Anthropic Uses October Revenue to Price a $2 Trillion Valuation, and This May Be the Most Absurd Principle in Investing

Author: Byron Gilliam (Blockworks)

Translated by TechFlow

TechFlow Editor's Note:In its closely watched S-1 prospectus, frontier AI giant Anthropic pulled off an astonishing valuation sleight of hand: annualizing the API consumption costs of the most recent 28 days by multiplying them by 13, adding the current monthly subscription fees multiplied by 12, and even leaving a conspicuous blank in a key spot, waiting for the latest single-month data for October to underwrite a potential $2 trillion astronomical valuation.

Blockworks senior macro analyst Byron Gilliam borrowed the Heisenberg Uncertainty Principle from quantum mechanics to sharply expose the most ironic valuation paradox in financial markets: the more precisely a financial metric can be measured, the less useful it is in gauging a company's long-term intrinsic value. Unable to forecast cash flows for decades ahead, investors settle for next year's price-to-earnings ratio (P/E); when facing huge losses, they retreat to the price-to-sales ratio (P/S); and amid the frenzied AI capital arms race, the market has even retreated to relying on annualized data from just one month.

A global semiconductor giant's stock rally hinges on a blank cell in a prospectus, making this the most absurd valuation pricing in investment history, yet also the unavoidable compromise of capital in a frenzied cycle.

The Heisenberg Principle of Investing

According to the laws of quantum mechanics, matter has wave-like properties: each discrete particle also exists as a probability wave.

This makes particles elusive. To know a particle's exact position, you must collapse its wave nature into a single point, but that causes it to lose a definite wavelength; to know its momentum, you must let the wave spread freely through space, but that causes the particle to lose a fixed position.

This is the famous Heisenberg Uncertainty Principle: the more precisely you know a particle's position, the less you know about its momentum.

A similar dynamic plays out in financial markets.

The most fundamental principle of investing is that the intrinsic value of any asset equals the net present value of all its future cash flows.

Logically, the most accurate way to value a company's stock would be to forecast all of its future cash flows.

But this is an almost impossible task. Who can accurately predict a company's earnings ten or twenty years from now? Who can even guarantee that it will still exist then? Typically, the limit of what we can achieve is forecasting earnings for the next one to two years.

Therefore, the trading logic in secondary markets follows suit. Stocks are usually valued at a multiple of expected earnings for the next year (price-to-earnings ratio, P/E), because this is the only metric that can be calculated with reasonable precision.

However, precision comes at a painful cost to usefulness: a short-term earnings forecast for just one year captures only a tiny fragment of a company's long-term intrinsic value.

This is the trade-off investors must face: the more precisely a financial metric can be measured, the less useful it is in determining an asset's true intrinsic value.

It may be called the Heisenberg Principle of Investing.

Take global memory chip giant SK Hynix as an example: its trading price corresponds to roughly 4 times expected earnings for the next year, a P/E that appears implausibly cheap on the surface.

But how useful is this metric as a gauge of the stock's intrinsic value? In fact, it is very limited. In the semiconductor industry, where cycles swing violently, enormous profits in one year can quickly turn into enormous losses in the next.

Yet this violent cyclical volatility is precisely why we are forced to use short-term forecasts to value SK Hynix: the true core driver of the stock's long-term intrinsic value is the supply-demand balance in the long cycle of global memory chips, which is impossible to predict accurately in reality.

And when a company does not even have positive earnings at present, the difficulty of valuation rises exponentially.

For persistently loss-making companies, the usual approach is to work upward along the income statement (P&L) until finding a positive number to apply valuation multiples to, such as revenue.

For example, cybersecurity software giant CrowdStrike is essentially operating without meaningful profits. Even so, it still commands a market capitalization of $274 billion, as investors choose to value it at about 40x expected revenue for the next year, a 40x price-to-sales (P/S) multiple.

A 40x P/S multiple is admittedly expensive, but cybersecurity is a red-hot sector, and without profits, revenue is the only metric we can use as a valuation anchor.

But what if even next year's revenue becomes elusive and unpredictable? In that extreme scenario, we may have to settle for current actual financial data as a fallback.

For example, frontier AI giant Anthropic is actively steering investors to value its equity at a multiple of its 'current revenue.' However, the precision of this measure is far more ambiguous than it sounds. What exactly does 'current' mean: this year's revenue, this month's revenue, or even just today's revenue?

In the widely anticipated S-1 registration statement, Anthropic provided its customized formula for calculating 'annualized run-rate revenue' (Run-rate Revenue):

This is an operational metric intended to reflect the estimated annualized value of our revenue-generating activities as of the measurement reference date. It is calculated as the sum of: (1) API usage fees incurred during the 28-day period immediately preceding the reference date, multiplied by 13; and (2) total monthly subscription fees for the current month as of the reference date, multiplied by 12.

In other words, Anthropic wants investors to annualize its most recent month of API revenue, which is based on pay-as-you-go on-demand usage, and to annualize its most recent day of subscription revenue, which is based on recurring subscription renewals.

Then it wants you to apply a highly ambitious multiple to that annualized figure, likely 15x or even 20x.

But wait: none of this will be revealed until November!

In the prospectus filed yesterday, a conspicuous blank was deliberately left, and Anthropic promised to fill in the figure as soon as the annualized data for October becomes available:

As the AI community often jokes, this small blank is carrying an unbearable weight. The success or failure of Anthropic's hundred-billion-dollar IPO hinges almost entirely on the number revealed by this blank.

This is also the core logic behind the sharp rally in global semiconductor stocks this week: market investors are convinced that if Anthropic were not absolutely confident that October's data would be impressive enough, it would not have deliberately left such a conspicuous blank in its prospectus.

If that logic holds, Anthropic will raise more capital to aggressively purchase semiconductor chips; and as long as it buys more chips, manufacturers such as SK hynix will earn fatter profits, thereby lowering their forward P/E multiples, which is exactly how we initially valued semiconductor giants.

Is this really a healthy way to invest? Can the market capitalization of an entire industry-chain sector be moved and repriced based solely on another private company's single-month operating performance?

The answer is clearly no. Crude annualization of just one month's revenue has slid to the extreme end of the 'Heisenberg principle of investing,' offering no practical valuation utility.

Pricing a $2 trillion-scale behemoth based on annualized run-rate revenue (Run-rate) is arguably the most absurd valuation trick.

Unfortunately, this is also the best approach we currently have to offer.

Source: TechFlow