An adjusted profit needs a bridge before it becomes a benchmark

Bloomberg, carried by Fortune, reports that Anthropic achieved positive adjusted operating income in the second quarter. That reported milestone is worth investigating.

I would keep the word “adjusted” attached every time I repeat it. The word does not invalidate the result. It tells us there is a definition to inspect before using the figure as a benchmark for another business.

The discipline is especially useful when an AI profitability headline becomes a claim about the economics of a whole sector. Without a bridge to the underlying operating result, the comparison can quietly depend on adjustments that differ between companies.

Unknown adjustments should stay unknown

An adjusted operating result is not automatically EBITDA. It is also not safe to assume that a particular compute cost has been removed merely because the company operates in AI. The exclusions have to come from the actual definition and reconciliation.

My rule is simple: if I cannot identify an adjustment, I will not supply one from the surrounding narrative. “They must have excluded the GPUs” is not analysis. It is an unsupported explanation for a number whose construction has not been established.

As an illustration, suppose I am reviewing a hypothetical supplier with an operating loss of 100 units and adjustments totalling 130. Its adjusted operating profit is 30. That arithmetic is straightforward; deciding what the 30 tells us is the real work.

I would ask what makes up the 130, whether each item recurs, which period it relates to and why management considers the adjusted view useful. A noncash item can still reflect an economic cost. An unusual cash payment can distort a period's comparison without disappearing from the company's funding requirements.

The purpose is to understand the measure, not to reject all adjustments. A well-explained bridge can make a result easier to interpret. An unexplained endpoint leaves several economically different stories compatible with the same headline.

Put comparability ahead of the league table

Imagine the next slide compares that hypothetical supplier with a competitor reporting unadjusted operating income. I would stop the ranking until the definitions were aligned. Otherwise the slide could reward a difference in presentation as though it were a difference in operating performance.

The same applies to periods. A quarterly result, an annual forecast and a revenue run rate cannot be placed beside one another as interchangeable observations. Each may be useful, but the units and time boundaries must survive the comparison.

For a practical review, make a compact working table with the reported measure, underlying measure, adjustments, period and source. Leave a field blank when the evidence is unavailable. That blank is more informative than a guessed normalisation presented to two decimal places.

Then write the conclusion the available evidence actually supports. It might establish a milestone under a specified management measure while leaving cash generation, capital requirements and cross-company comparability unresolved. Those are separate questions requiring additional disclosures.

I would also keep this distinction when evaluating an AI vendor operationally. A financial headline can prompt closer diligence, but it cannot replace an assessment of service continuity, contractual commitments and the particular dependency a customer is taking on.

There is no need to turn incomplete disclosure into an accusation. Ask for the bridge, inspect it when available, and change the conclusion if the evidence warrants it. The work starts with preserving the label on the number.

An adjusted profit becomes a useful benchmark only when the adjustments travel with it.