A discounted cash flow model of Anthropic, built ahead of the company's anticipated initial public offering, lands on an equity value of $1.45 trillion to $1.50 trillion, just below the rumored IPO pricing band of $1.5 trillion to $2.0 trillion. When the model was opened up for public challenge, the loudest objection was not the revenue forecast or the governance structure. It was the 9.74% cost of capital. That objection, on closer inspection, turns out to be the weakest part of the pushback, while the assumptions that genuinely deserve scrutiny sit elsewhere. Sorting the two apart is a useful exercise for anyone trying to value an AI company, because the same discount rate that values a firm from the outside also sets the hurdle for its investment decisions on the inside.
The model in brief
The valuation tells a story of Anthropic as a dominant winner in a rapidly expanding AI market. The headline inputs:
| Input | Value |
|---|---|
| Year 10 annual revenue | $1 trillion |
| Operating margin before regulation | 32.5% |
| Regulatory drag on margin | 5 percentage points |
| Effective operating margin | 27.5% |
| Revenue per $1 of reinvested capital | $3 |
| Cost of capital | 9.74% |
| Probability of failure | 10% |
| Equity value | $1.45 trillion to $1.50 trillion |
A few mechanics are worth spelling out because they drew questions. Any per-share figure in a pre-IPO filing is a placeholder until bankers settle the share count and offer price; changing the count moves the per-share quotient but not the total equity value. The 9.74% rate embeds a synthetic cost of debt consistent with a BB+ credit rating, which is below investment grade, and a recent Financial Times report described Anthropic's credit profile in the same tier. The prospectus discloses roughly $450 billion in contractual compute commitments plus several hundred billion in flexible ones; the contractual portion is discounted to a present value of about $373 billion and treated as debt. Existing restricted stock and options were valued and deducted, and dilution from future capital raises is already captured through the present value of negative free cash flows, so adding a second penalty would double count it.
Ranking the assumptions by fragility
Rather than defend the number, the more honest approach is to play devil's advocate and rank each narrative input by how easily it could break. That requires a parsimonious model, roughly 15 to 20 line items, so the path from story to terminal value stays traceable; a 400-line spreadsheet hides those links.

▲ Ranking valuation assumptions by fragility
1. The $1 trillion revenue target
This is the single most fragile pillar. A $1 trillion top line would make Anthropic larger by revenue than any operating company in the world today. It rests on four stacked assumptions: a total addressable market for AI of $5 trillion to $10 trillion by year 10, driven by AI moving from assistive software to direct labor replacement; a meaningful share of that market reserved for premium frontier models rather than commoditized tools; Anthropic winning and holding a leading share of that premium tier; and Anthropic pushing downstream into vertical businesses such as proprietary drug development. Each has a mirror-image failure mode. AI could remain a peripheral productivity tool, open-source models could erode premium pricing, share could splinter across competitors, or the company could lack the operational capability to run vertical businesses. Google's autonomous driving work offers an analogy: developing the technology did not translate into becoming a ride-hailing operator, because the logistics of running the business are a different problem from building the technology.
2. Governance
Anthropic is structured as a public benefit corporation with a multi-class share structure that leaves voting control with its seven founders regardless of how much outside capital comes in. Some investors see this as an asset, since technical founders understand AI best. The counterargument is that scaling to this size will involve technological, legal and competitive setbacks that require changing course and admitting error, and that a group sharing a common background and worldview, with no external check, may find that harder. On this reading, the structure warrants a deeper discount below intrinsic value before buying.
3. Profitability
The 32.5% pre-regulation margin depends on three things: gross margins at or above 80% as compute shifts from a variable to a fixed cost; fixed costs and headcount growing much more slowly than revenue; and a durable moat. In technology, that moat is customer stickiness rather than brand. Microsoft's Windows and Office entrenched themselves as business standards despite Apple's arguably superior early operating system, because switching costs locked customers in. The base case assumes Anthropic achieves similar stickiness by embedding premium products and customer data in workflows, which could fail if models commoditize.
4. Compute and reinvestment
The $3-of-revenue-per-$1-invested ratio assumes existing compute commitments cover near-term demand and that future capacity gets cheaper as accelerator chips become easier to manufacture. A compute crunch driven by power and infrastructure shortages would raise unit costs, and an aggressive push into capital-heavy domains like drug discovery would raise reinvestment needs.
5. Regulation
Regulatory risk is arguably the most worrying variable precisely because it follows political dynamics rather than market economics. The model shaves 5 percentage points off margins, but more onerous mandates could cut far deeper.
Notably, the cost of capital does not top this list. If Anthropic fails to scale revenue and reach operating profitability, it will fail regardless of the discount rate.
What the crowd said
About 70 people submitted their own valuations through an open spreadsheet version of the model. Roughly 65% to 70% came in below the baseline figure, and only two exceeded $2 trillion. In a parallel poll, 51% said the valuation was too high, 30% said it was about right and 19% said it was too low. Among those calling it too high, 41% cited gut instinct that no AI company could be worth that much, 35% said the 9.74% cost of capital was too low, 10% doubted the revenue target, 10% feared a regulatory crackdown and 4% saw no moat.
The revenue, regulation and moat objections map directly onto the fragility ranking above and are the productive kind. The cost of capital objection is the one that calls for a reality check, because the discount rate tends to become an emotional receptacle for every vague fear about a company, even though business drivers move value far more.
Three reality checks for a cost of equity
The cost of capital is a weighted average of the cost of equity and the after-tax cost of debt. The cost of equity is the risk-free rate plus a risk premium; the cost of debt is the risk-free rate plus a default spread, reduced by the tax rate. The cost of equity is fundamentally an opportunity cost, what an investor could earn elsewhere at equivalent risk, so it has to live in the actual market rather than in a wish. Three checks anchor it.
History
US equity returns averaged 8.74% a year over 2001 to 2025, 11.92% over 1976 to 2025, and 10.13% over the full 1928 to 2025 span. The geometric equity risk premium over Treasury bonds since 1928 is 5.54%; added to a 10-year Treasury yield of about 4.90%, that implies roughly 10.44% for an average firm. And US history already flatters the case, since the US was the most successful equity market of the twentieth century.
Market-implied returns
Solving for the internal rate of return that equates the S&P 500 at 7,651.54 on September 30, 2026 to expected cash flows gives an implied expected return of 8.99%, with early November updates at 8.83% and 8.76%. Subtracting the 4.90% risk-free rate yields an implied equity risk premium of about 3.70% to 4.09%. Looking back to 1960, implied returns only exceeded 15% in the late 1970s and early 1980s, when Treasury yields were 10% to 12%, and they have not topped 10% at any point since 1994. Demanding 15% to 20% today is a memory of that inflationary era.
Pricing multiples
Rearranging the Gordon Growth dividend discount model links cost of equity to a forward price-to-earnings ratio. Assume 5% nominal growth, a generous 20% return on equity and a 15% cost of equity, and the justified forward PE is 7.27. The market trades near 20. Anyone holding a 15% required return therefore faces a contradiction: sell every stock they own, or revise the rate. In practice an average US firm carries a cost of equity of 8% to 9%, and a lower cost of capital because debt is cheaper.
Debt, double counting and the loop
On the debt side, worries about debt-fueled data center construction, with CoreWeave as the most-cited example of heavy leverage, have bled into how people think about Anthropic's borrowing cost. But a cost of debt is a going-concern number, the marginal rate at which the company can borrow today, not a liquidation yield. Arguing that Anthropic must borrow at 10% or more conflates ordinary default spreads with catastrophic failure. The model expects a BB or BB+ rating after the IPO, implying a pre-tax cost of debt of about 6.5%, supported by a cash balance moving from $114 billion before the offering to over $200 billion after it, against commitments of about $60 billion next year and $76 billion the year after. Catastrophic failure belongs in an explicit probability, which is why the model carries a separate 10% failure rate, well below the 40% to 50% typical of early-stage startups.
Cross-sectional data for September 2026 shows how compressed discount rates really are. In information technology, 80% of firms fall between 9.88% and 14.50%, with a median of 10.75%; in energy the band runs from 7.16% to 10.96%, with a median of 7.6%. Fine-tuning a rate to two decimal places is misdirected effort.

▲ Market price, cost of capital and investment loop
Two caveats apply to anyone who still wants to push Anthropic's rate to the 90th percentile of tech, around 12%. First, drop the separate failure probability to zero; the baseline 9.74% combined with a 10% failure rate already imposes a penalty equivalent to roughly an 11% cost of capital, and stacking 12% on top of a 10% failure rate implies something closer to 13.5% to 14%. Second, a cost of capital cannot stay static for a company modeled to grow from under $100 billion in revenue toward $1 trillion while turning losses into large profits. As the firm matures, its rate must decline toward industry norms.
This is why the number matters beyond the valuation debate. The cost of capital works as the Swiss Army knife of finance. Inside a company, it is the hurdle rate that decides which projects create value, the benchmark for the right mix of debt and equity, and the test for whether cash should be reinvested or returned to shareholders through dividends and buybacks. From the outside, it becomes the discount rate applied to projected cash flows. A rate pulled from gut feeling corrupts every one of those uses at once.
Why inflated hurdle rates hurt everyone
The same error shows up far beyond Anthropic. Corporate surveys put hurdle rates two to three percentage points above market-justified costs of capital, and several CFOs responding to this valuation said their firms use about 14%. When a company's true cost of capital is 9% and management demands 15%, division managers inflate growth and margin forecasts to clear the bar, projects with negative true net present value pass on paper, and CFOs respond by raising the hurdle to 18%, so made-up cash flows get discounted at made-up rates. Global data across fifteen years of annual updates shows fewer than 30% of public companies earn a return on invested capital above an 8% to 9% cost of capital; at a 12.5% hurdle, fewer than 5% would.
Venture capital and private equity funds market target returns of 25% to 50%, yet realized returns across horizons from one to thirty years through September 2026 track, and often lag, the S&P 500 and NASDAQ. Individual investors demanding 20% find no undervalued company in the S&P 500 outside of distressed names, drift into cash drag or, more dangerously, into schemes promising exactly those returns. Private business owners face appraisers using a build-up method that layers on a small-cap premium absent from the data since about 1981, double-counts illiquidity in both the rate and the final value, and adds unsupported company-specific premiums, leaving sellers to hand value to public buyers with far lower costs of capital. Rates of 15% to 20% are justified only in high-inflation currencies such as the Turkish lira or Zambian kwacha, or in extreme country-risk settings like Afghanistan or South Sudan.
How to use this
A valuation of Anthropic will be wrong in hindsight; every long-horizon forecast is. The useful question is where it is most likely to break, and the answer is the revenue path, the governance structure, margins, compute costs and regulation, in roughly that order. If you want to disagree with the number, argue about whether AI becomes a labor substitute or stays a productivity tool, whether premium models keep pricing power, and whether a founder-controlled public benefit corporation can correct course.
If your disagreement is about the discount rate, run the three checks first. Compare your cost of equity to the 9% to 10% historical average, to the roughly 9% implied by current index prices, and to the forward PE near 20. If you still want 12%, remove the failure probability and let the rate fall as the company scales. Then decide how to act: a margin of safety of 20% to 40% trims the risk of buying the wrong stock but raises the risk of missing the right one, and pushing it to 40% to 80% leaves portfolios 50% to 70% in cash. Holding a highly uncertain name like Anthropic inside a basket of 20 to 25 positions lets estimation errors cancel out, which is a better defense than a discount rate chosen to match a feeling.


