Why Terminal Value Is the Most Fragile Part of Your DCF

Terminal value is the present value of all the cash flows a business will generate beyond your explicit projection period — everything from year eleven onward, collapsed into a single number. It appears at the end of a DCF model almost as an afterthought: you've spent considerable effort building a 10-year projection, estimating revenue growth and margins and capital expenditures year by year, and then the model needs one more input to be complete. That input, quietly, carries more weight than everything you've just built.

In most DCF analyses of durable, growing businesses, terminal value accounts for 60 to 80 percent of total estimated enterprise value. Not 30 percent. The majority. That isn't a practitioner quirk or a modeling shortcut — it follows directly from present value arithmetic. When a company's discount rate modestly exceeds its terminal growth rate, the denominator in the terminal value formula is small, which makes the terminal value large, which makes the discounted terminal value a dominant share of the total output. The math is implacable. A majority of your intrinsic value estimate sits in a number you cannot forecast with the precision the model implies.

The appropriate response isn't a better terminal value formula. It's treating the terminal period as a range of plausible outcomes rather than a single figure — understanding what moves the output and by how much, then refusing to report any DCF conclusion that doesn't acknowledge the sensitivity. This post works through the mechanics, the two methods used in practice, a worked example using Microsoft's FY2024 annual report, and the discipline that makes terminal value manageable rather than fictional.

Why Terminal Value Grows So Large

Start with the arithmetic of discounting. Every year of a DCF projection, a cash flow is divided by (1 + r)^t, where r is the discount rate and t is the year. A dollar received in year 10 at a 9% discount rate is worth about 42 cents today. A dollar received in year 1 is worth 92 cents. Near-term cash flows survive the discount process mostly intact. But the cumulative effect over a 10-year projection period — even at a modest 9% rate — reduces each year's contribution substantially. The 10-year discounted sum, for a typical quality business, is a fraction of the total DCF output.

Then you add the terminal value. For a business generating approximately $7.80 in owner earnings per share today, growing at 10% annually, the discounted sum of years 1–10 at a 9% discount rate is roughly $82 per share, with year-ten owner earnings reaching approximately $20.25. Apply a terminal growth rate assumption and the Gordon Growth Model to those year-ten earnings, and the terminal period dwarfs the projection period.

Gordon Growth Terminal Value = Final Year Owner Earnings × (1 + g) / (Discount Rate − g) Where g = assumed long-run terminal growth rate

At g = 3% and r = 9%, the denominator is 6 percentage points: terminal value equals $20.25 × 1.03 / 0.06 = $348 per share. Discounted back at (1.09)^10 — a factor of about 2.37 — that's $147 today. Add to the $82 from the projection period and total estimated fair value is approximately $229. The terminal period contributes $147 of $229 — 64 percent. Push g to 4% and the terminal share rises to 68 percent. Drop it to 2% and it's still 60 percent. Within the conventional 2–4% band practitioners use for quality businesses, a majority of DCF fair value sits in the terminal period regardless of where you set g.

Two Methods, Two Different Failure Modes

Practitioners use two methods to estimate terminal value. Each solves a real problem. Each introduces a distinct failure mode.

The Gordon Growth Model — the perpetuity formula above — forces an explicit growth assumption. That's useful: you know exactly what you're asserting about the long run, and you can test it against historical performance and industry structure. The fragility is mechanical. As g approaches r, the denominator approaches zero, and the formula becomes unstable. At g = 8.5% and r = 9%, the denominator is 0.5 percentage points. A 0.1-point move in either direction changes the terminal value by 20 percent. For businesses where anything between 3% and 6% terminal growth might be defensible, the imprecision in your assumption produces enormous imprecision in the output.

Exit Multiple Terminal Value = Final Year Owner Earnings × Multiple (Multiple calibrated to sector comparables — e.g., 20× for a wide-moat quality business, 15× for a more cyclical one)

The exit multiple method is often described as more grounded because it draws on observable market valuations rather than a growth rate assumed in perpetuity. But 'grounded in current market comparables' means 'assumes market pricing in 2034 resembles market pricing in 2024.' In a rising-rate environment, multiples compress. In a contraction, they collapse. The exit multiple method doesn't escape terminal value's fundamental uncertainty — it trades a growth rate assumption for a market sentiment assumption, anchoring the terminal period to a level of market pricing that may look very different at the model's terminal date. Different class of fragility. Not less of it.

Microsoft FY2024: What the Sensitivity Looks Like

Let me make this concrete using Microsoft (MSFT) and data from its Form 10-K for the fiscal year ended June 30, 2024, filed with the SEC in July 2024. Microsoft reported net income of approximately $88.1 billion for FY2024. Depreciation and amortization totaled approximately $14.4 billion. Capital expenditures were approximately $44.5 billion, elevated by the AI infrastructure spending that dominated Microsoft's capital allocation in the period. Owner earnings: $88.1B + $14.4B − $44.5B ≈ $58.0 billion. With approximately 7.43 billion diluted shares outstanding, that's approximately $7.80 per share.

For the projection period, assume owner earnings grow at 10% annually for ten years — roughly consistent with Microsoft's earnings trajectory between FY2019 and FY2024. Use a 9% discount rate, calibrated to Microsoft's wide-moat competitive position. The discounted projection period sums to approximately $82 per share; year-ten owner earnings reach approximately $20.25. Now apply three terminal growth rate assumptions: 2%, 3%, and 4%.

Gordon Growth Model — Terminal Value Sensitivity (r = 9%, projection period at 10% growth) g = 2%: TV = $20.25 × 1.02 / 0.07 = $295 → PV of TV ≈ $125 → Total FV ≈ $207/share (terminal share: 60%) g = 3%: TV = $20.25 × 1.03 / 0.06 = $348 → PV of TV ≈ $147 → Total FV ≈ $229/share (terminal share: 64%) g = 4%: TV = $20.25 × 1.04 / 0.05 = $422 → PV of TV ≈ $178 → Total FV ≈ $260/share (terminal share: 68%) Projection period PV ≈ $82 across all three scenarios

From the 2% case to the 4% case — a 2-percentage-point range that sits entirely within the conventional band practitioners apply to quality businesses — estimated fair value moves from approximately $207 to $260 per share. A 26 percent swing. Not from a different revenue model. Not from a different view on Microsoft's competitive durability. From one assumption about what happens starting in year eleven.

Microsoft traded around $415 to $430 per share in late 2024, well above these estimates — reflecting the market's expectation that Azure and Copilot would drive growth exceeding the 10% projection-period assumption used here. That gap is expected and beside the point. The exercise demonstrates the structure of the sensitivity: a 2-percentage-point terminal growth range produces a 26-percent fair value range before you've changed anything else about the business or the model.

The exit multiple version of the same analysis produces a comparable spread. At 18× final-year owner earnings, the terminal value is approximately $364 per share, discounting to roughly $154, for a total fair value of about $236. At 26×, the terminal value reaches $527, discounting to $223, for a total fair value of approximately $305. An 8-multiple-point range — fully within the historical variation in quality-business valuations across market cycles — produces a 29-percent range in estimated fair value. Same business. Same ten-year projection. Different terminal assumption.

Sensitivity Ranges as the Actual Output

Once you internalize that terminal value drives the majority of your DCF and carries genuine uncertainty in both major calculation methods, presenting a single-point DCF estimate becomes difficult to defend. The analyst who reports $229 as fair value isn't wrong — but they're implicitly asserting a 3% terminal growth rate with confidence the inputs don't support. The honest output of the Microsoft analysis above is $207 to $260 for a 2%–4% terminal growth range, and $236 to $305 for a plausible exit multiple range. That is the analysis. The single number is a stylistic choice among equally plausible alternatives.

Pairing terminal value sensitivity with a reverse DCF is, I find, the most clarifying combination. Rather than choosing terminal assumptions first and solving for fair value, a reverse DCF starts from the observable stock price and solves for the terminal growth rate already embedded in it. When a quality business trades at a price implying 8–9% terminal growth — above the conventional GDP-aligned ceiling of 2–4% — you're not calling the stock expensive. You're identifying the specific belief required to justify the price, which is a more honest framing for the decision than 'P/E seems rich.'

I'll admit this is an area where my own practice has shifted, and I'm less settled on it than I'd like to be: I used to present base-case DCF outputs as the primary result, treating sensitivity as a supplementary table. I'm no longer comfortable with that approach. The range is the result. The base case is one scenario within it. Any margin of safety worth trusting should exist relative to the conservative terminal scenario — the low growth, modest exit multiple end — not only relative to the base. A discount to the base case evaporates if the terminal growth assumption moves by a single point; a discount to the conservative case does not.

Goldman Sachs's 2026 valuation of MiniMax — the Chinese AI startup — applied a comparable structural discipline. The firm segmented the model into a detailed forecast period through 2030, a stable growth period extending through 2035, and Gordon Growth terminal value from 2035 forward, discounting everything at a 12% WACC that reflected the uncertainty of a pre-revenue business. By pushing the terminal period's starting point thirteen years out rather than ten, Goldman reduced terminal value's mechanical leverage over the total output — more of the estimated value sits in the explicitly modeled stages, where assumptions can be tested against near-term data. You're not eliminating the terminal value problem with this technique. But you're managing its leverage over the conclusion, which is the realistic goal.

Key Takeaways

• In most DCF models for quality businesses, terminal value accounts for 60–80% of total estimated fair value. This follows from present value arithmetic — it's a structural feature of the calculation, not a practitioner error.
• The Gordon Growth Model is mechanically transparent but unstable as g approaches r. The exit multiple method anchors terminal value in current market comparables but embeds a market sentiment assumption that may not hold at the model's terminal date. Both methods require named assumptions and sensitivity analysis before any conclusions are drawn.
• Using Microsoft's FY2024 10-K data (net income $88.1B, D&A $14.4B, capex $44.5B, owner earnings approximately $7.80/share), a 2%–4% terminal growth range produces a fair value range of roughly $207 to $260 per share — a 26% swing driven by one assumption. An 18×–26× exit multiple range produces a comparable 29% spread.
• The range of plausible terminal assumptions is the actual output of DCF analysis. Any margin of safety worth relying on should hold under the conservative terminal scenario — low growth, modest exit multiple — not only under the base case.
• A reverse DCF is the natural complement: start from today's observable stock price and solve for the terminal growth rate already embedded in it. Then assess whether that implied rate is realistic given historical performance and industry capacity.