Every beginner reads a P&L top-down: big revenue number, must be a big healthy business. A consultant reads it sideways. The question is not "how much revenue?" but "what kind of revenue is this, and will it still be here next year?" Two companies can both report $10 billion in sales and be worth wildly different amounts, because one earns it from thousands of independent customers paying market price, and the other earns most of it from a single partner it happens to be funding.

Picture a lemonade stand. You hand your cousin $5, he buys $5 of lemonade from you, and you record $5 of revenue. Your income statement looks alive. But no new wealth entered the system. You funded your own sale. Scale that from $5 to $80 billion and add three intermediaries so the loop is harder to see, and you have the structural worry hanging over AI infrastructure today.

That is the whole skill in one sentence: trace the cash to its origin and ask whether the buyer's ability to pay depends on the seller. If it does, the revenue is lower quality than the headline suggests.

When an interviewer hands you a company with explosive, concentrated growth, do not free-associate. Run a clean, MECE diagnostic with three buckets.

Source of the cash. Where did the buyer get the money to pay? If an independent third party funded the purchase out of its own operating cash, the revenue is high quality. If the seller, or an entity the seller invested in, supplied the capital, you have related-party revenue and you discount it. Nvidia putting equity into OpenAI while OpenAI's spending flows back toward Nvidia silicon is the textbook version.

Arm's-length test. Would this transaction happen at this price and this volume between two strangers? Vendor financing, equity-for-commitment swaps, and "strategic partnerships" all fail this test to some degree. The further a deal sits from a normal customer paying a normal price, the more you haircut the reported number.

What breaks the loop. Circular structures have single points of failure. Map the chain and ask which link, if it snaps, stops the music. If Nvidia decides OpenAI is too risky to keep funding, OpenAI cannot pay Oracle, and Oracle slows its chip orders. One decision, three revenue lines impaired. Naming that fragility out loud is what makes an interviewer lean in.

Here is where candidates go wrong: they stop at "the revenue looks circular" and feel clever. The teachable move is to go one level deeper inside each bucket. Quantify it. If 40% of a chipmaker's growth traces back to customers it funded, then 40% of that growth is hostage to its own balance sheet, not to end demand. That sentence is the difference between a label and an analysis.

The strongest answers connect the live problem to a precedent, and this precedent is almost perfect. During the dot-com telecom boom, equipment makers like Lucent and Nortel booked enormous sales to upstart carriers. The catch: the equipment makers were lending those carriers the money to buy the gear. Lucent carried billions in customer financing on its books. While capital was cheap and the story was hot, revenue compounded and the stock soared. When demand failed to show up, the carriers went bankrupt, the loans went bad, and the revenue they had supported evaporated at the same time. Lucent lost more than 90% of its value. The vendor financing did not just fail to help. It converted a revenue problem into a balance-sheet problem.

The AI loop is not a fraud, and saying so in an interview would be sloppy. Real compute is being bought and real models are being trained. But the structural lesson transfers cleanly: when a supplier finances its own demand, reported growth runs ahead of true end-market demand, and the gap is invisible until the cheap capital stops. OpenAI is reportedly on track to lose around $14 billion in 2026 while projecting $100 billion in revenue years out. Your job in the room is not to predict which way it breaks. It is to show you can hold both facts at once and locate exactly where the risk lives.

Re-underwrite one hot company by hand. Take any AI or growth name in the news, find its largest customer or partner, and ask the three questions: where did their cash come from, is the deal arm's-length, and what breaks the loop. Write the answer in four lines. Doing it once on a real company beats reading about it five times.

Practice the discount out loud. In your next mock, when given a revenue figure, say: "Before I trust this number, I want to know how concentrated it is and whether any of it is related-party." That one sentence signals you think like a diligence partner, not a calculator.

Separate the story from the cash. Train yourself to ask of any impressive metric, "what would have to be true for this to be durable?" Concentration, funding source, and arm's-length pricing are the three places the answer usually hides.

The cleanest way to feel this in your body is to work a case where a beautiful growth story hides a quality problem. Open the WeWork IPO case on BoardroomIQ and try to puncture the narrative before the diligence team does. For the foundation underneath this, read unit economics, the one calculation that tells you if a business actually works, then pressure-test it against the $740B AI capex boom.