“It’s very clear OpenAI moved around numbers to make things look better than they are, and I believe that inference costs are being dumped in sales and marketing.”
Our reading:
It is very clear that OpenAI moved numbers around to make things look better than they are, and he believes its inference costs are being booked under sales and marketing.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
Techstrong.ai (available 2026-06-20) reported on leaked audited 2025 statements that the Financial Times verified (e1). For 2025 they show $13.07B revenue, $7.5B cost of revenue and $5.73B sales and marketing. The article says cost of revenue reflects computing costs incurred at inference time. It does not say whether all inference costs sit in cost of revenue. It also does not say whether any costs were moved between line items. Epoch AI (available 2026-03-06) suggested that OpenAI's first-half 2025 S&M spending includes inference for free users (e2). Epoch presents this as its own explanation of a gap in reported figures, not as a disclosure. No evidence item provides the statements' notes on cost classification, a restatement, or an OpenAI statement on where inference for paid and free users is booked. No item explains the gap between reported inference spend and cost of revenue. The evidence neither shows nor rules out that numbers were moved to make results look better. The speaker hedged the sales-and-marketing part ('I believe'), so it is not scored separately.
Caveats
- The evidence includes no public filing or financial-statement notes on how OpenAI classifies inference compute for paid and free users.
- Techstrong.ai's account is a summary of leaked documents, not the statements themselves.
- Epoch AI's view is an analyst inference, not a company disclosure.
- Whether booking free-user inference as S&M would count as 'moving numbers' is a judgment the evidence cannot settle.
- The $8.67 billion nine-month inference figure comes from the speaker's own reporting and is not used here as evidence.
Evidence (2 items)
Context only: Techstrong.ai,
Cost of revenue rose to $7.5 billion from $2.65 billion, reflecting substantial infrastructure and computing costs incurred at inference time as user prompts scaled. Sales and marketing, meanwhile, expanded fivefold, from $1.11 billion to $5.73 billion
Why it matters: Confirms the 2025 line items the speaker discusses. It says nothing about how costs were classified or whether anything was moved.
- Where
- paragraph 5
- Publisher
- Techstrong.ai
- Kind of source
- News report, audited
- Measures
- OpenAI, cost of revenue; sales and marketing expense, USD billions, company reported, 2025-01-01 to 2025-12-31
- About events on
- Published
Context only: Epoch AI,
Our explanation of the discrepancy is that the 2025 H1 S&M spending includes the cost of inference for free users, OpenAI’s primary way of advertising their products, while the 2024 figure does not.
Why it matters: Independent analysts also suspect free-user inference sits in S&M. That is consistent with the speaker's hedged guess. But it is their inference, not a disclosure, and it does not show numbers were moved to mislead.
- Publisher
- Epoch AI
- Kind of source
- Secondary summary
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Conditions
- I believe (inference-cost allegation) (their words)
- very clear (moving numbers) (their words)
Where we had to interpret
- The firm claim is that OpenAI presented its 2025 costs so that results looked better than the underlying economics. For example, it put some costs outside cost of revenue, flattering gross margin, or shifted figures between line items.
- The 'I believe' clause about inference sitting in sales and marketing is the speaker's own hedged guess. It is not required for support. Evidence that inference costs were booked in S&M or another non-cost-of-revenue line would count toward the firm claim. So would any other documented reclassification.
- Target period: OpenAI's 2025 financial statements, which are the numbers he discusses.
Right if: A credible source documents that OpenAI classified or shifted costs, such as inference compute, outside cost of revenue in a way that makes its 2025 margins or losses look better. Sources could be a public filing, auditor notes, a restatement, an OpenAI statement, or reputable reporting on its accounting.
Wrong if: A credible source shows OpenAI's 2025 inference compute was fully in cost of revenue under consistent policies. Or a source explains the gap between the roughly $8.67B nine-month Azure inference figure and the $7.5B full-year cost of revenue without any flattering reclassification.
We wrote this test after the outcome was known.
What would settle it
A public S-1 or audited statements with notes showing where OpenAI books inference compute for paid and free users. An OpenAI statement or reputable reporting that documents a reclassification, or explains the gap between its inference spend and cost of revenue, would also settle it.
We found 24 claims in this post. See them all.
Full record
Source and dates
Where's Your Ed At, Cargo Culture, paragraph 95.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2026-06-23.
Where it was said (1)
Said once in the posts we read.
: Where's Your Ed At, Cargo Culture
It’s very clear OpenAI moved around numbers to make things look better than they are, and I believe that inference costs are being dumped in sales and marketing.
- Where
- paragraph 95
- Published
- 2026-06-23
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
A reviewer said our test wasn’t fair to what was actually said.