When Terminal Value Goes to Zero

How AI Breaks the Fundamental Assumption of Business Valuation

Seventy-one percent of a business's value can vanish with a single changed assumption. Not a recession. Not a competitor. Just the realization that "forever" might actually mean "five years." That's the math nobody in finance is running, and it changes everything about how we should value businesses in the age of AI.

A Quick Primer on How Businesses Are Valued

Most people think a business is worth some multiple of its profit. A company making $1 million a year might be "worth" $10 million. That multiple feels intuitive, but it actually hides a much more important assumption.

The formal way to value a business is called a Discounted Cash Flow (DCF) analysis. The logic is simple: a business is worth the sum of all the cash it will generate in the future, adjusted for the fact that money today is worth more than money tomorrow.

That adjustment is called the discount rate. If you could earn 10% per year in the stock market, then $100 next year is only worth about $91 to you today. $100 in two years? About $83. This is called Net Present Value (NPV), which is what all those future dollars are worth right now.

Here's where it gets interesting. Nobody can predict cash flows forever. So valuations typically forecast 5–10 years of specific projections, then add something called "terminal value," a single number representing all the cash flows from year 11 to infinity.

Terminal value typically accounts for 60–80% of a company's total valuation.

Read that again. Most of what a business is "worth" isn't the cash it'll generate in the next decade. It's the assumption that it will keep generating cash forever.

What Happens When Forever Becomes Five Years?

Let's run some simple math. Imagine a business that generates $1 million in profit per year. Using a 10% discount rate and assuming perpetual 2% growth (standard assumptions), the traditional DCF valuation looks like this:

That $13.1 million valuation implies a price-to-earnings multiple of about 13x. Seems reasonable for a stable business.

Now model the same business, but assume AI automation makes it obsolete in 5 years. No terminal value. No perpetuity. Just five years of cash, then zero.

Same business. Same cash flows today. 71% less value. That's the math of AI disruption that nobody is running.

Deterministic vs. Probabilistic: The Framework Nobody's Talking About

The standard AI disruption conversation goes like this: knowledge work gets disrupted first, physical trades are safe. This is wrong. Not because of the timeline, but because of how disruption actually works in each domain.

The crucial distinction isn't digital vs. physical. It's whether the output is deterministic or probabilistic.

Probabilistic Outputs: Knowledge Work

A marketing campaign, a legal brief, a consulting report. The quality of these outputs exists on a spectrum. There's no binary pass/fail. A strategy can be good enough, better, or brilliant. The output is probabilistic.

This means AI slots into knowledge work as augmentation. The marketing agency doesn't die overnight. It goes from 30 employees to 3 employees using AI tools. Value erodes gradually. Margins compress. The agency still exists, but it's a smaller business. Terminal value compresses but doesn't necessarily go to zero.

Knowledge work faces a slope. Margins compress over years. Businesses shrink but survive in reduced form. It's a slow bleed, not a cliff.

Deterministic Outputs: Physical Trades

A toilet flushes or it doesn't. A pipe leaks or it doesn't. An electrical circuit works or it doesn't. The output is deterministic: binary pass/fail.

This means there is no "AI-augmented plumber" middle state. You cannot half-fix a toilet with AI assistance. The human plumber is the complete solution until the day a robot can do the entire job, at which point the human is fully redundant. There's no graceful degradation, no margin compression phase, no augmentation era.

Physical trades face a cliff. Full employment until hardware catches up, then total displacement. No augmentation middle ground.

This is the opposite of what most people expect. The conventional wisdom says trades are safe because "robots can't crawl under a house." But that's confusing timeline with severity. Trades may be disrupted later than knowledge work, but when disruption arrives, it's absolute.

The Diagnosis Advantage Is Already Gone

The strongest counterargument to trade automation goes like this: even if a robot could physically fix a pipe, the real value of a plumber is diagnosis. Figuring out why the toilet is broken, navigating unpredictable environments, drawing on years of experience with different eras of construction. That's the human edge.

This was true in 2020. It's not true anymore.

Consider what a multimodal AI system already has access to: every building permit ever filed in the county, the complete Zillow listing history showing the 1987 renovation that used polybutylene pipes, the manufacturer recall database, thermal imaging data, moisture sensor readings, every plumbing forum post ever written, and the diagnostic patterns from every similar structure in the dataset. All synthesized simultaneously before anything is touched.

The human plumber is working off maybe 20 years of anecdotal experience. The AI has the complete informational history of the structure plus every similar structure ever built. On diagnosis, it's not even close.

So physical trades aren't protected on either axis. The cognitive side (diagnosis, pattern matching, contextual reasoning) is already solvable by large language models with multimodal inputs. The physical execution side is the only remaining bottleneck, and that's a pure engineering problem. Not if, but when.

The Valuation Implications

If you accept this framework, the implications for how we value businesses are dramatic:

1. Knowledge work businesses (agencies, consulting, legal services, content production): Trade at compressed multiples. Think 5–8x earnings instead of 12–15x. Terminal value exists but is significantly reduced. Cash flows decline as teams shrink.

2. Physical trade businesses (plumbing, electrical, HVAC, construction): Value on a 7–10 year DCF with zero terminal value. That plumbing company doing $500K EBITDA isn't worth $2.5M on perpetuity assumptions. It's worth roughly $1.8M on a truncated model.

3. High-level knowledge work (senior strategists, deal partners, expert consultants): Compressed but not zero. These roles are input- and accountability-heavy. Businesses built on genuine senior expertise retain more terminal value because institutional trust transfers slowly.

4. AI infrastructure businesses (chips, cloud, tooling, model providers): These are the only category where terminal value may actually expand. Don't confuse "uses AI" with "benefits from AI."

The value that gets destroyed doesn't disappear. It migrates. Capital flows from the businesses AI replaces toward the infrastructure that powers AI, the platforms that orchestrate it, and the human roles that sit at the trust and accountability layer. The investable insight isn't just "what dies." It's tracking where the terminal value relocates.

Time horizon matters more than ever. A company trading at 15x earnings might be cheap if it has a 20-year runway. It's catastrophically expensive if it has a 5-year runway.

The Conversation Nobody Wants to Have

Right now, governments and institutions are pushing a "learn a trade" message as the answer to white-collar AI displacement. This is good short-term advice for individuals. It's catastrophic long-term policy.

We're telling an entire generation to retool for careers that face the same existential risk on a slightly longer timeline, and with a harder landing when it arrives. A laid-off marketing manager who becomes a plumber has maybe a decade of good earnings before facing the same cliff, but this time with no augmentation fallback.

The people who understand this problem best aren't sounding the alarm. They're racing to win it.

Dario Amodei, CEO of Anthropic, is one of the few technology leaders actively warning about these dynamics from a position of deliberate restraint. Elon Musk identified the same risks years ago. He famously called AI "opening Pandora's box." But Musk has since concluded that the only way to manage the risk is to be the one holding the controls. Which requires winning the capital and governance race across Tesla, xAI, SpaceX, and the rest of his portfolio. His silence on AI safety isn't ignorance. It's strategy.

One says slow down and think. The other says go faster so I can steer. Both positions are internally coherent. Neither produces the policy conversation society actually needs.

What to Do About It

If You're Buying or Investing in Businesses

Stop using perpetuity multiples without an AI stress test. For every target, ask:

• Is the core output deterministic (pass/fail) or probabilistic (spectrum of quality)?
• Is the remaining human value in diagnosis, execution, or accountability?
• What does valuation look like with a truncated (not perpetual) terminal value?
• Does the business face a slope (gradual margin compression) or a cliff (binary displacement)?

If You're Building a Career

The safest positions aren't in trades and they aren't in easily-automatable knowledge work. They're at the edges of complex coordination: roles where you're navigating ambiguity between humans, institutions, and systems. Specifically: deal origination, stakeholder alignment across competing interests, regulatory strategy where the rules are still being written, and novel problem framing where the question matters more than the answer. Work where the output can't be verified against a simple success criteria.

If You're Making Policy

Start having honest conversations about timelines. "Learn to code" was bad advice for the wrong reasons: not because coding wasn't valuable, but because it assumed permanence. "Learn a trade" is making the same mistake. Good policy requires distinguishing between what buys people a decade of earnings versus what builds a career that lasts.

The Bottom Line

For 50 years, the fundamental assumption in business valuation has been that companies exist in perpetuity. Terminal value, that mathematical abstraction representing "forever," has accounted for the majority of what we think businesses are worth.

AI is stress-testing that assumption for the first time.

The businesses that survive won't just be the ones that "adopt AI." They'll be the ones whose core value proposition can't be replicated by a model that improves 10x every 18 months. And the crucial question isn't when your industry gets disrupted. It's whether you face a slope or a cliff when it does.

The math is simple. If AI can do what your business does in 5 years, your business is worth 3x earnings today, not 15x. If your output is deterministic, there is no augmentation phase to cushion the fall. The market hasn't figured this out yet. The question is whether you will before it does.

When terminal value goes to zero, so does most of your valuation. The only question is whether you face a slope or a cliff. Plan accordingly.

---

Appendix: The Math

The Standard DCF Formula

$PV = \sum_{t=1}^{n} \frac{FCF_t}{(1+r)^t} + \frac{TV_n}{(1+r)^n}$

Where:

• $PV$ = present value of the business today
• $FCF_t$ = free cash flow in year $t$
• $r$ = discount rate (typically WACC, usually 8–12%)
• $n$ = number of explicit forecast years
• $TV_n$ = terminal value at end of year $n$

Terminal Value (Gordon Growth Model)

$TV_n = \frac{FCF_{n+1}}{r - g}$

Where $g$ = perpetual growth rate (typically 2–3%). This single number, the value of "forever," accounts for 60–80% of most business valuations.

The AI-Truncated DCF

When AI obsoletes a business in $n$ years, terminal value is zero:

$PV = \sum_{t=1}^{n} \frac{FCF_t}{(1+r)^t}$

Worked Example

Business generating $1M/year. Discount rate $r = 10\%$. Growth rate $g = 2\%$.

Traditional (perpetuity):

$TV_5 = \frac{\$1M \times 1.02^6}{0.10 - 0.02} = \frac{\$1.126M}{0.08} = \$14.08M$

$PV(TV_5) = \frac{\$14.08M}{(1.10)^5} = \$8.74M$

$PV(FCF_{1-5}) = \$3.79M$

$Total = \$3.79M + \$8.74M = \$12.53M$

Truncated (5-year lifespan):

$PV = \$3.79M$

68% value destruction from one assumption.

Sensitivity Table: Lifespan × Discount Rate

Present value of $1M/year free cash flow:

The gap between 5-year lifespan and perpetuity at 10% discount rate: $8.71M on a business making $1M/year.

The Key Variable Nobody Is Debating

Finance argues about $r$ (discount rate) and $g$ (growth rate). Moving $r$ from 10% to 12% on a perpetuity changes valuation ~20%. Moving $n$ from perpetuity to 5 years changes valuation ~70%.

$\Delta V_{r \pm 2\%} \approx 20\% \quad \text{vs.} \quad \Delta V_{n: \infty \to 5} \approx 70\%$

AI is compressing $n$. That's a more fundamental repricing than anything the cost of capital can account for.

Modeling the Two Disruption Modes

Knowledge work (slope): declining cash flows:

$PV = \sum_{t=1}^{n} \frac{FCF_1 \times (1 - d)^{t-1}}{(1+r)^t}$

Where $d$ = annual margin compression rate (10–15%/year). Business survives but shrinks.

Physical trades (cliff): stable then zero:

$PV = \sum_{t=1}^{k} \frac{FCF_t}{(1+r)^t} + 0$

Where $k$ = years until robotic execution is viable. Full cash flow, then nothing. No signal to adapt before the drop.

Implied Multiples by Business Type

---

Last updated: February 10, 2026