OpenAI spent $34 billion to make $13.07 billion in 2025, losing $21 billion in a single year. Roughly $867 million of that revenue, about 6.6%, came from SoftBank, tied to the Crystal Intelligence initiative and SB OAI Japan, an entity that did not formally exist until November 2025. The company also obscured further losses using accounting labeled 'net losses attributable to noncontrolling members capital,' pointing to connected entities not yet publicly disclosed.
Ed Zitron's argument is not just that the numbers are bad. It is that the entire apparatus defending those numbers is the story. Cal Newport's New York Times piece, cited here, makes the structural point cleanly: AI companies describe their own products as existential dangers beyond their control, a posture no consumer product company could sustain without losing its customers immediately. Zitron builds on this to show how hype and doom operate as a single mechanism, keeping observers focused on a theoretical future model rather than the underperforming, expensive software shipping today.
The full piece is worth reading for its breakdown of how SoftBank revenue was likely recognized before the underlying joint venture legally existed, and for the specific accounting line items Zitron uses to suggest deliberate obfuscation. The question it leaves open is not whether OpenAI is in financial trouble. That part is settled. The question is who inside the industry already knows this, and what they are choosing to say publicly anyway.
[READ ORIGINAL →]