The AI Bubble Is An Information War
We recorded 38 claims from this post. 14 were checked against evidence. 24 can’t be checked. 20 more are held. Read the post at its source.
Claims from this post (38)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: Can’t be checkedSaid
Observation
“Worse still, NVIDIA’s earnings also mentioned $27bn in cloud commitments — literally paying its customers to rent the chips it sells, heavily suggesting that there isn’t the underlying revenue.”
Our reading: NVIDIA's $27bn in cloud commitments, which amount to paying customers to rent its chips, strongly suggest that the underlying demand revenue isn't there.
No horizon
Assessment: Can’t be checkedSaid
Observation
“It also suggests CoreWeave's customers — which include Meta, OpenAI, Microsoft (for OpenAI), Google, and a $6.3bn backstop from NVIDIA for any unsold capacity through 2032 — are paying like absolute crap.”
Our reading: CoreWeave's falling revenue per megawatt suggests that its customers, which include Meta, OpenAI, Microsoft, Google and an NVIDIA backstop for unsold capacity, are paying very poorly.
No horizon
Qualifier: suggests, inferred from revenue per megawatt
Assessment: Can’t be checkedSaid
Observation
“This suggests a fundamental weakness in the business model of renting out GPUs, which brings into question the value of NVIDIA’s $68.13bn in Q4 FY2026 revenue, or indeed, Coreweave’s $66.8bn revenue backlog.”
Our reading: CoreWeave's results suggest a fundamental weakness in the GPU-rental business model, which brings into question the value of NVIDIA's $68.13bn Q4 FY2026 revenue and CoreWeave's $66.8bn revenue backlog.
No horizon
Qualifier: suggests
Assessment: UnresolvedSaid
Observation
“Based on my own reporting, which has been ignored (I guess it’s easier to do that than think about it?) by much of the press, OpenAI made $4.33bn through the end of September, and spent $8.67bn on inference in that period.”
Our reading: Through the end of September 2025, OpenAI made $4.33bn in revenue and spent $8.67bn on inference.
No horizon Evidence as of
Limitation: The independently sourced accounting explanation relies on an unnamed source.
Assessment: UnresolvedSaid
Observation
“software is not as simple as spitting out code, neither is it able to automatically clone the SaaS experience.”
Our reading: Software is not as simple as generating code, and AI coding tools such as Claude Code cannot automatically clone the SaaS experience.
No horizon Evidence as of
Limitation: The result depends on what counts as the SaaS experience and automatic completion.
Assessment: SupportedSaid
Observation
“For some context, OpenAI raised around $42bn in 2025 alone.”
Our reading: OpenAI raised around $42bn in 2025.
No horizon Evidence as of
Limitation: Support depends on the ordinary approximate reading of around; exact equality is not established.
Assessment: UnresolvedSaid
Observation
“Stop Reporting That It Raised $110bn, It Is Factually Incorrect”
Our reading: Reporting that OpenAI raised $110bn is factually incorrect, because Amazon invested only $15bn (with $35bn contingent on AGI or an IPO) and NVIDIA and SoftBank are paying in installments.
No horizon Evidence as of
Limitation: The criticism has a supported factual foundation about financing structure.
Assessment: Can’t be checkedSaid
Rhetoric
“It’s so cool that OpenAI is just looting its non-profit!”
Our reading: He sarcastically condemns OpenAI as looting its non-profit, over a reported plan in which the nonprofit may sell several billion dollars of its shares to investors.
No horizon
Qualifier: nonprofit may sell shares depending on investment demand
Assessment: UnresolvedSaid
Observation
“OpenAI and AI companies are not a ponzi scheme. There’s real revenues, people are paying it money.”
Our reading: OpenAI and AI companies are not a ponzi scheme, because customers pay them real revenue.
No horizon Evidence as of
Limitation: The evidence supports the distinction between operating businesses and businesses with no real customers.
Assessment: UnresolvedSaid
Observation
“we’re at around $37bn — leaving OpenAI with about…$20bn in cash at the end of the year.”
Our reading: By his estimates, OpenAI spent about $37bn in the second half of 2025 and ended the year with about $20bn in cash.
No horizon Evidence as of
Limitation: Correct addition does not establish that the inputs were cash transactions.
Assessment: Can’t be checkedSaid
Observation
“In my mind, it’s far more likely that OpenAI’s losses were in excess of $10bn or even $20bn, especially when you factor in that OpenAI is paying an average of $1.5 million in yearly stock based compensation, per the Wall Street Journal.”
Our reading: OpenAI's 2025 losses were probably over $10bn, possibly over $20bn, not the reported $8bn.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“There’s also another possible answer: I think OpenAI is lying to the media, because it knows the media won’t think too hard about the numbers or compare them.”
Our reading: He offers as another possible answer that he thinks OpenAI is lying to the media because it knows the media won't think hard about or compare its numbers.
No horizon
Qualifier: another possible answer
Assessment: UnresolvedSaid
Observation
“Sure you can say “it’s projections,” but there is a clear attempt to use the media to misinform investors and the general public. For example, OpenAI claimed SoftBank would spend $3bn a year on agents in 2025. That never happened!”
Our reading: There is a clear attempt by OpenAI to use the media to misinform investors and the public; for example, OpenAI claimed SoftBank would spend $3bn a year on agents in 2025, which never happened.
No horizon Evidence as of
Limitation: Rollout timing cannot substitute for payment records.
Assessment: UnresolvedSaid
Observation
“What I’m trying to get at is that OpenAI (and, for that matter, Anthropic) has spent the last two years increasingly obfuscating the truth through leak after leak to the media.”
Our reading: For the past two years, OpenAI and Anthropic have increasingly hidden the truth about their finances through repeated leaks to the media.
No horizon Evidence as of
Limitation: Several original financial reports were inaccessible through read_page.
Assessment: Can’t be checkedSaid
Causal claim
“The numbers do not make any sense when you actually put them together, and the reason that these companies continue to do this is that they’re confident that these outlets will never say a thing, or cover for the discrepancies by saying “these are projections!””
Our reading: OpenAI's and Anthropic's leaked numbers are inconsistent with each other, and they keep leaking them because they expect the media not to challenge the discrepancies.
No horizon
Assessment: UnresolvedSaid
Observation
“But the biggest thing to take away from this is that one of the classic arguments against my work is that “costs will just come down,” but the costs never come down.”
Our reading: Contrary to the argument that AI costs will fall, the AI labs' costs have never come down.
No horizon Evidence as of
Limitation: The result depends materially on what costs means.
Assessment: Can’t be checkedSaid
Observation
“That, and it appears that both of these companies are deliberately obfuscating their real numbers as a means of making themselves look better.”
Our reading: OpenAI and Anthropic appear to deliberately obscure their real numbers to make themselves look better.
No horizon
Assessment: UnresolvedSaid
Observation
“There are members of the media who defend OpenAI saying that “these are annualized figures,” but OpenAI does not state that, because OpenAI loves to lie.”
Our reading: OpenAI's December 2025 revenue graphic was designed to mislead because it does not say its figures are annualized.
No horizon Evidence as of
Limitation: The central visual evidence was not transcribed.
Assessment: Can’t be checkedSaid
Observation
“Chief Executive Dario Amodei has spent the last few years massively overstating what LLMs can do in the pursuit of eternal growth.”
Our reading: For the last few years, Dario Amodei has greatly overstated what LLMs can do in order to drive growth.
No horizon
Assessment: Can’t be checkedSaid
Observation
“The military’s demands were for “all lawful uses,” though I don’t think Anthropic really gives a shit about whether the war in Iran is legal, because if it did it would have shut down the chatbot rather than supported the conflict.”
Our reading: He does not think Anthropic really cares whether the war in Iran is legal, because if it did it would have shut down the chatbot rather than supported the conflict.
No horizon
Qualifier: I don't think
Assessment: Can’t be checkedSaid
Causal claim
“In reality, Claude is likely being used to go through a bunch of images and to answer questions about particular scenarios. There is very little specialized military training data, and I imagine many of the demands for “full access to powerful AI” have come as a result of Amodei and Altman’s bloviating about the “incredible power of AI.””
Our reading: Claude is likely being used to go through images and answer questions about particular scenarios, there is very little specialized military training data, and he imagines many of the demands for 'full access to powerful AI' came from Amodei and Altman's bloviating about AI's power.
No horizon
Scope: many of the demands
Assessment: UnresolvedSaid
Observation
“Nevertheless, Dario Amodei gets fawning press about being a patriot that deeply cares about safety less than a week after Anthropic dropped its safety pledge to not train an AI system unless it could guarantee in advance that its safety measures were accurate.”
Our reading: Amodei gets fawning press as a patriot who deeply cares about safety less than a week after Anthropic dropped its pledge not to train an AI system unless it could guarantee in advance that its safety measures were accurate.
No horizon Evidence as of
Limitation: The cited policy evidence does not establish that all safeguards or every requirement to delay development were removed [e1, e2].
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Let’s stop pretending that Anthropic is some sort of ethical paragon! It’s the same old shit!”
Our reading: People should stop pretending Anthropic is some sort of ethical paragon, because it is the same old shit.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“While I think it’s far more likely that the US government itself shuts the door on Anthropic working with it for the foreseeable future even without the supply chain risk designation”
Our reading: The US government is more likely to stop working with Anthropic for the foreseeable future, even without the designation, than to make contractors cut it off.
No horizon
Qualifier: even without the supply chain risk designation
Assessment: Can’t be checkedSaid
Conditional forecast
“If Hegseth and the administration were to truly want to punish Anthropic, they could put pressure on Amazon, Microsoft and Google to cut off Anthropic, which would cut it off from its entire compute operation”
Our reading: If the administration wanted to punish Anthropic, it could pressure Amazon, Microsoft and Google to cut Anthropic off from all of its compute.
No horizon
Only if: if the administration truly wants to punish Anthropic
Assessment: UnresolvedSaid
Observation
“The Department of Defense had agreed to terms around not using Claude for mass domestic surveillance or fully autonomous killing machines (the former of which it’s not particularly good at and the latter of which it flat out cannot do)”
Our reading: Claude is not very good at mass domestic surveillance and cannot run fully autonomous weapons at all.
No horizon Evidence as of
Limitation: Classified systems may not have publicly available evaluations.
Assessment: Can’t be checkedSaid
Rhetoric
“All of this is to say that Altman definitely, absolutely loves war, and wants OpenAI to make money off of it”
Our reading: Altman definitely, absolutely loves war and wants OpenAI to make money off of it.
No horizon
Assessment: UnresolvedSaid
Observation
“Forgive me for being cynical about something from Sam fucking Altman, but I just don’t trust the guy, and this is an (as of writing this sentence) unsigned contract with bus-sized loopholes.”
Our reading: OpenAI's revised Pentagon contract language was unsigned at the time of writing and contains huge loopholes that allow domestic surveillance.
No horizon Evidence as of
Limitation: Unsigned status rests on a secondary relay of reporting.
Assessment: Can’t be checkedSaid
Observation
“This is ultimately a PR move to make Altman seem more ethical, and position Amodei as a pedant that rejects his patriotism and prioritizes legalese over freedom.”
Our reading: OpenAI's strengthened contract language is a PR move to make Altman look more ethical and to cast Amodei as a pedant.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“In reality, we should loathe both Altman and Amodei for their natural jingoism and continual deception.”
Our reading: People should loathe both Altman and Amodei for their jingoism and constant deception.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Both of them are willing and able to provide technology that will surveil or kill people, and while Amodei may have blushed at something to do with autonomous weapons or domestic surveillance, neither appear to have an issue with the actual harms that their models perpetuate.”
Our reading: Altman and Amodei are both willing and able to provide technology that will surveil or kill people, and although Amodei may have balked at autonomous weapons or domestic surveillance, neither appears to have an issue with the actual harms their models perpetuate.
No horizon
Qualifier: appear to
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Neither of these men deserve a shred of credit or celebration. Both of them were and are ready and willing to monetize war, as long as it sort-of-kind-of follows the law.”
Our reading: Neither Altman nor Amodei deserves a shred of credit or celebration, because both were and are ready and willing to monetize war as long as it sort-of-kind-of follows the law.
No horizon
Qualifier: as long as it sort-of-kind-of follows the law
Assessment: SupportedSaid
Observation
“Both of these companies are enthusiastic parts of America’s war machine.”
Our reading: OpenAI and Anthropic are both enthusiastic participants in America's war machine.
No horizon Evidence as of
Limitation: This assesses conduct and public statements, not private enthusiasm.
Assessment: Can’t be checkedSaid
Observation
“Dario Amodei seems far more interested in creating a world dictated by what the United States of America deems to be legal or just, and providing services to help pursue those goals, as does OpenAI and, I’d argue, basically every AI lab.”
Our reading: Amodei seems far more interested in a world dictated by what the US deems legal or just, and in providing services to pursue those goals, than in benevolent AI, as does OpenAI and, he would argue, basically every AI lab.
No horizon
Qualifier: seems
Assessment: Can’t be checkedSaid
Observation
“Generative AI isn’t intelligent, but it allows people to pretend that it is, especially when the people selling the software — Altman and Amodei — so regularly overstate what it can do.”
Our reading: Generative AI is not intelligent, and Altman and Amodei regularly overstate what it can do.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“None of this would be possible without the consistent repetition of the falsehoods peddled by OpenAI and Anthropic.”
Our reading: Military reliance on chatbots would not be possible without the falsehoods OpenAI and Anthropic keep repeating about AI.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Nothing about the military’s use of Claude makes it more useful or powerful than it was before — they’re probably just loading files into it and asking it long questions about things and going “huh” at the end.”
Our reading: Military use does not make Claude more capable; the military is probably just loading files into it and asking questions.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The vulgar dishonesty of Altman and Amodei puts blood on both of their hands, and it’s the duty of every single member of the media to remind people of this whenever you discuss their software.”
Our reading: Altman and Amodei have blood on their hands because of their dishonesty, and the media should point this out whenever it covers their software.
No horizon
Qualifier: whenever discussing their software
Held from this post (20)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews reached different assessments. Our reading, not assessed: Anthropic is totally fine with the US military using Claude for anything beyond its stated exceptions, even though hallucinations are an inevitable result of using a large language model.
- The two reviews read the claim differently
- The two reviews read the claim differently
- The two reviews read the claim differently
- The two reviews reached different assessments. Our reading, not assessed: Deaths from military use of LLMs will be a direct result of people lying about what LLMs can do.
- The record does not meet the current rules. Our reading, not assessed: He thinks OpenAI will need to burn far more than the projected $230bn to reach its revenue projections through 2030.
- Attribution is uncertain
- The criteria are ambiguous. Our reading, not assessed: Claude Code made only $203m in revenue ($2.5bn ARR), and Anthropic at times spends $8 to $13.50 for every dollar it brings in.
- The record does not meet the current rules. Our reading, not assessed: He thinks it will eventually come out that sources were lying to multiple media outlets about OpenAI's burn rate.
- The record does not meet the current rules. Our reading, not assessed: Revenue per megawatt fell from Q3 to Q4 2025, so CoreWeave's underlying business appears to be weakening as it adds capacity.
- The two reviews reached different assessments. Our reading, not assessed: If OpenAI started from $17.5bn in cash at end-June 2025, added $8.8bn in revenue, $8.3bn and $22.5bn in funding, and had a negative cash burn of $8bn, it would have $49.1bn, so the reported 'about $40 billion in cash' cannot be a rounding of that figure.
- The two reviews reached different assessments. Our reading, not assessed: No hyperscaler discloses its actual revenue from renting out GPUs or its own silicon, which he says is not something companies do when things are going well.
- The record does not meet the current rules. Our reading, not assessed: CoreWeave is a time bomb: its operations are deeply unprofitable and need huge capex, $10bn in 2025 and expected to double in 2026.
- The criteria are ambiguous. Our reading, not assessed: LLMs still cannot produce unique ideas, and outside coding (which needs massive subsidies to be even somewhat useful) they are of doubtful efficacy and untrustworthy.
- The two reviews reached different assessments. Our reading, not assessed: If first-half cost ratios held, OpenAI spent about $23.58bn on training and $28.6bn in total burn in the second half of 2025.
- The two reviews reached different assessments. Our reading, not assessed: If Hegseth enforces his ban on military contractors doing any business with Anthropic, Anthropic's business will collapse.
- The two reviews read the claim differently
- The record does not meet the current rules. Our reading, not assessed: CoreWeave's margin fell from 17% (excluding debt) in FY2024 to negative 1% in FY2025, and he sees this only as evidence against the idea that margins might improve.
- The criteria are ambiguous. Our reading, not assessed: Anthropic's Claude appears to be the only AI model available for classified military operations.
- The two reviews reached different assessments. Our reading, not assessed: Anthropic wants to help the military make more accurate autonomous kill drones.