The article presented leaked OpenAI numbers as evidence of catastrophic losses, centered on a reported $38.5 billion 2025 net loss and massive spending on compute, R&D, and sales. The useful framing that emerged is that the headline number is not the business. Multiple commenters pointed to Financial Times reporting that about $30 billion of the 2025 loss came from a non-cash accounting charge tied to OpenAI’s restructuring from its earlier ownership setup into a public benefit corporation. On that view, the operating loss is closer to $8 billion, still enormous but much less sensational than the headline suggests.
From there, the conversation settled on a narrower and more important question: does OpenAI at least make money on serving tokens today. The leaked figures show $13.07 billion in revenue against $7.5 billion in
cost of revenue, which many took as evidence that
inference itself is not obviously being sold below cost. That was enough for some to argue the core business looks healthier than critics claimed, especially if 2026 revenue really reaches the widely cited $25 billion to $30 billion range. Others pushed back that this is only comforting if you pretend model training and refresh cycles are optional. For frontier labs, R&D is not a side project. It is the product pipeline and a large part of the competitive
moat. If new models must keep arriving every few months, then treating training spend as ignorable overhead flatters the economics.
That led to the main business split. Bulls argued OpenAI now looks like a hypergrowth company with ugly but legible numbers: high
gross margin on usage, brutal but potentially temporary expansion spend, and a path to break-even if revenue keeps compounding. Bears argued AI is not normal software because the marginal cost of serving output is real, the shelf life of top models is short, and model quality is converging across OpenAI, Anthropic, Google, DeepSeek,
Qwen, and others. In that world, scale helps but does not create the kind of lock-in that made prior software monopolies so profitable. Price pressure, cheaper
open-weight models, and enterprises routing work from premium planners to cheaper executors could compress margins before these labs ever earn back their training spend.
The mood was skeptical of the article’s framing but not especially charitable to OpenAI. People largely agreed the leaked figures do not prove imminent collapse. They also do not justify easy confidence in a trillion-dollar story. The live issue is not whether OpenAI can sell tokens profitably this quarter. It is whether any frontier lab can keep funding constant retraining, customer acquisition, and infrastructure buildout while model differentiation erodes faster than costs do.