“AI startups and enterprise customers pay Anthropic and OpenAI on a per-million token rate (losing money in the process) so that Anthropic and OpenAI can rent GPUs from hyperscalers (losing tens of billions of dollars a year) so that hyperscalers can buy a trillion or more dollars’ worth of GPUs (putting them in such a hole that they’ll never, ever be able to make the money back).”
Our reading:
Anthropic and OpenAI lose tens of billions a year, and hyperscalers will never recoup their trillion-plus GPU spending.
Assessment: Unresolved
What happened
Did the forecast come true?
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
The annual-loss figures reach tens of billions on a reported net-loss basis: Anthropic disclosed $42.0 billion for 2025, and Ars Technica reported nearly $39 billion for OpenAI [e1, e2]. Anthropic's total includes $34.3 billion in fair-value changes, while OpenAI's includes a large valuation-related charge [e1, e2]. Ars Technica separately reports OpenAI's 2025 operating loss as $20.92 billion, illustrating why the loss measure matters [e2]. These records do not establish lifetime recovery of hyperscaler GPU spending [e1, e2]. S&P Global's August 27, 2026 outlook forecasts improving monetization but documents neither verified recovery nor a closed investment lifecycle for a defined trillion-plus GPU investment [e3]. The forecast that hyperscalers will never recover that spending therefore remains unresolved [e1, e2, e3].
Caveats
- The reported net and operating losses do not settle an unspecified loss measure or isolate losses attributable to GPU rentals [e1, e2].
- OpenAI's figures are secondary reporting on financial statements [e2].
- The infrastructure outlook does not establish a defined trillion-plus GPU purchase cohort, its attributable cumulative returns or final recoveries [e3].
Evidence (3 items)
Context only: Anthropic,
net losses
Why it matters: The surrounding disclosure reports a USD42.0 billion 2025 net loss, including USD34.3 billion of fair-value changes. This establishes a substantial accounting loss but does not identify GPU rental losses or settle the speaker's unspecified loss measure.
- Publisher
- Anthropic
- Kind of source
- Regulatory filing, audited
- Measures
- Anthropic, Annual net loss, USD billions, gaap, 2025-01-01 to 2025-12-31
- About events on
- Published
Context only: Ars Technica,
nearly $39 billion
Why it matters: The report distinguishes OpenAI's net loss from operating losses and describes a large accounting charge. It is secondary reporting on audited company statements, not evidence that the entire loss arose from GPU rental.
- Publisher
- Ars Technica
- Kind of source
- News report, audited
- Measures
- OpenAI, Annual net loss reported from company financial statements, USD billions, company reported, 2025-01-01 to 2025-12-31
- About events on
- Published
Context only: S&P Global Ratings,
monetization improves
Why it matters: This competing outlook anticipates improving monetization. It is not realized investment recovery, and the report's broad capital-expenditure measure cannot establish trillion-plus GPU spending or its lifetime return.
- Publisher
- S&P Global Ratings
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Forecast
- Deadline
- None given. We don’t add one.
Where we had to interpret
- This is a compound claim: a present annual-loss observation and an indefinite forecast about recovery of hyperscaler GPU investment. The forecast determines the overall horizon.
- For the loss clause, I use calendar 2025, the latest completed annual period before publication. I do not annualize a later half-year figure.
- I read tens of billions as at least USD20 billion annually for the two labs collectively. The wording could also be read as applying separately to each.
- The source does not specify net loss, operating loss or cash burn. Those measures must not be substituted for one another.
- Making the money back means cumulative attributable returns after operating costs and recoveries covering the relevant GPU purchase cost. Company-wide revenue alone would not demonstrate recovery.
- The contract metric_scope describes the forecast's GPU investment. The separate loss clause concerns Anthropic and OpenAI, USD billions, realized annual flows for January–December 2025, with an unspecified accounting basis.
- No deadline, fixed GPU investment cohort or exhaustive list of hyperscalers is supplied.
Right if: Comparable annual figures establish the lab-loss assertion, and a closed investment lifecycle establishes that the relevant trillion-plus GPU spending was not recovered and can generate no further recoveries. Present losses or slow returns alone cannot establish never.
Wrong if: Comparable annual figures refute the loss assertion, or verified cumulative attributable net returns and recoveries cover the relevant GPU investment. Because the financial basis is unspecified, a supported or contradicted score requires resolving that basis first.
We wrote this test after the outcome was known.
What would settle it
Establish comparable calendar-2025 losses for both labs on an explicit accounting basis. Define the hyperscalers and GPU investment cohort. Verified cumulative attributable returns after operating costs, plus recoveries, covering the purchase cost would contradict 'never.' Support would require final retirement or disposal and closure of all economic interests, with total returns below cost and no possible further recoveries.
We found 37 claims in this post. See them all.
Full record
Source and dates
Where's Your Ed At, The More You Buy, The More You Lose, paragraph 89.
- 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 2026-07-28.
Where it was said (1)
Said once in the posts we read.
: Where's Your Ed At, The More You Buy, The More You Lose
AI startups and enterprise customers pay Anthropic and OpenAI on a per-million token rate (losing money in the process) so that Anthropic and OpenAI can rent GPUs from hyperscalers (losing tens of billions of dollars a year) so that hyperscalers can buy a trillion or more dollars’ worth of GPUs (putting them in such a hole that they’ll never, ever be able to make the money back).
- Where
- paragraph 89
- Published
- 2026-07-28
- 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
The final check failed: our summary didn’t match their words.