When AI Startups Trade Credits and Call It Revenue

Look closely at any AI startup pitch deck right now and something feels off. Revenue doubles every quarter, but operating cash flow doesn’t keep up. The whisper word making the rounds in venture circles: revswap. Mutual revenue laundering, dressed up as commerce.

How One Trade Becomes Two Sales

The mechanics are almost embarrassingly simple. Startup A hands Startup B $1M in service credits. B hands A $1M in its own credits back. Not a dollar of cash changes hands. Both companies book $1M in revenue.

For VCs, top-line growth is the metric that prints valuations. Triple-digit ARR growth can justify a 2-3x markup at the next round. Which is exactly why it matters whether that ARR reflects real market demand — or just two startups bartering vapor.

The Pattern Nobody Wants to Name

No one will point fingers on the record, but the shape of it is unmistakable. An AI infra company gives compute credits to a model lab. The model lab gives API credits back. Vector databases, agent platforms, data-labeling shops — the same kind of round-trip swaps keep surfacing.

Here’s the accounting problem. Revenue recognition under both GAAP and IFRS requires transactions to be at arm’s length between independent parties. When the same VC sits on both cap tables and the “customers” are each other’s portfolio companies, that test gets very hard to pass.

Why It’s Surfacing Now

Investors are starting to do the math. There’s a widening gap between the ARR numbers AI startups put in board decks and the actual inference traffic the hyperscalers report. One side claims 200% growth. The other side’s GPUs don’t seem to know about it.

The other catalyst: IPO prep. Once SEC auditors get involved, what looked like a “strategic partnership” in a Series C deck starts getting reclassified as a related-party transaction — with all the disclosure baggage that entails. A few late-stage AI companies are reportedly restructuring deals before filing S-1s.

The Dot-Com Echo

If this rhymes with anything, it’s 2000. Online ad startups swapped banner inventory and booked the trades as revenue. PointCast, the early portal players, a chunk of the B2B exchanges — all played some version of this game. It was sold as innovative partnership. It ended as accounting fraud cases and a lost decade for the sector.

Not every credit swap is fraudulent. Legitimate co-marketing and integration deals exist. But the line between real economic exchange and bookkeeping theater is getting blurry, and from the outside it’s nearly impossible to tell which side of it any given deal sits on.

What to Actually Watch

Reading an AI startup’s numbers now means going past the headline ARR. Look at operating cash flow as a percentage of reported revenue. Look at customer concentration. If you can get it, ask what share of revenue is non-cash or settled in credits. The distance between a real business and a clever ledger is shrinking, and it’s where the next year of pain will come from.

AI is a real revolution. But every revolution attracts a thicker layer of financial engineering than the last. Whether what we’re watching is genuine industrial liftoff — or a giant circular trade inflating everyone’s books — earnings season will start to tell us, one footnote at a time.