whole years, 1 to 7. | |||
| 8 | Fab equipment depreciation life | years | 5 | Straight line from the first year the tools produce. Lag plus life at most 7, the reported years laid out. | |||
| 10 | Costs and operating expenses | ||||||
| 11 | Cost inflation: materials, labour and overhead | % | 2.5% | ||||
| 12 | Power price growth | % | 4.0% | ||||
| 13 | R&D | % | 8.5% | % of revenue: process development for the next node | |||
| 14 | SG&A | % | 3.5% | % of revenue | |||
| 15 | Tax rate on EBIT | % | 15.0% | An effective rate after investment incentives; floored at zero in a loss year | |||
| 16 | Maintenance and technology spending | % | 9.0% | % of revenue: tool upgrades, R&D lines and the rest of the fab, adding no wafer capacity | |||
| 17 | Net working capital | % | 12.0% | % of revenue: receivables and wafers in process, less payables | |||
| 19 | Valuation | ||||||
| 20 | Cost of capital (WACC) | % | 9.5% | ||||
| 21 | Terminal growth | % | 2.5% | Of free cash flow after FY7E | |||
| 22 | EV / EBITDA for the cross-check | x | 7.5x | Applied to FY1E EBITDA | |||
| 23 | Net cash at the end of FY0A | $m | 3,000.0 | ||||
| 24 | Shares in issue | m shares | 2,000.0 | ||||
| 26 | Capacity and utilisation by node | ||||||
| 27 | Node group | Leading edge and the node before | Mid-generation FinFET and planar | Older planar, specialty and analog | |||
| 28 | Capacity, FY0A | '000 wafers a year | 360 | 720 | 1800 | ||
| 29 | Utilisation, FY0A | % | 94.0% | 86.0% | 80.0% | Wafers shipped / capacity | |
| 30 | Normal utilisation, forecast | % | 95.0% | 87.0% | 82.0% | Before the cycle; never above 100% | |
| 31 | Sensitivity to the cycle | x | 0.60 | 1.00 | 1.30 | Times the year’s utilisation swing on the Scenarios sheet: mature fabs feel a downturn first | |
| 32 | Capital cost of new capacity, FY0A | $m per '000 a year | 30.00 | 8.00 | 4.00 | Spending for a thousand wafers a year of capacity | |
| 33 | Capital cost of new capacity, rise a year | % | 6.0% | 2.0% | 2.0% | Each new node needs more tools and steps per wafer | |
| 35 | Price per wafer | ||||||
| 36 | Price per wafer, FY0A | $ per wafer | 17500 | 8200 | 2900 | Blended across the nodes in the group | |
| 37 | Price change a year | % | 3.0% | 0.0% | -2.0% | Before the scenario’s change, which applies to every node | |
| 39 | Yield and cost per wafer | ||||||
| 40 | Line yield, FY0A | % | 86.0% | 96.0% | 98.0% | Wafers shipped / wafers started | |
| 41 | Yield improvement a year | % | 3.0% | 0.5% | 0.0% | Percentage points, up to the ceiling | |
| 42 | Yield ceiling | % | 95.0% | 97.5% | 98.0% | ||
| 43 | Materials and chemicals per wafer started, FY0A | $ per wafer | 2600 | 1300 | 600 | Wafers, gases, photoresist, slurries and masks; grows with cost inflation | |
| 44 | Power per wafer started, FY0A | $ per wafer | 900 | 500 | 250 | Grows with the power price | |
| 45 | Fab labour and overhead, FY0A | $m per '000 a year | 1.60 | 0.90 | 0.45 | Per thousand wafers a year of capacity, full or not: fixed | |
Select a cell to see its row, its schedule and the type of cell. Font colours: blue for inputs, black for formulas, green for links to other sheets. Scroll sideways to see every period.
Download
Semiconductor foundry model: the workbook
An Excel workbook with working formulas, no macros and no external links: the same model you can work with on this page.
- Browse without an accountPreview every model online, sheet by sheet, with its outputs and checks.
- Free accountWork with every model online: change the inputs, switch the case, see every formula. And every starter workbook, with its key schedules left blank to build.
- L3VLUP Plus$4.99/monthEverything in Free, and one worked model each billing month as an Excel file: every schedule built, formulas live, checks at zero, yours to keep.
- L3VLUP Pro$25/monthEverything in Plus, every worked model in the library, and the Pro applications.
- The worked model
- 12 sheets and 1,145 live formulas, all 7 schedules built and linked, every check reading zero. Change an input in Excel and the whole model moves.
- The starter
- The same inputs and layout with the capex and depreciation (PP&E roll-forward) left blank for you to build, each blank cell marked. The worked model is its answer.
What the base case says
- Revenue, FY1E
- $17,632.9m
- Gross margin, FY2E
- 39.2%
- Gross margin points per point of utilisation, FY2E
- 0.41
- Fixed cost, % of revenue, FY2E
- 36.0%
- Revenue growth a year, FY0A to FY7E
- 11.1%
- Advanced, % of revenue, FY7E
- 64.4%
- Capital intensity, FY1E
- 28.3%
- Free cash flow, FY1E to FY7E
- $32,224.3m
- Enterprise value, DCF
- $72,739.3m
- Value per share, DCF
- 37.87
- Value per share, EV / EBITDA
- 34.56
Read from the workbook as served, every input at its default. Periods: FY0A, FY1E, FY2E, FY3E, FY4E, FY5E, FY6E, FY7E. The figures are invented and move with whatever you type in.
What this model is
An operating and valuation model of a fictional pure-play foundry: three node groups (advanced, mainstream and mature), their wafer capacity, utilisation, price per wafer and cost, seven forecast years after a reported FY0A, and a DCF of the free cash flow. Every figure is invented.
Capacity is bought: spending on each node adds wafers out a year later, at that node’s capital cost, and is depreciated over five years from then. The Capex sheet lays out seven reported years of spending so that the depreciation waterfall carries every vintage still being written off.
Most of a fab’s cost is fixed. Depreciation, labour and overhead are the same whether the fab is full or idle; only materials and power follow the wafers. So gross profit falls further than revenue when the fab empties: ten points of utilisation are worth about four points of gross margin in the base case, which the Cost of Revenue sheet and the gross margin grid show.
Set Scenario on the Assumptions sheet to 2 (bear) for a downturn in FY2E, with mature fabs hit hardest, price cuts and spending reduced, or to 3 (bull).
Seats: Equity research and hedge funds, Investment banking, Private equity.
Careers that do this work: growth equity, venture capital, equity research, corp dev / m&a, bizops / strategy & ops.
How the schedules connect
Every row in the workbook is tagged with the schedule it belongs to; the inspector above shows the tag when you click a row. These are the schedules this model is made of and where each one lives.
Wafer capacity, utilisation and shipments
Fab capacity by node, the spending lag and the cycle
Assumptions, row 7 · Assumptions, row 16 · Assumptions, rows 27–33 · Scenarios, rows 8–11 · Scenarios, rows 20–23 · Scenarios, rows 26–28 · Capacity, rows 6–10 · Capacity, rows 13–17 · Capacity, rows 20–24 · Capacity, rows 27–32 · Capex, rows 8–14
Revenue build
Where the top line comes from
Assumptions, rows 36–37 · Scenarios, rows 14–17 · Revenue, rows 6–14 · Revenue, rows 17–22
Fab cost structure: yield, fixed costs and margin
Variable and fixed cost of revenue, and operating leverage
Assumptions, rows 11–12 · Assumptions, row 17 · Assumptions, rows 40–45 · Cost of Revenue, rows 6–7 · Cost of Revenue, rows 10–15 · Cost of Revenue, rows 18–23 · Cost of Revenue, rows 26–31 · Cost of Revenue, rows 34–49
Capex and depreciation (PP&E roll-forward)
Opening, plus capex, less depreciation, closing
Assumptions, row 8 · Capex, rows 17–33
Operating costs and margins
Cost of sales, opex, EBITDA
Assumptions, rows 13–15 · Profit and Cash Flow, rows 6–16
Discounting and terminal value
Discount factors, two terminal methods, reconciled
Assumptions, rows 20–24 · Profit and Cash Flow, rows 19–25 · Valuation, rows 6–18 · Valuation, rows 21–22
Sensitivity tables
Two assumptions at once, without a data table
Sensitivity, rows 5–8 · Sensitivity, rows 11–16 · Sensitivity, rows 20–28 · Sensitivity, rows 31–43
What you should be able to explain
- Why a foundry’s gross profit moves further than its revenue when utilisation falls, and how to measure it in points of margin per point of utilisation.
- Why most of a fab’s cost of revenue is fixed: depreciation, labour and overhead are paid whether the fab is full or idle.
- How capital spending becomes capacity a year later and depreciation for five years after that, and why today’s depreciation is the spending of the last five years.
- Why the blended price per wafer rises even where no node’s price does, and how to separate mix from price.
- Why yield is a cost rather than a revenue line under wafer pricing, and what a node’s yield ramp is worth.
- Why capital intensity is the number that decides a foundry’s free cash flow, and how much of the valuation sits in the terminal value.
What a reviewer looks for
- Cost of revenue as a fixed percentage of revenue, so gross margin holds steady through a downturn that would take ten points off it.
- Depreciation as a share of revenue or of opening assets, rather than built from the spending of each year it comes from.
- Capacity that arrives the year the money is spent, with no lag for installing and qualifying the tools.
- One blended price per wafer, which hides that most of the growth in price is a shift towards the advanced nodes.
- A terminal year whose capital spending is far below the depreciation it implies, so the perpetuity values a fab that never replaces its tools.
Conventions this workbook uses
Stated on the cover sheet too. A model is only as trustworthy as the decisions it tells you it made.
- Annual periods. FY0A is the last reported year and the valuation date is its end. Wafers are 12-inch equivalents, in thousands; prices and costs per wafer are in dollars.
- Capacity added in a year is the node’s capacity spending a lag earlier divided by its capital cost per thousand wafers a year, and is in service for the whole year. Utilisation is wafers shipped over capacity: the node’s normal level plus its cycle sensitivity times the year’s swing, held between 0% and 100%.
- Revenue is wafers shipped times the node’s price per wafer, which changes each year by the node’s own rate plus the case’s change for every node. The blended price at FY0A prices isolates the effect of mix.
- Yield is a cost: wafers started are wafers shipped divided by line yield, and materials, chemicals and power are paid on every wafer started. Under wafer pricing the customer bears die yield; line yield is the foundry’s. Fab labour and overhead are a cost per unit of capacity, and with depreciation make up the fixed cost of revenue.
- Depreciation is straight line over the equipment life, from the first year a vintage’s tools produce (a lag after the spending) to the end of its life. Maintenance and technology spending is a share of revenue and is depreciated the same way; it adds no capacity.
- R&D and SG&A are shares of revenue. Tax is charged on EBIT at the effective rate, floored at zero, with no losses carried forward. Working capital is a share of revenue. No debt is modelled: free cash flow is unlevered and net cash at FY0A is added to enterprise value.
- The DCF discounts each year’s free cash flow from its year end at WACC and adds a perpetuity on FY7E free cash flow. The EV / EBITDA cross-check applies a multiple to FY1E EBITDA. The grids hold capacity, yield and fixed costs as in the selected case.
What this model does not cover
Where its answer stops, so you know what to add before relying on it.
- Debt, interest and a balance sheet. Free cash flow is unlevered and net cash is added at the valuation date.
- Fab shells and buildings apart from tools. Every dollar of spending is depreciated over the same equipment life.
- Die yield and pricing per good die. Under wafer pricing the customer bears it; a foundry that guarantees die yield on a new node carries a risk this model does not.
- Customer concentration, take-or-pay commitments and prepayments for capacity.
- Government subsidies for new fabs, tax losses carried forward and currency.
- Quarterly seasonality and inventory held by customers, which is what turns a slowdown in end demand into a sharper fall in wafer orders.
Build it yourself
The starter workbook
The depreciation waterfall has been cleared from the Capex sheet: for each year’s spending from FY-6A to FY7E, the depreciation it carries in each year once its tools produce, the total, the net book value of the vintages and depreciation as a share of revenue. Rebuild them from the spending above and the lag and life on the Assumptions sheet. The Checks sheet shows when the depreciation over every vintage’s life ties to what was spent.
Blanks: Capex and depreciation (PP&E roll-forward). Download it beside the worked model.
The path around this model
Understand it, drill it, read the build, then apply it to a real company.
Build · Lab · ~10 min
Depreciation Waterfall
Forecast depreciation from a capex plan rather than a ratio, read a capex-to-depreciation multiple, and say which part of the spend is buying growth and which is only standing still.
Build · Lab · ~15 min
DCF Builder
Build a DCF by hand from revenue to a share price, know how much of it sits in the terminal value, and name the one assumption the answer hinges on.
Read · Guide · 16 min
How to Build a DCF Model: From Operating Assumptions to a Defensible Valuation
Read · Guide · 10 min
Scenario and Sensitivity Architecture: Building a Model That Can Be Stress-Tested
Read · Guide · 7 min
Equity Research Interview Questions: The Coverage Mindset
Apply · Skill
Model Audit
Find the errors in a financial model before someone senior does.
Apply · Skill
Company Profile
Build a banker-grade one-pager on any company, fast.
Vocabulary: Utilisation, Gross Margin, Contribution Margin, CapEx, Depreciation & Amortisation (D&A), Useful Economic Life, Free Cash Flow (FCF), Terminal Value.
Questions about this model
How far does gross margin move with utilisation?
Further than revenue, because 36.0% of revenue in FY2E is fixed cost: depreciation, fab labour and overhead. In the base case each point of utilisation moves FY2E gross margin by 0.41 points. In the bear case utilisation falls from 81.8% to 65.3% in FY2E, revenue falls 14.3% and gross margin falls from 37.7% to 27.4%, a degree of operating leverage of 2.64. At that lower level each point of utilisation is worth 0.62 points of margin, because the fixed cost is spread over fewer wafers.
How does capital spending become capacity and depreciation?
With a lag of a year, then over five. The $2,600m spent on advanced capacity in FY0A, at $30m for each thousand wafers a year, adds 86.7 thousand wafers a year of capacity in FY1E. All $4,300m spent in FY0A is depreciated at $860m a year from FY1E to FY5E. FY1E depreciation of $4,020m is one fifth of the spending of FY-4A to FY0A, which is why the Capex sheet lays out seven reported years beside the forecast.
Why does the blended price per wafer rise so fast?
Mostly mix. The blended price rises from $6,329 in FY0A to $9,084 in FY7E. Priced at FY0A prices, FY7E’s wafers would still average $8,175, so two thirds of the rise comes from shipping more advanced wafers rather than charging more for any of them. Advanced nodes go from 39.0% of revenue to 64.4%.
How is yield treated?
As a cost. The customer pays for each good wafer shipped and bears die yield; the foundry pays materials, chemicals and power on every wafer it starts. In FY1E an advanced wafer started costs $3,601 in materials and power, and at an 89% line yield each wafer shipped carries $4,046. Wafers scrapped across the three nodes cost $263m in FY1E. The advanced yield rises three points a year to a 95% ceiling, which lowers cost per wafer shipped as the node matures.
What is the foundry worth?
In the base case, an enterprise value of $72.7 billion from the DCF, 8.3 times FY1E EBITDA, and $37.87 a share after $3.0 billion of net cash. The terminal value is 70.3% of enterprise value and implies 6.1 times FY7E EBITDA. At 7.5 times FY1E EBITDA the cross-check gives $34.56. The bear case is worth $26.06 a share and the bull case $43.52.
Related models
Infrastructure and project finance · Advanced
GPU cloud infrastructure model
GPU cohorts, contracts and backlog, power, facility debt
Core modelling · Foundation
Integrated three-statement model
The operating model everything else sits on
Core modelling · Core
Discounted cash flow model
Unlevered free cash flow to a share price
Questions about this model
Discussions open here soon.