Anthony Scilipoti on Forensic Accounting, the AI Bubble, and Spotting Corporate Collapse

Anthony Scilipoti is a forensic accountant who built an independent research firm, Veritas Investment Research, on two calls almost nobody wanted to hear: sell Nortel in 2000 and sell Valeant Pharmaceuticals in 2012, both years before each collapsed. In this conversation with Shane Parrish, he turns the same footnotes-first, cash-flow-over-narrative method on the AI boom — reading circular financing between Nvidia, OpenAI, Microsoft, and CoreWeave, and the accounting shortcuts that let a euphoric market look past the numbers.

Key ideas

1. Circular financing is the AI boom’s central symptom, and it rhymes with the dot-com telecom bust. Scilipoti draws a direct structural parallel between Nortel, Lucent, and Cisco vendor-financing their own telecom customers in 1999–2000 and today’s Nvidia–OpenAI–Microsoft–CoreWeave web: Nvidia invests in OpenAI and supplies its chips; Microsoft invests in OpenAI and sells it cloud capacity; Nvidia is also a meaningful investor in CoreWeave, which buys Nvidia’s chips and sells capacity to Microsoft. ‘No one’s doing anything bad. No one’s cheating’ — but the same money is circulating through supplier, customer, and investor relationships at once, the way it did before Nortel’s wheels came off.
2. His ‘flammable items’ framework replaces red flags with a three-stage test: environment, item, spark. Stage one is understanding the business, its control environment, and how management is paid. Stage two is spotting a flammable item — negative cash flow, an aggressive non-GAAP metric, a change in how adjusted EBITDA is calculated year to year — which is not, by itself, a problem (a company burning cash to fund an investment returning 20%+ is fine). Stage three is watching for the spark — a new competitor, a funding-market wobble — that ignites the flammable item into an actual collapse.
3. He reads the footnotes before the financial statements, because that is where management discloses its accounting choices. The notes tell you how a company decided to account for a transaction — capitalising a cost versus expensing it, for instance — and only once you know that can you correctly interpret the numbers that follow. Nortel’s own footnotes classified long-term customer receivables as a long-term asset rather than a current one, which kept the exposure out of both the current ratio and operating cash flow — a disclosure choice almost no one read closely until it mattered.
4. Free cash flow has no fixed formula — ‘it depends’ on the facts, the constraints, and the objective. Scilipoti rejects the textbook shorthand of operating cash flow less capex as too rigid: Nortel’s structure made its free cash flow look healthy when the underlying customer-financing exposure made it negative. He teaches investors to derive the metric from the specific facts of a transaction, the constraints on the business (public versus private, its regulator, its debt covenants), and the objective the number needs to serve, rather than applying a classroom formula by rote.
5. EBITDA earns his harshest line: ‘the mother of all disastrous measures.’ It measures operating performance before interest, tax, depreciation, and amortisation — nothing more — yet investors treat it as a proxy for cash flow or debt capacity. The distortions compound when companies redefine adjusted EBITDA from one year to the next, or fold recurring acquisition costs back in as one-off add-backs — a pattern he flags as one of the biggest flammable items of all, and one he says was central to Valeant’s collapse.
6. The Valeant call shows why forensic scepticism is lonely and slow to pay off. Veritas held the only sell rating on Valeant in 2012–13; the company did not blow up until 2015. Scilipoti’s explanation for why sophisticated, well-respected investors missed it: ‘price creates narrative’ — each successful acquisition and each rise in the share price reinforced belief in management’s story, making it progressively harder for underperforming sceptics to hold their position. He suggests the same dynamic is now protecting AI-boom narratives from scrutiny of the underlying financial statements.

See also

• Alex Sacerdote on S-Curves, the AI Boom, and Finding Technology Winners — the bull case for the same boom, argued from adoption curves and competitive moats rather than financial statements