بِسْمِ اللَّهِ الرَّحْمَٰنِ الرَّحِيمِ
In the name of Allah, the Most Gracious, the Most Merciful
Imagine you run two restaurants on the same street. Both serve dinner to 100 people every night. Both charge $30 per plate. But at the end of the year, one owner takes home $500,000 and the other takes home $8,000.
Same revenue. Same customers. Wildly different outcomes.
That's exactly what happened in the semiconductor industry in 2023. NVIDIA earned a 91.5% return on every dollar of shareholder equity. Intel earned 1.6%. They sell to many of the same customers. They compete in the same industry. But the financial statements tell completely different stories.
The question isn't just "who's doing better?" That's obvious. The real question is: WHERE is the difference coming from? And what does that tell you about whether it will last?
- ROE (Return on Equity) tells you how much profit a company generates per dollar of shareholder money
- DuPont breaks ROE into three gears: profit margins x asset efficiency x financial leverage
- Each gear reveals a different story: NVIDIA wins on margins (49% net profit), TSMC wins on efficiency (foundry scale), Intel struggles on all three
- Beyond DuPont: Liquidity ratios and the cash conversion cycle reveal whether a profitable company can actually pay its bills
This post is for you if:
- You've heard of ROE but don't know what drives it up or down
- You want to compare companies in the same industry and understand why their numbers differ
- You read Post 1 and want to go deeper into the financial ratios behind strategy
In How to Read a Company's Strategy From Its Numbers, we learned that NVIDIA, TSMC, and Intel have three fundamentally different strategies. Now we'll see exactly how those strategies show up in the numbers — ratio by ratio.
The Engine Analogy
Think of a company as a car engine. ROE — Return on Equity — is the speedometer. It tells you how fast you're going, but not why.
Three gears determine the speed:
- Gear 1 (Margins) = fuel efficiency. How much energy (profit) you extract from each unit of fuel (revenue). A high-margin company converts more of each sale into profit.
- Gear 2 (Efficiency) = how hard the engine works. How many times your assets cycle through revenue. A high-turnover company squeezes more sales from fewer assets.
- Gear 3 (Leverage) = turbo boost. Borrowed money that amplifies your returns — but also amplifies your risk. More leverage = more speed, but blow the engine if it fails.
Understanding the speedometer reading (ROE) is useful. Understanding the three gears that produce it? That's how you predict what happens next.
Before we decompose ROE, it helps to know that managers increase company value through four types of decisions:
The first two are operating decisions — how you run the business day-to-day. The last two are financing decisions — how you fund it. DuPont analysis captures all four, but (as we'll see later) it doesn't cleanly separate them. That's both its power and its limitation.
ROE — The One Number CEOs Care About
If there's one number that defines corporate performance, it's ROE — Return on Equity. It answers a simple question: for every dollar shareholders invested, how much profit did the company generate?
Let's walk through a real calculation. Intel in 2021:
- Net Income: $19,868 million
- FY2020 Equity: $81,038 million
- FY2021 Equity: $95,391 million
- Average Equity: ($81,038 + $95,391) / 2 = $88,215 million
- ROE = $19,868 / $88,215 = 22.5%
Not bad for 2021. But fast-forward to 2023, and Intel's ROE dropped to 1.6%. Meanwhile, NVIDIA's ROE hit 91.5%.
Same industry. Same customers. One returns $0.016 for every dollar of equity, the other returns $0.915. A 57x difference. The question is: where did those returns come from — and where did they go?
Think of the income statement as a movie — it captures the full year of activity. The balance sheet is a photograph — it captures one moment in time.
If a company issued $10 billion in new stock mid-year, the ending equity is much higher than what was actually "in use" for most of the year. Using the average approximates the capital base that was working throughout the period.
This especially matters when big equity changes happen: new stock issuances, large dividends, share buybacks, or significant profits that grow equity over the year.
Charis Ji, a research analyst at Dodge & Cox (a San Francisco firm that actively manages $300 billion in assets), looks for returns on invested capital, margin structure, capital intensity, and balance sheet strength. Their investment process emphasizes bottom-up research with a 3-5 year investment horizon.
She points to UPS vs. FedEx as a case study: both deliver packages, but UPS runs an integrated network between air and ground, while FedEx runs separate networks. The result? Very different margins and return on capital — same business, different operational architecture, completely different financial profiles.
Neal Nathani, CEO of Totem Point (20 years investing in technology), analyzes companies through the S-curve framework: early stage → growth → maturation → decline. He points to Microsoft as an example — for about a decade, Microsoft was in a maturation phase linked to the PC cycle. Wall Street had written them off. Then cloud computing changed everything, and the S-curve restarted.
As Satya Nadella puts it: "The technology industry doesn't respect tradition, it respects innovation."
The lesson: when you look at a company's ROE, ask yourself where on the S-curve it sits. Intel's 1.6% ROE looks like late maturation (or early decline). NVIDIA's 91.5% looks like the peak of a growth explosion. Understanding which phase a company is in tells you whether current ratios are likely to improve, hold, or deteriorate.
Same industry. Same customers. Three completely different readings. But why?
The DuPont Decomposition — Three Gears
In the 1920s, executives at DuPont Corporation realized that staring at ROE as a single number was like watching the speedometer without opening the hood. They developed a decomposition that breaks ROE into its three component gears:
Notice how the math works: Sales cancels out between the first two terms. Assets cancels out between the last two. You're left with Net Income / Equity — which is just ROE. But by keeping the intermediate terms, you can see which gear is driving performance.
Let's decompose Intel's 2023 ROE to see where the 1.6% comes from:
- Net Profit Margin: $1,675M / $54,228M = 3.1%
- Asset Turnover: $54,228M / $186,838M = 0.29
- Financial Leverage: $186,838M / $106,626M = 1.75
- ROE: 3.1% x 0.29 x 1.75 = 1.6%
Same formula. Three completely different stories. NVIDIA's margin drives its ROE. Intel's collapsed margin drags everything down.
Now the picture is clear. NVIDIA's 91.5% ROE is driven overwhelmingly by its 48.8% net profit margin — nearly half of every dollar in sales becomes profit. Intel's leverage (1.75) is actually similar to NVIDIA's (1.80), but it doesn't matter because the margin collapsed to 3.1%.
TSMC sits in the middle — good margins (38.4%), lower turnover (0.47, because foundries need massive assets), and conservative leverage (1.45). A stable, efficient machine.
In fact, the source material identifies which gear matters most for each business model:
- NVIDIA → Margin is the gear to watch. R&D is expensive (15-20% of revenue). If those designs don't command premium pricing, the whole model collapses.
- Intel → Efficiency is the critical gear. As a vertically integrated company with resources stretched across design AND manufacturing, efficiency determines whether all those assets earn their keep.
- TSMC → Leverage is the key gear. Each advanced fab costs $10 billion+ and takes years to build. How TSMC finances those enormous investments — and at what cost — determines whether the returns justify the capital.
Before DuPont, analysts looked at ROE as a single number. A company with 20% ROE was "good." One with 5% was "bad." End of analysis.
DuPont executives realized in the 1920s that decomposing the number revealed where to focus management effort. Two companies can have identical 15% ROE through completely different paths:
- Company A: 15% margin x 1.0 turnover x 1.0 leverage = 15% ROE (luxury brand)
- Company B: 3% margin x 2.5 turnover x 2.0 leverage = 15% ROE (grocery chain)
Same ROE. Completely different businesses. Different risks. Different futures.
You can use DuPont two ways: time series (compare your own gears over years — is margin improving or declining?) and cross-sectional (compare gears across peers — who wins on which dimension?). This framework is still the starting point for professional financial analysis today.
Retailers win on volume: thin margins (5%), but assets cycle rapidly (turnover 2.5) and they use more leverage (2.0). Luxury brands win on value: fat margins (25%), but assets move slowly (turnover 0.4) and they need less leverage (1.5). Same ROE, completely different business models.
Think about it: retailers sell massive volumes at thin margins (Walmart's net margin is ~2.5%). Luxury brands sell fewer items at premium prices (high margins). High turnover + thin margin = volume business. Low turnover + fat margin = premium business. The DuPont gears reveal the strategy.
Gear 1: Margin Ratios — From Gross to Net
Four Layers of Profit
Think of revenue as a waterfall. At the top: $100 of sales. As it cascades down, each layer of cost peels away some water. Gross margin is the first ledge. Operating margin is the second. Net margin is the pool at the bottom.
The question is: how much water reaches the bottom?
Each layer peels away costs. The drop from gross (55%) to net (25%) tells you how much the "business of running the business" costs.
Here are the four key margin ratios, from top to bottom:
- Gross Margin = (Sales - COGS) / Sales — How much you keep after the direct cost of making the product. This is closest to the product itself.
- Operating Margin = Operating Income / Sales — How much you keep after running the business (R&D, sales team, administration).
- EBITDA Margin = EBITDA / Sales — Operating profit before depreciation and amortization. Useful for comparing capital-heavy companies where depreciation varies wildly.
- Net Profit Margin = Net Income / Sales — What's left after absolutely everything: costs, operations, interest, taxes, unusual items.
Now let's compare our three companies in 2023:
| Metric | NVIDIA | TSMC | Intel |
|---|---|---|---|
| Gross Margin | 72.7% | 54.4% | 40.0% |
| Operating Margin | 54.1% | 42.6% | 0.1% |
| Net Profit Margin | 48.8% | 38.4% | 3.1% |
Intel's story is stark. Even at gross margin (40%), they're behind. But the collapse from 40% gross to 0.1% operating means their R&D and operating costs are eating nearly everything. By the time you get to net margin (3.1%), almost nothing is left.
Interestingly, Intel's 2021 income statement reveals something subtle: net income ($19,868M) was actually close to operating income ($22,082M) — but only because a $2,729M gain on sale of investments offset $2,291M in legal settlements and other charges. When non-operating items happen to cancel out, net and operating margins look similar. When they don't — and they often don't — the gap between operating and net margin tells you how much of the bottom line comes from the actual business versus one-time events.
Gross margin is the ratio closest to the product itself. It strips away everything except the direct cost of what you sell:
- NVIDIA 72.7%: Sells chip designs, not silicon. TSMC does the manufacturing. NVIDIA's COGS is mostly licensing fees and packaging — minimal physical cost.
- TSMC 54.4%: Runs massive fabrication plants ($10B+ each). High costs, but spreads them across enormous volume and near-maximum capacity utilization.
- Intel 40%: Pays for BOTH design AND manufacturing. Neither world-class at this point — AMD took server market share from 5% to 15-20%, and TSMC makes better chips.
When gross margin declines over multiple years, it signals competitive pressure or a structural cost problem. Intel's decline from 62.3% (2018) to 40% (2023) is not a blip — it's a five-year structural collapse.
NVIDIA's R&D as a percentage of revenue dropped from 27.2% (2022) to 14.2% (2023). At first glance, that looks like they're cutting investment in innovation. They're not.
Absolute R&D spending: $7,339M (2022) → $8,675M (2023) — an 18% increase.
Revenue: $26,974M → $60,922M — a 126% explosion driven by AI chip demand.
The ratio dropped because the denominator (revenue) grew much faster than the numerator (R&D). This is a critical lesson in financial analysis: always check absolute numbers, not just ratios. A dropping ratio can mean "spending less" or "earning way more." Context is everything.
Here's an even more striking comparison: in 2022, both Intel and NVIDIA spent about 27-28% of revenue on R&D. Nearly identical investment intensity. Yet NVIDIA's R&D produced the chips that dominated the AI boom, while Intel's R&D couldn't prevent market share losses in servers and PCs. Same spending level, wildly different outcomes — proof that how you spend matters far more than how much.
Looking at these margin numbers, a professional analyst would have pointed questions for each company:
- Intel CFO: "Operating expenses — especially R&D — keep rising while revenue is falling. At what point do you restructure the cost base to match declining sales?"
- NVIDIA CFO: "R&D as a percentage of revenue dropped from 27.2% to 14.2%. Are you under-investing in future chip architectures, or is this just the denominator effect of the AI revenue explosion?"
- TSMC CFO: "Operating margin jumped from 41% to roughly 50% in a single year (2021 to 2022). What drove that sudden spike — pricing power, utilization, or one-time factors?"
Notice how each question targets a different concern: Intel's cost discipline, NVIDIA's reinvestment rate, TSMC's margin sustainability. The ratios surface the questions; the answers require judgment.
When margins decline, management typically pulls these levers:
- Check for redundant capacity: Close unprofitable plants or production lines
- Cancel non-productive projects: Kill R&D programs that aren't generating returns
- Renegotiate supplier contracts: Push for better terms on raw materials
- Examine pricing power: Can you raise prices without losing customers?
Intel's challenge in 2023: Operating margin hit 0.1%. R&D and SG&A expenses were rising while revenue was falling 14%. That's not just a market problem — it's an execution problem. Revenue dropped 20% in 2022 and another 14% in 2023, yet costs didn't shrink proportionally.
This is Intel. A consistent multi-year decline across ALL margin levels — gross, operating, and net — signals a fundamental business model challenge, not a temporary blip. AMD gained server market share (5% → 15-20%), and TSMC out-executed Intel in manufacturing. When every margin ratio declines for five straight years, it's structural.
Cyclical downturns affect a single year, maybe two. This is a five-year slide across every margin metric. One-time charges would show up in net margin but not gross margin. When gross margin — the closest ratio to the product — declines by 22 percentage points over five years, the company is losing its competitive edge.
Gear 2: Efficiency Ratios — Are Assets Working?
Asset Turnover — How Hard is the Engine Working?
Imagine two coffee shops. Both invested $100,000 in equipment. Shop A generates $500,000 in sales. Shop B generates $200,000. Same investment, but Shop A's equipment is working 2.5x harder. That's asset turnover — sales per dollar of assets.
But asset turnover is a blunt instrument. You can break it further into working capital efficiency (short-term assets) and long-term asset efficiency (factories, equipment):
Working Capital Ratios
- Working Capital Turnover = Sales / Avg Working Capital — How well the company uses its current assets (minus current liabilities) to maintain sales. This is the big-picture working capital metric.
- Accounts Receivable Turnover = Sales / Avg AR — How fast do customers pay you?
- Inventory Turnover = COGS / Avg Inventory — How fast does stock move off the shelf?
- Accounts Payable Turnover = Purchases / Avg AP — How fast do you pay your suppliers? (Note: Purchases = COGS + net change in inventory)
Long-Term Asset Ratios
- Long-Term Assets Turnover = Sales / Avg Long-Term Assets — How efficiently a company generates sales from all its non-current assets (including intangibles, investments, etc.)
- PP&E Turnover = Sales / Avg PP&E — How much revenue does each dollar of factories and equipment generate? This is the narrower view focused on physical assets only.
| Metric | NVIDIA | TSMC | Intel |
|---|---|---|---|
| Asset Turnover | 1.04 | 0.47 | 0.29 |
| PP&E Turnover | Highest (no fabs) | Medium | Lowest |
| LT Assets / Sales | ~62% | 137% | 244% |
NVIDIA's efficiency advantage is structural: it's fabless. No factories, no massive PP&E. Every dollar of assets generates $1.04 in sales. Intel's assets generate only $0.29 per dollar — its fabs are enormous but underutilized as market share erodes.
But here's the surprise: TSMC (0.47) is more efficient than Intel (0.29) despite having more physical assets than Intel. How? Capacity utilization. TSMC runs near maximum capacity because it makes chips for everyone — Apple, NVIDIA, AMD, Qualcomm. Intel's fabs increasingly sit idle.
And here's a counter-narrative worth noting: Intel actually has the best working capital efficiency and fastest receivables collection of the three companies. NVIDIA, despite its dominant ROE, is the slowest at collecting receivables and turning over inventory. This is why you can't judge a company by one ratio — Intel manages its short-term assets well; it's the long-term asset utilization and margins that are killing it.
There's also a revealing supplier strategy difference: TSMC pays its suppliers the fastest of the three (highest accounts payable turnover). That's not inefficiency — it buys better relationships, creditworthiness, and potential early-payment discounts. Intel, by contrast, appears to have deliberately slowed payments to suppliers to conserve cash for working capital needs. When you're profitable, you can afford to be generous with suppliers. When you're struggling, every day of float matters.
NVIDIA is fabless — no factories. Its PP&E is offices, servers, and design tools. Every dollar of PP&E generates massive sales because the actual manufacturing is TSMC's problem.
TSMC has huge PP&E — each advanced fab costs $10 billion+ to build. But with 52% global foundry market share and 92% of advanced 5nm chips, their utilization rates are extremely high.
Intel has huge PP&E AND declining sales. Their long-term assets represent 244% of sales in 2023, versus TSMC's 137% and NVIDIA's ~62%. That's a company with too many factories for its shrinking customer base.
When asset turnover drops, management can:
- Negotiate faster payment terms: Move from 45 to 30 days receivable
- Offer discounts to move inventory: Reduce inventory sitting on shelves
- Assess manufacturing utilization: Are fabs running at capacity?
Common mistakes:
- NOT: Replace long-term assets with short-term ones (gaming the ratio, not improving the business)
- NOT: Simply increase inventory (makes turnover worse)
The real question: Is turnover declining because assets grew (you invested heavily) or because sales shrank (customers left)? The first might be temporary. The second is a warning.
TSMC runs at near-maximum capacity because it manufactures for the entire industry — Apple, NVIDIA, AMD, Qualcomm. Intel's fabs are increasingly underutilized as it loses market share. Efficiency isn't about how FEW assets you have — it's about how HARD those assets work.
Equipment age doesn't explain a 60% efficiency gap. And debt is about leverage (Gear 3), not efficiency (Gear 2). The answer is utilization — TSMC's fabs run near maximum capacity serving hundreds of customers, while Intel's fabs sit increasingly idle as AMD takes market share. Same assets, different workload.
Gear 3: Leverage — Power and Risk
The Turbo Boost
Leverage is like a turbo boost in a car. It makes you go faster — BUT if the engine can't handle the extra power, it explodes. A company that borrows $5 for every $1 of equity can multiply returns 5x... or multiply losses 5x.
Four ratios measure leverage and financial health:
- Leverage Ratio = Total Liabilities / Total Equity — Overall reliance on other people's money
- Debt-to-Equity = Total Debt / Total Equity — Just the "expensive" money (bonds, loans — not accounts payable or accruals)
- Interest Coverage = Operating Income / Interest Expense — Can the company comfortably pay interest on its debt?
- Dividend Payout = Cash Dividends / Net Income — How much goes to shareholders vs. reinvestment? Companies pay dividends to return profits, signal confidence about future prospects, and attract income-seeking investors. Alternatively, they retain earnings to fund new growth.
Here's the surprising part: NVIDIA's leverage (1.80) and Intel's (1.75) are nearly identical. Both borrow similar amounts relative to equity. But NVIDIA generates 48.8% net margin on that leverage while Intel generates 3.1%.
The issue isn't how much you borrow — it's whether your business earns more than the borrowing costs.
And look at interest coverage: TSMC peaked at 272x — generating 272 times more operating income than its interest costs. Intel's interest coverage dropped to nearly zero by 2023. Same leverage, wildly different ability to service debt.
Intuition says: less debt = safer. But that's not always true from a shareholder's perspective.
If borrowing costs 5% and your business earns 20%, every borrowed dollar generates 15% profit for shareholders. Zero debt is safe, but you're leaving returns on the table.
NVIDIA (1.80) and Intel (1.75) have nearly identical leverage ratios. But NVIDIA generates vastly more profit per dollar of assets. The leverage amplifies NVIDIA's strong operating performance, while Intel's leverage amplifies... not much.
The real question: Does the business earn more than the cost of borrowing? If yes, leverage is smart. If no, leverage is dangerous.
If leverage becomes dangerous, management can:
- Issue additional stock (increases equity, but dilutes existing shareholders)
- Pay down long-term debt from operating cash flow
- Sell non-core assets and use proceeds to retire debt
Common mistakes:
- NOT: Increase short-term payables (this actually increases total liabilities = more leverage)
- NOT: Just reduce inventory (doesn't help unless proceeds go to paying down debt)
The Safety Net: Liquidity and Working Capital
Can You Pay Your Bills?
A company can be profitable and still go bankrupt. Imagine earning $10 million per year, but all your cash is locked in 2-year contracts. When a $5 million bill comes due tomorrow, your profit margin doesn't help.
Liquidity is about WHEN cash is available, not how much you earn.
Three ratios test this, each stricter than the last:
- Current Ratio = Current Assets / Current Liabilities — Can all short-term assets cover short-term debts?
- Quick Ratio = (Cash + ST Investments + AR) / Current Liabilities — Same test, but exclude inventory (harder to sell quickly)
- Cash Ratio = (Cash + ST Investments) / Current Liabilities — Nuclear option: can you pay everything with cash on hand?
| Metric (2022) | NVIDIA | TSMC | Intel |
|---|---|---|---|
| Current Ratio | 3.52 | 2.17 | 1.57 |
| Quick Ratio | 2.61 | 1.90 | 1.02 |
| Cash Ratio | 2.03 | 1.65 | 0.88 |
NVIDIA leads on all three — fabless companies tend to be cash-rich because they don't sink billions into factories. Intel's cash ratio of 0.88 means it can't cover all current liabilities with cash alone. Not critical yet, but a warning sign given its declining profitability.
The Cash Conversion Cycle — The Hidden Metric
How long does it take to turn a dollar spent on inventory back into cash in the bank? That's the Cash Conversion Cycle:
Every day in the cycle = money you need sitting idle or borrowed. Amazon flipped the equation entirely.
Amazon collects payment immediately (credit card at checkout). Holds inventory briefly (just-in-time fulfillment + marketplace model where sellers hold their own inventory). Pays suppliers on 60-90 day terms.
Result: Amazon operates with other people's money. Suppliers are essentially financing Amazon's growth — they deliver products today and get paid months later. In the meantime, Amazon invests that cash.
This is the same principle as Netflix using studio payables to finance its content library (we covered this in Post 2). Both companies turned their suppliers into involuntary investors.
This is why Amazon could grow at massive scale despite near-zero profits for years. When your cash conversion cycle is negative, you don't need much external capital to fund operations.
Putting It All Together: Three Business Models
Now we've seen all the ratios individually. Let's see what they reveal when we look at the complete picture — all three companies, all ratios, side by side.
NVIDIA — "The Margin Machine"
Model: Fabless. Low CapEx (1-10% of revenue), high R&D (15-20%). Designs chips, outsources manufacturing to TSMC.
Wins on: Margins (72.7% gross, 49% net), PP&E turnover (no fabs). Generated $60.9 billion in revenue in FY2023 — a 126% increase driven by AI demand. Holds 83% of the discrete GPU market and powers 70% of the world's top 500 supercomputers.
Weakness: Entirely dependent on TSMC for manufacturing. And crucially, NVIDIA is the most volatile of the three — gross margin dipped to 56.9% in 2022 before rebounding to 72.7% in 2023, and net profit margin swung from 16.25% in 2022 to 48.8% in 2023. The 91.5% ROE is an AI-boom peak, not a steady state. If AI demand cools, those margins can compress just as dramatically as they expanded.
ROE story: 91.5% driven almost entirely by that staggering 48.8% net profit margin.
Founded in 1993 as a Silicon Valley startup, NVIDIA's core belief was that the future of computing was accelerated, graphics-based computing. Their third chip — the RIVA 128 — was a breakout hit in gaming.
The name? "NV" for "next version" + "invidia," Latin for "envy." They went public in 1999 and are credited as the inventor of the GPU — a chip category that didn't exist before them.
Through the 2010s, NVIDIA expanded from gaming GPUs into scientific computing, AI training, data science, autonomous vehicles, and robotics. The company's own motto captures it: "We deliver GPU performance improvements on a pace ahead of Moore's Law."
That bet on GPU computing — which seemed niche for decades — turned out to be the exact architecture needed for the AI revolution. Sometimes the best strategy is being early and patient.
TSMC — "The Efficiency Engine"
Model: Foundry. High CapEx (10-45% of revenue), lower R&D (~8%). Makes chips for everyone: Apple, NVIDIA, AMD, Qualcomm. Operates 17 fabs, produces 11,617 distinct products for 510 customers.
Wins on: Inventory turnover, capacity utilization, stable margins (54-60% gross, peaking at 59.6% in 2022 — its best year ever). 52% global foundry market share. 92% of advanced 5nm chips.
Strength: Dubbed "the Hope Diamond of semiconductors" by G. Dan Hutcheson of TechInsights. Their scale and expertise create a moat that's nearly impossible to replicate — each advanced fab costs $10B+ and takes years to build.
Surprise: You'd expect a capital-heavy foundry to have the lowest return on assets (ROA) of the three. But TSMC does NOT — because its foundries are more efficient than Intel's. Despite Intel also having design revenue, Intel can't boost its overall ROA because its foundry operations drag it down. TSMC, by focusing exclusively on manufacturing excellence, achieves better asset returns than the company trying to do both.
ROE story: 26.1% — solid margin (38.4%) on lower turnover (0.47, capital-heavy) with conservative leverage (1.45).
Founded in 1987 in Hsinchu, Taiwan, as a joint venture between the Taiwanese government and Philips Electronics of the Netherlands, TSMC didn't enter an existing market — it created one. Before TSMC, if you wanted to design chips, you had to own a factory. TSMC pioneered the dedicated foundry model: "You design it, we'll make it."
They broke the 1-micron feature size barrier in 1991, IPO'd on the Taiwan Stock Exchange in 1994, and became the first Taiwanese company listed on the NYSE in 1996.
By 2020, TSMC had a market cap of $560 billion (11th largest in the world). Their core strategy hasn't changed in 35+ years: achieve the best yield at the most advanced process through high-capacity utilization. They were the first to produce 5nm chips in large quantity — a process so precise that each transistor is roughly the size of a strand of DNA.
Intel — "The Warning Sign"
Model: IDM (Integrated Device Manufacturer). Does BOTH design AND manufacturing. Must excel at both to compete.
The decline: ROE fell from 29.3% (2018) to 1.6% (2023). Every single ratio declined. Gross margin: 62.3% → 40%. Operating margin: 30% → 0.1%. Interest coverage: nearly zero. Annual sales growth tells the trajectory clearly: 12.9% (2018) → 1.5% (2021) → -20% (2022) → -14% (2023). That's not a blip — it's a structural unraveling.
What happened: The introduction of the iPhone in 2007 triggered a seismic shift from PCs to mobile devices, demanding smaller, more powerful chips. This shift accelerated specialization — companies like NVIDIA focused purely on design while TSMC focused purely on manufacturing. Intel, which had dominated by doing both, found itself stuck between two fronts: competing against TSMC on manufacturing (where TSMC had greater scale, the best lithography machines from ASML, and 50%+ market share) and against AMD and NVIDIA on design (where AMD gained server market share from 5% to 15-20%, literally at Intel's expense).
ROE story: 1.6% — margin collapsed (3.1%), turnover worst in class (0.29), leverage average (1.75) but now amplifying weakness instead of strength.
Intel was founded in 1968 by Gordon Moore (of Moore's Law fame) and Bob Noyce — originally named "NM Electronics." Three years after Moore's famous observation that transistor density doubles roughly every two years, they set out to make it happen.
Intel started as a memory company, but pivoted to microprocessors in the 1970s after struggling to compete with Japanese memory manufacturers. Their chips powered the world's first personal computer in 1974.
The key to Intel's decades of dominance was the "sole source" strategy. Originally, Intel had to share its x86 instruction set IP with AMD. But they figured out how to become the sole supplier, which allowed them to leapfrog technology — staying generations ahead of competitors. The iconic "Intel Inside" marketing campaign cemented their brand with consumers, creating demand pull that PC manufacturers couldn't ignore.
By the early 2020s, Intel had over 100,000 employees in 46 countries. But the advantages that built an empire — vertical integration, sole-source dominance, PC-centric design — became liabilities in a world that moved to mobile, cloud, and AI.
| Metric | NVIDIA (Fabless) | TSMC (Foundry) | Intel (IDM) |
|---|---|---|---|
| ROE | 91.5% | 26.1% | 1.6% |
| Net Margin | 48.8% | 38.4% | 3.1% |
| Asset Turnover | 1.04 | 0.47 | 0.29 |
| Leverage | 1.80 | 1.45 | 1.75 |
| Gross Margin | 72.7% | 54.4% | 40.0% |
| Interest Coverage | Very high | 272x (peak) | Near 0 |
| Current Ratio | 3.52 | 2.17 | 1.57 |
Three companies in the same industry. Every metric tells the same story: NVIDIA dominates on profitability, TSMC on stability, Intel on decline.
Traditional DuPont is powerful, but it has two blind spots:
- It mixes operating and financing decisions. Net income includes interest expense and tax effects. Two companies with identical operations but different capital structures (one debt-heavy, one debt-free) will show different net margins — even though their actual business performance is the same.
- It ignores cash effects. The leverage ratio doesn't distinguish between a company sitting on $50 billion in cash and one with $0. Both could have the same leverage ratio, but one can pay off all debt tomorrow while the other can't.
Key insight: A company sitting on cash can pay down debt instantly, which would lower leverage, reduce interest expense, improve net margin, AND increase asset turnover. Traditional DuPont misses all of this. Analysts call this concept "net debt" — total debt minus cash holdings. Two companies can have identical leverage ratios but very different net debt, meaning very different actual risk.
This is exactly why the next post in this series introduces Modified DuPont — which separates operating performance from financing decisions, giving you an X-ray view of what the business actually does vs. how it's funded.
This is TSMC. The 0.47 asset turnover looks low, but it's expected for a capital-heavy foundry model — you NEED massive factories. What matters is that those factories earn 54% gross margins, the company can cover interest 272 times over, and leverage is conservative. Low turnover in a foundry is structural, not a weakness.
Low asset turnover isn't automatically bad — it depends on the business model. Foundries and manufacturers NEED heavy assets. TSMC's 0.47 turnover comes with 54% gross margins, 272x interest coverage, and $560B market cap. Context matters: always interpret ratios through the lens of the business model, not in isolation.
Practice Mode
Test your DuPont analysis skills with real scenarios
The Three Gears
- ROE = Margin x Turnover x Leverage
- Margin = profit per sale
- Turnover = how hard assets work
- Leverage = borrowed amplification
Key Ratios
- Gross Margin = (Sales-COGS)/Sales
- Operating Margin = OpInc/Sales
- Asset Turnover = Sales/Avg Assets
- Current Ratio = CA/CL
- CCC = Days AR + Inv - AP
Three Business Models
- Fabless — High margin, high turnover, low CapEx (NVIDIA)
- Foundry — Medium margin, low turnover, high CapEx (TSMC)
- IDM — Must excel at both or lose (Intel)
Was this helpful?
Your feedback helps me improve these guides