“which is why you have noted venture capitalists claiming that AI labs have “90%+ inference margins,” a completely unproven statement that AI boosters cling to and repeat often enough that it’s taken as gospel”
Our reading, the sentence that was checked:
Claims that AI labs have 90%+ inference margins are completely unproven.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, The OpenAI Bubble, paragraph 44.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-the-openai-bubble-c85548-a1
We recorded 28 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-07-15. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- AI labs
- 90%+ inference margins
- a completely unproven statement
Our interpretation
- This concerns whether the margin assertion had been substantiated publicly by publication, not whether such margins are economically possible.
- The plain reading of inference margin is inference revenue less attributable serving costs, divided by inference revenue, excluding training and unrelated corporate expenses.
- Allow evidence for the specific lab, product or API business actually covered by the margin assertion; do not require company-wide net profitability.
- Use July 15, 2026 as a snapshot of the public evidence record. A financial reporting quarter would answer a different question.
- Audited accounts would be strong evidence, but credible disclosures or independently substantiated measurements could also demonstrate the figure.
Counts as supported if: A sufficiently comprehensive examination of the relevant public claims and their underlying evidence establishes that the 90%+ assertions remained assumptions or unsupported estimates as of July 15, 2026.
Counts as contradicted if: At least one relevant 90%+ inference-margin assertion had already been demonstrated through disclosed or independently substantiated revenue and serving-cost figures on a matching basis.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The readable sources distinguish estimated serving economics from demonstrated realized margins. None establishes the relevant 90%+ figure. That does not prove Zitron's stronger assertion that it was completely unproven: the underlying evidence for the specific investor claims was not fully accessible. Neither side is settled by the admitted record.
Limitations
- The claim does not define inference margin or identify the exact business coverage.
- Relevant analyst material was excluded because read_page marked it PARTIAL.
- The admitted examples concern other dates and narrower serving configurations.
Evidence (2 items)
Context only: Max Kan,
concurrency from the outside
Why it matters here: This original experiment combines external performance measurements with assumed hardware, utilization and serving configurations. It documents why an estimated margin is not automatically a demonstrated realized margin, but cannot establish the absence of all relevant proof in July.
- Publisher
- Max Kan
- Kind of source
- Independent measurement
- Measures
- Google, Microsoft, Fireworks, Together, Nebius and DeepInfra, Modeled inference gross margins for DeepSeek R1-0528 serving percent, estimate, 2026-02-09 to 2026-02-09
- Describes events of
- Public from
Context only: DeepSeek,
all user requests
Why it matters here: DeepSeek disclosed serving statistics but calculated hypothetical revenue by applying one pricing schedule to traffic that included free services. This is relevant disclosure, not proof of realized 90%+ margins for the labs covered by the claim.
- Publisher
- DeepSeek
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- DeepSeek, Theoretical inference revenue and estimated rental cost for V3 and R1, USD dollars per day, company reported, 2025-02-27 to 2025-02-28
- Describes events of
- Public from
What would settle it
The original 90%+ assertions and their supporting records as available by July 15, 2026, including matching revenue, serving costs, product coverage and calculation methods.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, The OpenAI Bubble
which is why you have noted venture capitalists claiming that AI labs have “90%+ inference margins,” a completely unproven statement that AI boosters cling to and repeat often enough that it’s taken as gospel
- Where
- paragraph 44
- Published
- 2026-07-15
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-the-openai-bubble-c85548-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.