Exclusive: OpenAI Losses Increased Nearly 8X in 2025, With Spending Hitting $34 Billion

Source: Where’s Your Ed At Date: 2026-06-16 URL: https://www.wheresyoured.at/exclusive-openai-financials/

Summary

Ed Zitron’s exclusive report on OpenAI’s 2025 financials: revenue grew to $13.07 billion but net losses hit $38.53 billion — nearly 8× the prior year’s $5.09 billion loss — against $34 billion in total spending. R&D alone consumed $19.18 billion, with $17.2 billion flowing to Microsoft across expense categories. The conversion from nonprofit to for-profit contributed fair-value adjustments to the headline loss figure, but the underlying cash burn is structurally steep: the company is spending roughly $2.60 for every dollar of revenue.

Implications

This feeds directly into the AI economics/subsidies thread and has second-order pressure on every other thread:

• Subsidy dependency is load-bearing. At current burn, OpenAI’s product pricing is not remotely sustainable on revenue alone — it is a subsidy funded by investor capital, Azure compute credits, and Microsoft’s balance sheet. Every dollar of below-cost API access competitors price against is backed by that same subsidy. The floor can shift abruptly if investor sentiment or Microsoft’s strategic calculus changes.
• The “closed capability” clock. The $19 billion R&D line is what maintains OpenAI’s frontier position. If that spending has to compress to approach sustainability, the gap between the closed-lab capability curve and open-weight models shrinks faster than the lab would choose. The economics create structural pressure toward the open clock even when the lab prefers the closed one.
• Host-ownership war sub-text. $10.59 billion of R&D flowing to Microsoft is model training on Azure. That is OpenAI’s model infrastructure being owned by the host. The platform-lock dynamic runs both ways: OpenAI depends on Azure, Azure depends on OpenAI’s frontier position for its AI services revenue. Watch what a renegotiation of that relationship does to both companies’ reported numbers.