Anthropic just completed a $93.5 billion public offering, including the underwriters' extra shares. The amount raised is staggering. The valuation is worse. At the $83 IPO price, the prospectus's pro forma diluted share count, including the full overallotment and its IPO option assumptions, implies roughly $1.5 trillion of equity value. That is about 43 times revenue for the twelve months ended September 30, or almost 19 times the company's approximately $81 billion October annualized revenue run-rate. From a value investor's perspective, the price is fucking absurd.
The growth is extraordinary, so a value investing framework would always look silly, but let's play around with it anyways: using last year's $4.6 billion of revenue implies a roughly 331x revenue valuation. At that unchanged revenue level, common shareholders would need the company to distribute 100% of revenue, with no expenses at all, for 331 years just to get their money back. That is absurd. This company is wildly, fabulously unprofitable on a GAAP earnings basis. In my view, common stock investors will never see their cash back through distributions unless pretty absurd assumptions come true on revenue growth or cost reductions. The question I care about is how much durable cash this business can return to shareholders after paying for the continuing race to build and serve better models. Forty-three times trailing sales asks me to pay heavily today for exceptional growth and pricing power far into the future. I see very little margin of safety.
In our Second Mind business, clients largely do not care whether the underlying model is Opus, Astra, Sol, Kimi or GLM. Often they cannot tell the difference. They care whether the work gets done accurately, reliably and at a sensible cost. For much of that work, the models are already smart enough. That is why I view the underlying token business as a commodity business and do not see a durable model moat that justifies this valuation. When a customer can get an acceptable result from several suppliers, competition pushes the economic value toward the customer. Model providers still have to fund the next training cycle while defending their prices against everyone else's next release.
Anthropic has real distribution, enterprise integrations and compliance capabilities. Those can make switching harder, and its own applications can capture value beyond a raw API call. I am unconvinced those advantages protect the level of long-term pricing power embedded in this price. We can build a very useful business with these models while declining to own the model company at almost any price. I expect AI adoption to grow enormously. I also expect competition to make capable intelligence cheaper. At this valuation, Anthropic shareholders need those two forces to resolve very generously in their favor. I would rather sell useful outcomes than pay a monopoly price for tokens.
Happy Friday.