The most popular objection to the $1.45 trillion Anthropic valuation is also the weakest. Critics say the 9.74% cost of capital is far too low. The number that actually decides the valuation sits elsewhere: a single revenue assumption, a trillion dollars within ten years, whose failure would drag margins, compute utilization and risk down with it. When an investor raises a discount rate to 15% to express skepticism, the result is an internally inconsistent model that hides the real bet instead of pricing it. Raising the rate is not conservatism.
The interesting part of the author's self-critique is not that he defends his number. It is the mechanism he names. His value drivers do not fail independently. They fail together, in the same direction. A discount rate cannot represent that kind of correlated failure. It can only blur it.
The Discount Rate Is the Wrong Place to Put Doubt
The revised valuation moved only slightly, from $1.475 trillion to $1.45 trillion. The author folded in leaked third-quarter figures and lowered year-one growth so that forward twelve-month revenue still lands near $100 billion. The cost of capital of 9.74% combines a risky-technology cost of equity with a cost of debt priced off a BB+ rating. That places Anthropic near the 90th percentile of US companies in 2026. A rate at the 90th percentile is not a rate that treats the business as safe.
The crowd still disagreed. Across inputs submitted to his shared spreadsheet and comments on several platforms, the consensus value clustered near $1 trillion rather than $1.5 trillion. Apart from blanket dismissal, the most common reason given was a cost of capital that was too low.
The rate draws the anger for a simple reason. It is the one input that feels like a risk dial. If you are uneasy, you turn it up. Valuation does not work that way. A cost of capital measures an opportunity cost: the return available elsewhere on an investment of equivalent risk. That claim can be checked against markets, and when it is checked, the 15% camp loses badly.
What a 15% Hurdle Implies About Everything Else
There are three independent checks, and they point the same way.
- History. US stocks have historically earned premiums of roughly 5.5% to 5.75% over the 10-year Treasury. With the 10-year at 4.9% on October 26, 2026, that implies a cost of equity of about 10.4% to 10.65% for an average-risk firm. That figure comes from the most successful equity market of the past century, so survivorship pushes it higher than a fair forward estimate, not lower.
- Market-implied. The author backs an expected return out of the S&P 500's current price and cash flows each month. At the start of October 2026 it stood at 8.99%. It has not exceeded 10% since 1994. Reaching 15% to 20% requires going back to the early 1980s, when the 10-year yield itself sat at 10% or more.
- Pricing multiples. This check is the most damaging to the critics, and it is easy to reproduce. Assume nominal growth of 5% and a generous 20% return on equity, which implies a 75% payout ratio. A stable-growth dividend model then gives a P/E of 0.75 × 1.05 / (r − 0.05). At r = 15%, the market should trade near 7.9 times earnings. At r = 9%, it should trade near 19.7 times. The actual forward multiple is close to 20.
The arithmetic leaves the 15% advocate with a hard choice. Either the entire US market is overvalued by roughly 60%, or 15% is not the opportunity cost of equity today. Holding both views at once, a 15% hurdle alongside a belief that 15 times earnings is a fair multiple, is a contradiction, not caution. The same logic applies to enterprise value. A 12% cost of capital implies market-wide multiples near 9 times operating income and 6 to 7 times EBITDA, far below where those multiples trade.
The rates channel matters here because the risk-free leg is not small. At a 4.9% Treasury yield, a 15% cost of equity requires a risk premium of roughly ten percentage points. That is close to double the historical premium over the 10-year bond, and the critics are applying it to a company already rated at the risky end of the distribution.
Revenue Is the Variable That Drags the Others
If the rate is not the fragile input, the trillion-dollar revenue target is. The author ranks it first, and the reason is structural. That figure would put Anthropic at or near the top of global companies by revenue, alongside Amazon and Walmart, which reach that scale only with thin margins. Anthropic's story needs scale and a 32.5% target operating margin together. That combination depends on three things: agentic AI delivering real economic work, that demand being monetized, and Anthropic winning the premium tier of the market with pricing power.
Any one of these can break, and a break spreads through the model. Lower revenue growth leaves contracted compute idle, and idle compute pushes margins down. Weaker margins slow the path to profitability, which raises failure risk. The author makes a subtle point here: risk in his model is largely an output, not an input. Anthropic becomes riskier because revenue disappoints. A disappointing revenue line is not the result of a higher discount rate.
This is where the crowd's correct instinct turns into the wrong edit. Someone who believes revenue tops out at $500 billion should cut revenue to $500 billion. That change feeds lower margins, stranded commitments and higher distress probability through the model in a coherent way. Raising the discount rate instead discounts the full trillion-dollar story harder. The valuation still assumes the success case and only treats it with more suspicion, so the model can no longer show which part of the story the investor actually rejects.
Moats are part of the same chain. The strongest critique is commoditization. The author's reply is that technology moats come from switching costs, not from brands or patents. Claude agents embedded in client data on customers and employees could become sticky enough that customers stay even when cheaper alternatives exist. That stickiness is not proven yet. It is also a revenue and margin question, which is another reason the disagreement belongs in the operating inputs and not in the discount rate.
Commitments, Dilution & the Double-Count Trap
The second confusion comes from the debt-fueled data-center boom. CoreWeave has become the example of AI infrastructure that abandoned corporate-finance first principles, and Anthropic's compute commitments have pulled it into the same conversation. The critics' syllogism runs like this: if Anthropic would pay 10% or more to borrow, its equity must cost much more.
The author handles these exposures directly rather than through the rate. He estimates contractual compute commitments at just over $450 billion in aggregate, with a present value of about $373 billion. Outside estimates range from $400 billion to $800 billion, because the prospectus scatters the figures through the text instead of tabulating them by year. Those commitments sit in the cash flows and the balance sheet, which is where they belong.
Dilution is handled the same way. Employee options are valued at option value, $472 million, and netted out. Restricted shares are added to the share count. The convertible bonds are carried at their mark-to-market value, the same marking that produced a $72 billion hit to 2026 earnings, which strips the conversion option out of equity. Future share issuance to cover negative free cash flow is deliberately left out of the share count, because those negative cash flows already reduce present value. Adding the shares too would penalize the same dollars twice.
A higher discount rate stacked on top of all this commits the same error at the level of the whole firm. The commitments, the convertibles and the funding gap are already in the numbers. Raising the rate for them a second time is double-counting, not prudence.
Where the Critics Have a Point
The best version of the opposing case is not about the discount rate. It is about failure risk, and it deserves a fair hearing.
The author keeps the cost of capital as a going-concern measure and treats default separately. That is the right design, because distressed costs of debt at the lowest ratings reflect imminent default that a perpetuity-style rate cannot represent. The design only protects the valuation if the separate failure probability is set honestly. Anthropic is a money-losing company carrying hundreds of billions in obligations, whether as debt or as contracts. The defense that it "is no CoreWeave" depends on two things: the commitments being spread over time, and a cash balance large enough to cover the first few years. The first depends on contract terms the prospectus does not clearly disclose. The second depends on revenue arriving on schedule. Both lead back to the same weak link.
Two further risks sit outside any discount-rate argument. The first is regulatory. The author calls regulation the input he is most worried about getting wrong, because it is driven by politics more than economics. One election could turn a small haircut on operating profit into a large cut to revenue. The second is governance. The public-benefit-corporation structure and founder-controlled voting shares leave little room for outsiders to force a change in course after a technological or competitive setback. Both are real, and neither is fixed by moving a rate from 9.74% to 15%. They are fixed by explicit probabilities and lower revenue paths.
So the crowd's $1 trillion figure may well be closer to the eventual outcome. If so, it will be for the wrong reason: a revenue shortfall that the 15% rate was disguising, not a market opportunity cost of 15%.
What Would Move the Number
Three observable items will test this reading before any offering price is set. The first is a firm-level schedule of compute commitments by year and contractual standing. If that schedule lands nearer $800 billion than $450 billion, the present-value drag rises sharply and the cash-runway defense weakens. The second is revenue growth after the third-quarter slowdown. Forward twelve-month revenue that cannot hold near $100 billion would be the first evidence that the trillion-dollar target is slipping, and every other input would follow. The third is the final share count. That changes the per-share price but not aggregate equity value, so it should not be confused with a change in value.
The valuation breaks on revenue, commitments or politics, and the useful disagreement is over the probability attached to each one. A discount rate set to express unease answers none of those questions.
