The two companies building the infrastructure for the agent economy can't agree on what a dollar looks like.
The Summary
- OpenAI and Anthropic report revenue using different methodologies, making direct comparisons between the AI leaders impossible for investors trying to value the private companies
- The accounting discrepancy matters more now because Bank of America sees strong institutional and retail demand for potential OpenAI and Anthropic IPOs, with public markets ready to absorb massive tech listings
- Investors measuring the AI race are flying blind on the most basic metric: how much money these companies actually make
The Signal
The revenue calculation gap between OpenAI and Anthropic reveals a deeper problem in the foundation model economy. These are not widget companies. They sell API calls, enterprise licenses, consumer subscriptions, and credits that may or may not get used. The definition of "revenue" gets messy when your product is inference tokens, not physical goods.
One company might count committed enterprise contracts as revenue the day they're signed. The other might recognize revenue only when tokens are actually consumed. The difference could be hundreds of millions of dollars in reported performance for the same underlying business activity.
"Public markets have the capacity to absorb massive potential IPOs from companies including OpenAI and Anthropic."
Bank of America's analysis points to SpaceX's listing as proof that institutional and retail investors have appetite for transformational technology companies, even at eye-watering valuations. But SpaceX builds rockets. You can count rockets. You can measure payload capacity and launch cadence. The metrics are hard.
AI foundation models operate in a different universe. How do you value a company when you can't even agree on what revenue means? Abundant private capital has let OpenAI and Anthropic delay this reckoning. They've raised billions without the scrutiny of public market accounting standards. But that grace period is ending.
Key valuation challenges:
- No standardized revenue recognition for API-based AI services
- Enterprise contracts with variable usage make forecasting unreliable
- Consumer subscription revenue mixed with volatile API call volume
The IPO market has changed. Companies now stay private longer, accumulating massive valuations before facing public scrutiny. When they finally list, investors get less time to understand the business before the market sets a price. If the two leading foundation model companies can't even measure revenue the same way, what chance do public market investors have?
The Implication
This is not just an accounting curiosity. It's a warning that the agent economy is being built on financial infrastructure that doesn't exist yet. Before foundation model companies go public, the industry needs standard revenue recognition practices. Otherwise, investors will be comparing numbers that mean different things, pricing companies based on metrics that aren't comparable.
Watch for pressure from late-stage investors and bankers to standardize these metrics before any IPO. The first foundation model company to go public will set the template. If they choose aggressive revenue recognition, everyone else will follow. If they choose conservative, the entire sector's valuations could compress overnight. The accounting rules matter more than the technology right now.