Real Options Theory: Why Traditional Valuation Breaks Down for AI Companies

Standard corporate finance approaches derive an organisation's value by discounting future expected cash flows to today. For stable and predictable organisations, a discounted cash flow (DCF) model is an acceptable method of estimating that value. However, for development-stage organisations (like Anthropic), much of the value will likely come from their potential for future growth rather than from current revenue generation. Rather than DCF, a methodology known as Real Options Theory is needed to estimate the value of Anthropic. Real options theory was developed by economists (including Stewart Myers, Avinash Dixit, and Robert Pindyck), and it also incorporates ideas similar to those used in option pricing models.

Why DCF doesn't work here

Building a DCF model means forecasting Anthropic's cash flows for the next several years and discounting them all back to today using a single rate. That's a reasonable way to value a SaaS company with steady, predictable revenue. It's a much tougher fit for a frontier AI lab. Anthropic's annualised revenue was around $1 billion in December 2024 and had climbed past $30 billion by the end of March 2026 — roughly tenfold growth in about 15 months. At the same time, the company is expected to spend close to $19 billion on training and inference compute in 2026, nearly matching that year's revenue, and it isn't expected to turn a profit until 2028. Plug numbers like that into a DCF model and you'll get wildly different valuations depending on which assumptions you tweak. So rather than treating Anthropic's future as one predictable cash-flow stream, analysts tend to break it into separate bets, each behaving more like an option than a guaranteed payout.

Future products as options, not certainties

Several of Anthropic's newer bets haven't proven themselves in revenue yet — agentic coding tools, enterprise API expansion, and cybersecurity products. There's no guarantee any of these scales, but Anthropic keeps funding R&D, compute, and talent behind them incrementally, so that if demand does take off, the company can scale fast without starting from zero. Claude Code is the clearest example: its revenue run rate hit roughly $2.5 billion in early 2026, more than doubling in just a few months, and it's reportedly behind something like 4% of all public GitHub commits during that stretch. With talk of an eventual Anthropic IPO building, the market increasingly treats these unproven product lines the way it would treat a call option — a bet on future upside — rather than something as dependable as a wage.

Why irreversibility matters

There are two important features to keep in mind when thinking about real options: they are irreversible once created, and this irreversibility raises the value of retaining flexibility. This means that capital tied up in computing, whether as a compute cluster or a machine learning model built using cloud computing resources, becomes a sunk cost that cannot be reallocated elsewhere after the investment has been made.

The investor's expectation (i.e., the price they will pay) is based not only on the outcome that is expected, but also on the degree of flexibility or potential for further investment that Anthropic (as the developer/creator of the compute cluster) has to adapt as new information regarding AI's capabilities is released.

Discount rates struggle with capability uncertainty

When discounting cash flows from AI development, one of the biggest challenges is the uncertainty about the capabilities of that development effort. Given that AI 'outcomes' are perceived as binary (i.e., the company either successfully develops additional capabilities or it does not), the classic single-discount-rate approach cannot adequately capture the risk associated with uncertainty about AI development and its ability to generate future profits for the company. Amodei stated on a public forum that it only takes a one-year delay in the company's advancement towards developing robust AI capabilities to bankrupt Anthropic, which speaks to just how intertwined the company's value is with its capabilities' paths of development as compared to a typical business model's risk profile. Therefore, real options will become the model used to price future AI capabilities, which in turn will shape how investors value information regarding the Anthropic IPO.

The market is already behaving this way

In May 2026, Anthropic raised $65 billion in Series H funding, resulting in an estimated post-money valuation of $965 billion, compared to a valuation of $380 billion just three months prior. This marks a significant jump in the company's valuation, implying an approximate 30x revenue multiple, which is an extremely high valuation when compared to traditional public businesses — but is not unreasonable when valuing companies based upon their anticipated future revenue, rather than their historical sales figures. Amazon provides a good case study; in the late 1990s, Amazon's trading was significantly higher than any number that a discounted cash flow model would produce, because investors valued Amazon based on its future potential for growth beyond just selling books. Thus, as this uncertainty continues to increase, the use of real options theory (real options pricing) will likely become the primary valuation framework for pricing future IPOs and potentially future offerings of AI products and solutions.