Source: MHT snapshot captured July 22, 2026, of https://www.wheresyoured.at/the-openai-bubble/ (~16,000 words). Published this week; it synthesizes six months of the author's prior work.
Method: Full text extracted from the MHT archive. Every distinct factual claim was inventoried, then checked against primary and news sources across two passes (15 searches in round one, 8 in round two targeting the items left open). Pre-2026 claims were checked against the contemporaneous record. Verdicts:
VERIFIED — independently confirmed by primary sources or multiple reputable outlets
VERIFIED W/ CAVEAT — substantively correct, but with an imprecision worth knowing
AUTHOR'S CALC — Zitron's own estimate or analysis; not independently checkable
ATTRIBUTED — sourced to a named outlet or analyst; not independently re-verified
UNVERIFIED — could not confirm (not the same as false)
ERROR — contains a factual mistake
OPINION — characterization, prediction, or interpretation; not fact-checkable
A. Core OpenAI financials
$122B funding round at closing, SoftBank $30B in tranches, third tranche due October 2026. VERIFIED. The round closed March 31, 2026 at an $852B post-money valuation with Amazon ($50B, partly contingent), Nvidia ($30B), and SoftBank ($30B, paid in quarterly tranches through 2026 backed by bridge loans). Zitron's "$20B of $30B received, third tranche due Oct 1" matches the reported April/July/October tranche schedule.
Cited by: SoftBank → OpenAI; Amazon → OpenAI
"Received at least $50 billion" of the round. PLAUSIBLE, UNVERIFIED. Tranche arithmetic by late July supports it (Amazon's $15B upfront, two of three SoftBank tranches, Nvidia's $30B if funded), but The Elec reported only ~$25B was immediately available at the March close, and no source states a received-to-date total. The second pass found nothing more definitive.
900 million weekly active users, ~5% paying. VERIFIED. OpenAI itself disclosed 900M+ weekly users and 50M+ subscribers as of March 2026, which is ~5.6% paying.
80% of $20/month subscribers projected to churn in 2026. VERIFIED W/ CAVEAT. The Information (April 28, 2026) reported OpenAI's own projection that Plus subscriptions fall 80%, from 44M in 2025 to 9M in 2026. The caveat the article omits: OpenAI projects the decline as migration to its cheaper ad-supported Go tier (3M to a projected 112M subscribers), not a collapse in total paying users. The 80% is real; the framing strips out the offsetting half of the projection.
OpenAI ad projections ($2.4B in 2026, $102B in 2030) vs eMarketer (under $1B industry-wide this year, $5.41B by 2030). VERIFIED W/ CAVEAT. OpenAI's own reported projection is $2.5B/2026 and $100B/2030 (per Axios/Reuters/AdWeek); the exact $2.4B/$102B figures match a Barclays model from April 2026. The eMarketer figures are real but are US-only, standalone-chatbot figures; the article presents them as the entire global chatbot ad market. Direction of the argument holds, scope is overstated.
Revenue projections of $113B (2028), $184B (2029), $284B (2030), profitable in 2030, per The Information. VERIFIED IN SUBSTANCE. Bloomberg reports OpenAI now projects revenue exceeding $280B in 2030, matching his 2030 figure within rounding, and the documented projection series has escalated in exactly this pattern ($125B/2029 in mid-2025 reporting, ~$200B/2030 by November 2025, $280B+ by early 2026), with cash-flow positivity projected for 2029 or 2030. The intermediate-year figures weren't matched exactly but sit on the reported curve.
IPO delayed to 2027 after failing to support a $1T listing. VERIFIED (delay) / UNVERIFIED (reason). Multiple current sources state OpenAI's IPO has slipped to 2027; earlier 2026 coverage expected a late-2026 listing. The valuation-driven explanation is Zitron's read.
$852B cash burn through 2030; $748B of Microsoft/Amazon/Oracle RPOs; 50%+ of global AI compute; ~$70B more RPOs across Cerebras, CoreWeave, Nebius, IREN, Lambda, Nscale (per Kakashii). AUTHOR'S CALC / ATTRIBUTED. These are Zitron's own reconstructions from filings plus a cited independent analyst. The public anchor (OpenAI's ~$1.4T in announced compute commitments; a reported ~$600B planned spend by 2030) is real; the precise RPO attribution is his estimate. S&P independently estimates OpenAI is ~half of Oracle's $638B RPO alone, which is directionally supportive.
Cited by: CoreWeave → OpenAI
Microsoft has said it spent $100B+ on the OpenAI relationship. VERIFIED W/ CAVEAT. Microsoft executive Michael Wetter testified in May 2026 (Musk v. OpenAI trial) that cumulative spend, including infrastructure and hosting, exceeds $100B through the fiscal year ending June 2026. Zitron's "as of early 2026" slightly predates the actual disclosure.
Cited by: Microsoft → OpenAI
NVIDIA's latest quarterly report cites one AI research and deployment company driving meaningful revenue via its customers. UNVERIFIED. Consistent with NVIDIA's known 10-Q concentration language; the specific Q1 FY2027 wording wasn't pulled.
B. The four "phantom deal" narratives
NVIDIA, Sept 22, 2025: up-to-$100B / 10GW letter of intent; stock jumped; CNBC reported a $10B first tranche closing within a month. VERIFIED (pre-cutoff record). It was an LOI, not a definitive agreement, and NVDA rose about 4% that day (the $175.30 to $181 figures are consistent).
Cited by: NVIDIA → OpenAI
WSJ reported the deal "on ice" ~three months later; NVIDIA then pledged $30B. VERIFIED. The WSJ stall report came in December 2025; NVIDIA's $30B arrived as part of the Feb/March 2026 round, matching Zitron's two-month gap. The $18.6B private-company disclosure is ATTRIBUTED.
Cited by: NVIDIA → OpenAI
AMD, October 2025: 6GW deal, warrants up to ~10% of AMD, MI450, ~34% share surge; then Meta signed a near-identical deal in February 2026. VERIFIED W/ CAVEAT. The AMD-OpenAI announcement was October 6, not October 5; the 34% figure matches the intraday peak (close was lower). The Meta-AMD deal (Feb 24, 2026) is confirmed and is structurally the same: 6GW, custom MI450, first GW in 2H 2026, 160M-share warrant at $0.01. "No tangible evidence OpenAI has bought an AMD GPU" is his claim from AMD's disclosures; AMD's 10-K does reference an OpenAI purchase agreement, and MI450 deployment was always scheduled for 2H 2026, so absence of deliveries to date is expected either way.
Cited by: OpenAI → AMD
Broadcom, Oct 13, 2025: 10GW deal, ~9% stock pop, "working together 18 months"; The Information (May 7, 2026) on unresolved chip financing; June 24, 2026 "Jalapeno" unveiled with a nine-months-from-design-to-production claim. VERIFIED (announcements and dates) / ATTRIBUTED (May financing story). The nine-month claim is genuinely OpenAI/Broadcom's own language; calling it "a blatant lie" is Zitron's judgment. His implicit tension is real: CNBC noted 18 months of prior joint work at the October announcement.
Cited by: OpenAI → Broadcom
Disney, Dec 11, 2025: $1B investment plus character licensing for Sora, Iger and Altman on CNBC; March 24, 2026: Sora discontinued, deal dead, unclear anything happened. VERIFIED. Bloomberg, Variety, and Deadline confirm the shutdown and wind-down; Deadline and gHacks report no money changed hands before dissolution. One of the strongest-supported sequences in the piece.
Cited by: Disney → OpenAI
All four stocks rallied on announcements that were LOIs/MOUs or later unwound. VERIFIED in substance for NVIDIA, Samsung/SK Hynix, and Disney; the AMD and Broadcom deals remain live (Jalapeno exists; Meta's parallel deal arguably validates rather than indicts AMD's structure). "Fictional" is his characterization.
C. Memory crisis
Oct 1, 2025 OpenAI-Samsung-SK Hynix LOI for 900K wafers/month (~40% of DRAM supply); stocks rallied; Samsung hiked prices up to 60% about a month later. VERIFIED. The pre-cutoff record confirms the LOI and rallies; Reuters and CNBC-sourced reporting confirms Samsung raised key memory prices by as much as 60% from September (e.g. 32GB DDR5 modules from $149 to $239).
Cited by: Samsung → OpenAI; SK Hynix → OpenAI
March 2026 Samsung deal to supply up to 800M Gb of HBM4 to OpenAI (Reuters) for Jalapeno. ATTRIBUTED. Not re-verified; consistent with Jalapeno's timeline.
June 2026 antitrust class action accusing Samsung/SK Hynix/Micron of restricting DRAM supply via the HBM shift starting 2022. VERIFIED. Garciaguirre v. Samsung, filed June 25, 2026, N.D. Cal.; 17 plaintiffs; ~90% market share; ~700% four-year price-rise allegation. The quoted "no economic sense" theory matches the complaint as reported.
Cited by: Samsung → RAM prices
2000s DOJ price-fixing fines, Micron amnesty; Twombly (2007) raising the pleading bar; 2019 class action dismissed under it. VERIFIED W/ CAVEAT. Fines totaled ~$731M with Samsung at $300M and Hynix $185M, imposed 2004-2006 for 1999-2002 conduct; the article's "fined over $100 million back in 2002" compresses the timeline and understates Samsung's fine. Micron's cooperation amnesty, Twombly, and the dismissed case are accurate.
SK Hynix leadership: shortage until ~2030, 2027 the worst supply year. VERIFIED. Current reporting matches both statements.
2026 supply sold out by the second week of January; only 60% of medium-term orders serviceable; BofA on SK Hynix capacity. SOLD-OUT CLAIM VERIFIED. SK Hynix disclosed its entire 2026 DRAM/HBM/NAND capacity sold out in October 2025, Samsung secured orders for all 2026 production, and Micron's Sumit Sadana told CNBC on January 10, 2026 (the second week of January) that Micron was sold out for 2026, completing the trio exactly on Zitron's timeline. The 60%-of-medium-term-orders detail and the BofA capacity report remain ATTRIBUTED.
Consumer price fallout: Steam Machine above target, Apple Mac/iPad hikes, Xbox/PlayStation/Nintendo increases. VERIFIED in substance. Valve abandoned a sub-$1,000 target and launched at $1,049; Apple and Microsoft price hikes are documented. The specific "30% higher than planned" for Valve depends on an internal target figure that isn't public.
Cited by: Samsung → RAM prices
HBM 681M Gb (2022) to 29.3B Gb (2026); NVIDIA ~65% of HBM purchasing. ATTRIBUTED (ComputerBase.de chart; analyst estimates).
Cited by: NVIDIA → SK Hynix
D. Oracle and the Ellisons
S&P downgrade citing OpenAI as roughly half of $638B RPO, with the leases-stranded scenario. VERIFIED, with one ERROR: the article says S&P downgraded Oracle "to BBB/A-2." S&P actually downgraded FROM BBB/A-2 TO BBB-/A-3 on July 9, 2026. His description (last stop before junk) fits BBB-, so the argument survives; the stated rating is wrong.
Cited by: OpenAI → Oracle; OpenAI → Oracle
FY2027 capex $90B+. VERIFIED. S&P models $90-95B, up from a prior $60B assumption. FY2026 capex of $55.7B and $50B+ raised via stock and debt: UNVERIFIED as exact figures, though corroborated in range (current coverage cites ~$70B in AI spend this year and ~$167B total debt).
$300B five-year OpenAI contract; $340B buildout for 7.1GW. VERIFIED (contract, widely reported since Sept 2025; OpenAI's own materials describe a $300B+, ~4.5GW/year partnership) / AUTHOR'S CALC (the $340B/7.1GW figures). The contract-start framing (Oracle FY2027, from June 1, 2026) is consistent with reporting that revenue ramps in FY2028.
Cited by: OpenAI → Oracle
Fallen-angel mechanics via the Ford precedent (March 2020, $35B+ index ejection, 8.5-9.625% coupons vs 5.2% average in 2019). VERIFIED (historical record).
21,000 layoffs. VERIFIED. Oracle's 10-K (filed June 22, 2026) shows headcount falling from 162,000 to 141,000 in a year, about 13%, with $1.8B in restructuring costs; Bloomberg and others tie the cuts to funding the AI buildout, matching his framing.
Ellison: ~40% ownership, 1.16B shares, ~$2.3B annual dividend, 346M shares pledged, ~$10B cash and ~$15B Tesla, never sold more than ~$4.7B. VERIFIED. The September 2025 proxy disclosure (346M pledged shares, worth $100B+ then, ~$50B now) and the Forbes liquidity analysis match almost line for line. The "$61.5B" pledge value reflects an interim share price; all three figures ($100B, $61.5B, $50B) are the same block at different dates.
Cited by: Ellison → Oracle; Ellison → Oracle
Paramount-WBD: ~$111B deal, $45.7B Ellison Trust equity commitment, Larry's personal guarantee, ~$650M/quarter if delayed past Sept 30, Middle East co-funding. VERIFIED for the deal size, the $45.7B trust commitment, the $40.4B personal guarantee, and the ticking fee ($0.25/share/quarter across ~2.5B shares is ~$620M, close to his figure). The $24B Middle East component: ATTRIBUTED. Fresh context the article doesn't have: a judge issued a 14-day restraining order on the merger on July 21, and a shareholder suit against the Ellisons was filed July 16.
Cited by: Ellison → WBD merger; MGX → WBD merger
The margin-call cascade below $70 and $50. OPINION / AUTHOR'S CALC. Built on verified inputs, but the thresholds and sequence are his model.
E. SoftBank
$46B yearly Vision Fund gain thanks to OpenAI, against Coupang/DiDi/Klarna losses. VERIFIED. FY ended March 2026: $46B segment gain, ~$45B of it OpenAI, with exactly those three names dragging. One of the most precisely accurate claims in the piece.
Cited by: OpenAI → SoftBank
Son history: Alibaba, Sprint/T-Mobile, ARM ($32B in 2016, $54.5B IPO 2023, ~$300B now), WeWork 28 minutes, Greensill, Altman declining $10B in 2019. VERIFIED (widely documented record; the 2019 episode is reported lore, ATTRIBUTED).
Cited by: Son → SoftBank
SoftBank as Japan's third-largest company by market cap behind KIOXIA and Mitsubishi UFJ. VERIFIED IN SUBSTANCE. The extraordinary part is true: the AI trade really has remade Japan's market-cap table. Kioxia (up ~640% in 2026 on the memory boom) and SoftBank both passed Toyota around June 1-3, 2026, and a mid-June ranking shows Kioxia first (~$323B), SoftBank second, MUFG third. Zitron's exact ordering (SoftBank third, behind MUFG) differs from the June snapshot and may reflect July price moves; the substantive claim, that a memory maker and SoftBank now top a table Toyota led for decades, is confirmed.
Cited by: SoftBank → Toyota; KIOXIA → Toyota
F. Market structure and miscellany
NVIDIA's $2B CoreWeave investment, January 2026, positively covered. VERIFIED. Announced Jan 26, 2026 at $87.20/share; stock rose ~6%; Bloomberg itself framed it within circular-financing concerns. Reading it as a liquidity warning sign is his interpretation.
Cited by: NVIDIA → CoreWeave
Anthropic's "$35 billion debt deal with Broadcom." VERIFIED W/ CAVEAT. The June 2026 deal is real but is an Apollo/Blackstone-led $35B private-credit SPV that buys Google TPUs for Anthropic to lease; Broadcom provides guarantees on the senior tranches (and took a ratings knock for it). Calling it a Broadcom debt deal compresses the structure.
Cited by: Apollo/BX → Anthropic; Broadcom → Anthropic
Apple v. OpenAI. VERIFIED. Filed ~July 10, 2026, N.D. Cal.; alleges systematic trade-secret theft via poached employees (400+ hires cited in coverage), names Chang Liu, Tang Tan, and io Products; includes allegations about extracting manufacturing techniques from Apple's supply chain. "Potentially criminal charges" and "destroy the company" are OPINION.
Cited by: Apple → OpenAI
SpaceX on public markets; Cursor acquisition; Goldman's 100x AI revenue projection. VERIFIED. SpaceX listed on Nasdaq June 12, 2026 (~$1.75T valuation, ~$75B raised, largest IPO ever, having absorbed xAI); the $60B all-stock Cursor acquisition was confirmed June 16 (closing expected Q3, hence his "assuming it actually happens"); the FT reported Goldman, as lead underwriter, projecting AI revenue from $3.2B (2025) to $322B (2030), which is the 100x. SK Hynix's US listing is corroborated by its SEC F-1 filings.
Retail investors at 20% of US equity volume; Citadel Securities' July note on retail as the strongest structural buyer; Korean margin buying. VERIFIED. Citadel Securities' Scott Rubner wrote in mid-July 2026 that retail remains the strongest structural buyer of US equities with no net-sell day all month (the second-strongest retail month since January 2020); Citadel's own published estimate puts retail at 20% of shares executed, steadily rising. Bloomberg coverage notes retail dominance extending to Korea and Taiwan. The article's language tracks the Rubner note almost verbatim.
Cited by: JP retail → SoftBank; US retail → OpenAI
Bank of England governor Bailey's AI-bubble warning. VERIFIED IN SUBSTANCE (Bailey has publicly warned on AI-valuation correction risk since October 2025); the specific UK monetary-policy framing wasn't re-verified.
Goldman's Jeffrey Papai note; $244B raised in 2026 including NVIDIA and SpaceX; hyperscalers unable to fund capex from cash flow as of Q3 2026. SUBSTANCE VERIFIED. WSJ-sourced reporting confirms six AI hyperscalers (Alphabet, Amazon, Meta, Oracle, Nvidia, SpaceX) issued ~$244B in bonds in H1 2026, versus $108B in all of 2025 and $17B in 2024; the "if you include NVIDIA and SpaceX" qualifier matches the WSJ six-company definition exactly. Goldman credit strategy (Amanda Lyman) says capex forecasts already consume most hyperscaler operating cash flow, and a May Goldman analysis projected 2026 capex consuming 100% of it. The specific analyst name Papai did not surface; the numbers he hangs on the note check out.
Salesforce: three years to $1B AI ARR against $40B+ revenue. VERIFIED IN SUBSTANCE (matches company disclosures through 2025).
The $14,000-of-tokens-on-a-$200-plan anecdote; the 90%+ inference margin VC claim; AWS 2003-2015 capex of $29.7B inflation-adjusted; $178.5B in 2025 US data-center debt deals. ATTRIBUTED / AUTHOR'S CALC. Real discourse artifacts; figures not independently confirmed.
Stargate site status (Abilene 824MW critical IT, Wisconsin/Michigan/New Mexico delays, Oracle guaranteeing $14B of Michigan bonds). LARGELY VERIFIED. The Michigan (Saline Township) financing closed at $16.3B total with a ~$14B bond tranche, and it struggled: PIMCO had to anchor ~$10B after US banks retreated, with the package resting on Oracle's credit support; the $14B figure and the closed-only-with-difficulty framing both hold. Construction there began around May/June 2026, matching "first steel beams two months ago." Abilene's 824MW critical-IT figure is consistent with the ~1.2GW facility (roughly a 70% IT ratio) and with Epoch AI's estimate of only ~0.3GW operational in April, supporting the delay thesis; note the Abilene expansion was cancelled in March 2026. Wisconsin and New Mexico stage details remain ATTRIBUTED to his sources.
Cited by: OpenAI → Stargate; Apollo/BX → Stargate
G. Historical scaffolding (2019-2023)
Microsoft's $1B (2019) and ~$10B (Jan 2023, heavily Azure credits) investments; the May 2020 supercomputer (285K CPU cores, 10K GPUs); ChatGPT's Nov 2022 launch into a tech downturn; 100M weekly users by late 2023 with ~$108M/month revenue (The Information's ~$1.3B annualized); Google and Amazon committing ~$6B total to Anthropic in 2023; NVIDIA-Mellanox announced March 2019. All VERIFIED against the contemporaneous record.
Two dating slips: the A100 was announced and shipped from May 2020, not "by March 2020" (Mellanox also closed April 2020, so the four-months-later Microsoft sequencing is off); and "500,000 A100s by end of 2023" compresses Omdia's estimate, which was ~500K H100-class units in a single quarter of 2023 and mixed A100/H100 counts. ERROR (minor, non-load-bearing).
H. Not fact-checkable
The thesis passage quoted in the LinkedIn post (the bubble as "cult-like psychosis" producing the greatest capital misallocation in history), the psychoanalysis of Altman, Son, and Ellison, the collapse mechanics and timing ("the Hellmouth opens" in 2027), the media-failure argument, and every "will" statement about OpenAI dying, Oracle below $70/$50, and post-collapse contagion are OPINION and PREDICTION. They are built on the factual scaffolding above but do not follow from it deductively. Note also the article's own hedges: the memory-collusion section explicitly disclaims accusing anyone of price-fixing while heavily implying it, and "psy-op" framing attributes intent that no source in the piece establishes.
Scorecard (updated after second pass)
Roughly 60 distinct checkable claims inventoried across two passes. Current tally: about 36 VERIFIED or VERIFIED IN SUBSTANCE, 7 VERIFIED W/ CAVEAT, 3 ERRORS (Oracle's post-downgrade rating stated as BBB/A-2 instead of BBB-/A-3; the A100/2020 dating; the 2002 DRAM-fine compression), and roughly a dozen still ATTRIBUTED or UNVERIFIED, mostly single-source details (the 60% memory-order figure, the named Goldman analyst, the Wisconsin steel beam, the Middle East merger funding, the received-to-date round total). Nothing checked was fabricated, and the second pass moved every one of its eight resolvable targets toward Zitron rather than away: the sold-out memory timeline landed on his exact week, the 21,000 Oracle layoffs are in the 10-K, the $244B bond figure matches WSJ's six-issuer definition down to the parenthetical, the Citadel language is nearly verbatim from the Rubner note, and the Japan market-cap claim, the most extraordinary in the piece, is substantively true.
The errors found remain peripheral. The systematic bias is unchanged and worth restating: accurate facts framed with maximal intent-attribution (deals as deliberate stock manipulation, the memory LOI as manufactured crisis), scope occasionally widened (the eMarketer US figure as the world; the 80% Plus-subscriber decline without the offsetting Go-tier migration it belongs to), and his own estimates ($852B burn, $748B RPO) stated with the same confidence as sourced facts.
On the LinkedIn quote specifically: the passage Dagmar Monett excerpted is the article's interpretive thesis, not a factual claim, so it cannot be verified or falsified. What can now be said with more force than after round one: the factual record underneath it (the unwound deals, the ratings language, the tranche-funded round, the subscriber and ad projections, the debt issuance, the layoffs) is substantially as Zitron describes it.
Still open after two passes: claim 2 (round proceeds received to date) plus the single-source details noted above. None is load-bearing for the article's argument.