Coffee Break: OpenAI Is Sinking Fast, Who Will It Pull Down With It?

Sam Altman’s OpenAI empire is imploding fast. Ed Zitron’s reports on the financial blackhole at its heart have been confirmed by the FT and key execs are exodusing faster than even Sam can say, “IPO at a trillion dollar valuation or nothing.” Tons of news, let’s move fast. The Mighty FT Confirms Ed Zitron’s Reporting And if you know the kinds of things Ed Zitron’s been reporting about OpenAI’s finances, you know that’s really really really bad…for Altman and OpenAI. And maybe for the whole LLM-based US economy. Here’s what Ed found and FT confirmed: OpenAI spent $34bn last year as the ChatGPT maker poured money into a race to dominate the fast-growing AI market ahead of a planned stock market listing.Audited financial figures confirmed by people familiar with the matter show the company spent about $19bn on research and development in 2025 and nearly $6bn on sales and marketing, as well as other costs. The spending figures, up sharply from the previous year, offer a rare glimpse into the economics underpinning the AI boom, particularly OpenAI’s lavish outlay to build models, fund data centres and recruit top researchers. The numbers, which were first shared with the FT by independent journalist Ed Zitron, suggest OpenAI’s revenues are outstripped by rising costs. OpenAI booked about $13bn in revenue last year. By the end of 2025 it was generating $2bn in monthly revenue, up from $1bn a quarter at the end of 2024, making it one of the fastest-growing businesses in history. But heavy spending contributed to a nearly eightfold increase in the net loss attributable to OpenAI, which soared from $5bn in 2024 to around $39bn in 2025. A person familiar with the matter said the large majority of that jump reflected a non-cash accounting charge linked to the company’s previous structure rather than its underlying operations. Before we go back to the FT’s “person familiar with the matter”, let’s also look at some numbers from the WSJ, via Gary Marcus: | OpenAI, 2026 | Q1 | Q2 | Move | |---|---|---|---| | Revenue | $5.7bn | $6.7bn | +18%, or $1bn | | Operating loss (incl. stock comp) | $9.3bn | $12.3bn | +$3bn | | Anthropic revenue, same quarter | — | $11.6bn | more than doubled, small profit | | Figure | Source | |---|---| | All OpenAI Q1 and Q2 2026 figures | WSJ, 18 Aug 2026, by Berber Jin and Corrie Driebusch (free version: Yahoo Finance) | | Anthropic Q2 revenue and operating profit | Same WSJ report | | Anthropic’s “small profit” | Two months of that quarter ran on discounted SpaceX compute, per the SpaceX S-1 — noted by Ed Zitron in Vanity Fair | In the quarter before it hoped to list, OpenAI added a billion dollars of revenue and three billion dollars of losses. It lost roughly $1.84 for every dollar it took in — worse than the $1.22 Ed Zitron has been citing for a while. Now back to that whisperer who talked to the Pink Lady, er the Financial Times. The FT’s anonymous source goes on to “explain” how OpenAI’s family blogging conversion from a non-profit to a for-profit (so it could IPO at a $1 trillion valuation because it’s incredibly desperate for cash — and it’s easy to see why, that furnace needs to be fed!) forced the conversion of early investors “convertible interest rights” (ie not conventional equity) which racked up an eye-popping $41.55 billion charge. This allegedly cuts OpenAI’s loss to “only” $8 billion — after “stripping out the charge and other non-cash expenses, such as stock-based compensation of staff and computing credits from Microsoft.” Ha. Ha. Ha. That’s like me saying “after stripping out my excess fat, skeleton and muscular system I weigh a healthy 95 pounds.” This financial hemorrhaging explains the executive exodus. Thirteen, Make That Fourteen Top Execs Exit OpenAI Business Insider has the list: The ChatGPT maker has lost a string of senior leaders in 2026, including former operating chief Brad Lightcap, applications CEO Fidji Simo, chief revenue officer Denise Dresser, and executives overseeing marketing, enterprise products, science, safety, and ethics. Some left to start new ventures, while some stepped back for health reasons. - Chief revenue officer Denise Dresser, started December 2025, left August 2026. - Chief operating officer Brad Lightcap, started 2018 as CFO, became COO in 2022, left August 2026. - CEO of applications Fidji Simo, started May 2025, stepped down July 2026. - Chief product officer, later VP of OpenAI for Science, Kevin Weil, started 2024, left April 2026. - Head of Sora Bill Peebles, started 2023, left April 2026. - Chief technology officer for business applications Srinivas Narayanan, started 2023, left April 2026. - Chief marketing officer Kate Rouch, started December 2024, stepped down April 2026. - Enterprise AI sales lead Barret Zoph, returned January 2026 after first leaving in September 2024, left June 2026. - Head of ethics Chloé Bakalar, started August 2025, left July 2026. - Head of safety systems Johannes Heidecke, started 2021, took charge of safety systems in 2024, left July 2026. - Chief futurist Joshua Achiam, started 2017, left July 2026 after nearly nine years. - Head of robotics and consumer hardware Caitlin Kalinowski, started November 2024, resigned March 2026. - Vice president of sales for the Americas Kaylin Voss, start date not given, resigned August 2026. - Head of data centers Chris Malone, started March 2025, left August 2026. Chris Malone was in the Biz Insider piece but not counted, WSJ reported on his exit. A Company About to do the Biggest IPO Ever Doesn’t Bleed Out C-Suite Talent As Eli the Computer Guy keeps reinforcing in his video on the OpenAI meltdown this is really weird: I hear this from some of the viewers. They’re like, “Eli, Eli, this is normal before the IPO. You’re just being melodramatic. This is clickbait.” Here’s the deal. Most of the time when you see departures before an IPO, essentially what they do is they hire adults. You will hear this in the startup technology world — the whole question of when do you bring the adults into your company. What that means is you are bringing in people that have experience running publicly traded companies. They know what is required out of a publicly traded company. They know what the investors are looking for. And so if you’re going to IPO and you have this snot-nosed college dropout that is currently the CEO, the idea is you bring in other executives, you flip the executive roles to bring in adults, to make it seem better for the whole IPO thing. The thing that’s weird here with OpenAI is that this is a massively valuable company that has already had access to basically whoever the hell they want. Whether or not people actually understand what it is, everybody thinks they understand what artificial intelligence is. And so they should not have had a problem being able to bring in good executives. And again, the executives that are leaving, these are not kids. The chief revenue officer was not a kid. This was not some snot-nosed girl getting her master’s degree in economics that happened to get the chief revenue officer position. She was poached from Salesforce, from Slack, had spent a decade there, and she bounced out. Her subordinate, the vice president of sales for America — again, years of experience — she bounced out. And that’s the thing to understand about this. This is not the kids being replaced by the adults. This is the adults running for the f*cking door. These are the experienced technology executives that have cut their teeth in Silicon Valley over a decade or more, looking at OpenAI and going “f*ck no,” and running away. This should theoretically be the third most valuable IPO in US history. Even if we’re not talking about generational wealth being created, simply having that token to say “I was there” is incredibly valuable in the technology world. These guys aren’t willing to stick around five or six more months. Really? It’s weird, dude. It’s f*cking weird. So, we’ll let that suffice for our case that OpenAI is screwed. What does it all mean? OpenAI Might Pull Oracle Down With It Summarizing Ed Zitron’s mid-August piece “How Much Money Does AI Need?“, OpenAI’s contractual obligations include: - $750bn compute through 2030 (WSJ) - $50bn compute in 2026 (OpenAI’s own figure, from the Musk trial) - $800bn needed over three and a half years to meet all commitments (Zitron’s calculation) - $300bn five-year Oracle deal OpenAI and Anthropic together need to raise more than $1.1trn — they raised $217bn combined across all of 2026. (WSJ) Zitron also reports that the Stargate Abilene project that Sam Altman, Larry Ellison, and POTUS Trump announced with such fanfare in January 2025 is behind. Way behind. Only three of eight buildings energised and monetised, per sources with direct knowledge of Oracle Cloud Infrastructure. Wisconsin has been even less accommodating. In July the state’s Public Service Commission tightened its credit rules for very large power users, leaving Oracle facing a $7bn collateral requirement to guarantee power for the Port Washington campus — a bill that could run north of $100m a year. The trigger was Oracle’s own credit rating: the tariff bites on developers rated below S&P A−, and S&P had just cut Oracle to BBB−. The rule exists so that ratepayers, not Oracle, aren’t left holding the bill for a data center that never gets finished. Then in August the PSC unanimously revoked the “completeness determination” for the transmission project meant to power the site. American Transmission Company had filed or refiled 564 documents since the original sign-off, and regulators gave up trying to evaluate what they called a moving target. Chair Summer Strand said the application was “clearly underwater, nearly underwater” and that pulling the determination was “akin to tossing it a life preserver.” ATC has to start the whole application over — and neither the company nor a ratepayer watchdog can remember the commission ever doing that before. And let’s not forget the debt that OpenAI funders like Microsoft are carrying is even bigger than reported, via the WSJ — Nat Wilson Turner (@natwilsonturner) September 2, 2026 Then there’s Ed Zitron’s Hater’s Guide to Circular Financing, Part One which reports that, if OpenAI can’t make it to a public offering, Nvidia is an obvious candidate to feed it at least another $10bn. He also notes Nvidia’s $6.3bn deal to buy back unused CoreWeave compute has never been triggered — because the compute is mostly being used by OpenAI. He also reports that Nvidia’s Q2 FY2027 revenue was over $96bn, up 106% YoY. But 70% of accounts receivable came from five customers and 44% from three, and days-sales-outstanding jumped from 45.4 to 59.6 days — Nvidia is now letting some customers pay three months to a year after delivery, booking revenue long before cash arrives. CFO Colette Kress guided to 70% revenue growth in FY28, about $674bn. The 10-Q shows $366bn in commitments and $25bn of data-centre leases not yet commenced. It takes Zitron tens of thousands of words to explain all this crap, but I’ll just let Bloomberg do a few billions worth of work with this graphic from their “AI Circular Deals: How Microsoft, OpenAI and Nvidia Keep Paying Each Other” — Nat Wilson Turner (@natwilsonturner) September 2, 2026 So How Much Trouble Is Oracle In? This much: | The Oracle commitment | Figure | Against | |---|---|---| | Contract value | $300bn | five years, compute begins 2027 | | Annual obligation | $60bn | vs $26.8bn annualising OpenAI’s Q2 | | Oracle total backlog (RPO), end FY2026 | $638bn | up 363% year over year | | Share of that backlog tied to OpenAI | ~$300bn | roughly half, per S&P | | Figure | Source | |---|---| | $300bn contract, five years, compute begins 2027 | WSJ, 10 Sept 2025 (Reuters wire); 4.5GW and 2027 start via Tom’s Hardware | | $60bn a year | Fierce Network; Yahoo Finance | | OpenAI Q2 2026 revenue of $6.7bn (annualised here) | WSJ, 18 Aug 2026 (free version: Yahoo Finance) | | $638bn backlog, up 363% | Oracle FY2026 Q4 results; Oracle FY2026 10-K | | Roughly half the backlog tied to OpenAI | S&P, via Yahoo Finance; DCD on the RPO trajectory | It’s not just that what Oracle is doing is bone stupid, it’s that the normies at CNBC noticed back in March and called it out in a brutal piece titled “Oracle is building yesterday’s data centers with tomorrow’s debt.” Feel Larry Ellison’s pain: Oracle secured the site, ordered the hardware, and spent billions of dollars on construction and staff, with the expectation of going bigger. An Oracle spokesperson declined to comment. It’s a logical decision for OpenAI, which doesn’t want older chips. Nvidia used to release a new generation of data center processors every two years. Now, CEO Jensen Huang has the company shipping one every year, and each generation offers a leap in capability. Vera Rubin, unveiled at CES in January and already in production, delivers five times the inference performance of Blackwell. For the companies building frontier models, the smallest improvement in performance could equate to huge gaps in model benchmarks and rankings, which are closely followed by developers and translate directly to usage, revenue, and valuation. That all points to a bigger problem at play. For infrastructure companies, securing a site, connecting power and standing up a facility takes 12 to 24 months at minimum. But customers want the latest and greatest, and they’re tracking the yearly chip upgrades. Oracle’s added challenge is that it’s the only hyperscaler funding its buildout primarily with debt, to the tune of $100 billion and counting. Google, Amazon and Microsoft, by contrast, are leaning on their enormous cash-generating businesses. Meanwhile, Oracle partner Blue Owl is declining to fund an additional facility, and [Oracle] plans to cut up to 30,000 jobs. Oracle did cut at least 20,000 jobs and things have only gotten worse since because they’re talking about cutting more jobs and I’m hearing rumors that the cuts were BONE deep. I’m talking critical HR, legal and billing functions are not functioning according to my sources who are Oracle clients. I’ve shown in previous pieces how the Ellison Empire depends on OpenAI more than ever because: - the Gulf States are on the hook to pay for up to 25% of Paramount’s WBD acquisition and they’re suddenly broke because Iran war - Oracle’s share price has declined and its corporate debt has exploded — S&P has since cut it to BBB−, one notch above junk - The crisis in private credit means that pool of cash is also drained — banks have spent months trying to offload more than $50bn of construction debt on data centres leased to Oracle. And as for the Trump administration, they’ve already come down hard on Anthropic and taken the side of OpenAI, not smart when enemies are circling. Please follow and support my work at NatWilsonTurner.com. Related: - Stressed Oligarchs at Meta, OpenAI, and the Ellison Empire (August 19, 2026) - Ellison Empire Besieged On All Fronts (July 29, 2026) - Oracle’s Exploding Debt and Diving Stock Threaten Paramount Empire (July 22, 2026) - Will the Trump Admin Buy Into OpenAI & Save Softbank? (July 2026) - OpenAI, Amazon Colluding With Trump Against Anthropic (June 15, 2026) - Has the Ellison Infotainment Empire Peaked With Paramount-WBD? (April 2026) - Weak Links Oracle, OpenAI, UAE Are Hammered by Iran War (April 6, 2026) - AI Subprime Crisis Both a Victim and Expression of Idiocracy (April 1, 2026) - Mask-Off Moment for Paramount at Nexus of Global Conflict & AI (March 9, 2026) - Larry Ellison Goes Beyond Oracle Into Military and Media (August 11, 2025) May I ask a stupid question? With such circularity one gets lost. Oracle main business in AI is as data centre owner to be leased to OpenAI and others? Otherwise this is a very good job Nat. This week NC posts are clearing a big chunk of AI fog. “This week NC posts are clearing a big chunk of AI fog.” 👍 I wonder if The Town is gonna pick it up. They already failed last week when denouncing Ruffalo. Of course not doing so could ruin the show. They did bring up the interesting point that instead of silencing the Ruffalo story Ellisons highlighted it in an alleged intent to divert attention from the problems over the Warner/Paramount merger p.s. film actor Mark Ruffalo was criticizing Ellisons for supporting the genocide via IDF contracts, thus opposes any merger with Paramount and was dencounced as antisemitic of course. And sadly The Town – of course – did not defend him. Yes. Senile Larry Ellison got himself into the least lucrative and most expensive part of the business and doesn’t have the resources to compete while simultaneously financing his son to slap together the most ambitious and least liked media empire of all time in service of the dreaded genocidal entity. He might have underestimated also the liberal “bias” in tinsel town which may turn out to be a bit useful for once. “George Clooney Defends Mark Ruffalo’s ‘Ability to Speak Freely’ About Paramount-Warner Bros. Deal and Doesn’t See How the Merger ‘Financially Ends Up Making Sense’” https://variety.com/2026/film/festivals/george-clooney-defends-mark-ruffalo-paramount-warner-bros-merger-1236848168/ “Mark Ruffalo Defended by Joaquin Phoenix, Hannah Einbinder and More Than 180 Jewish Figures After Paramount Accuses Him of Using ‘Antisemitic Tropes’ — Read the Open Letter (EXCLUSIVE)” https://variety.com/2026/film/global/mark-ruffalo-open-letter-antisemitism-claims-paramount-1236848093/ Norman Finkelstein predicted in 2012 that once liberal (Jewish) America would learn of the true nature of Project Israel they would not support any more because structurally he argues those very same American liberals politically are on a completely different page. And to some limited extent this may factor in how Ellison´s case is at last seen and pictured in public. Even Matt Belloni admits what you say “least liked media empire”. And Belloni usually just provides us with what money thinks. Counterfactually Belloni however fails to connect the dots: Obviously the genocide, IDF, Israeli suprematism (none of which he would ever pronounce on his show) are the cause. yea Belloni is a must-read for me since I follow Paramount closely for NC and The MMA Draw both and you have him nailed. Their mistake was openly allying with Trump AND Israel at the same time. That’s one toke over the line for most Americans. I wish I’d had more time & space to update about the Paramount case and OpenAI’s links to Trump but next time! I, for one, look forward to the next installment #Ditto Pretty sure the US couldn’t build a garden center in a year, let alone a multi-billion data center. As a movie fan who has been following the Skydance – Warner Brothers merger challenge, I’m looking on with keen interest at what would happen if/when Oracle implodes. I wonder if they will be able to hold out until the trial starts next March, let alone how long it takes the judge to deliver her decision. Those poor Ellisons. Might be worth while knowing that Trump has plans for Hollywood so might get involved here and put his thumb on the scales- ‘Donald J. Trump @realDonaldTrump I just had a GREAT and very interesting meeting with my Hollywood Ambassador and Movie Legend, Jon Voight! He is a fantastic man who loves our Country, and feels strongly about the Motion Picture and Television Industry. He hates what’s happened to it! It is being dissipated in its entirety. It has moved to Canada, and other Countries, with very little work being done anymore in the United States. Hollywood is a Complete and Total Disaster! Despite the name, it is getting very little work. There is no incentive to be there, and it is hurting California very badly. Jon, and many others in the Industry, are suggesting we do Federal Tax Incentives in order to Make our Movie and Television Production Business GREAT AGAIN, perhaps GREATER THAN EVER BEFORE! The amount of money spent on Tax Incentives will be made up tenfold by the money pouring into the Treasury’s coffers. Meetings are being set up with the Leaders of both Parties in order to get this done. It should be Bipartisan, especially since so much money is being lost in California, and other largely Blue States. I am going to suggest that Republicans and Democrats get together, and immediately craft Legislation to save the Movie, Television, and Entertainment Business in America. Congress should approve, immediately, a Federal Production Incentive to create Entertainment Jobs in America. It can be done quickly, accurately, efficiently and, importantly, will benefit ALL of America. What we watch on the Silver Screen should be made in what was once the Movie and Motion Picture Capital of the World. Let’s get this done! President DONALD J. TRUMP’ https://truthsocial.com/@realDonaldTrump/posts/117192406705033763 “Feel Larry Ellison’s pain” Nothing. I feel not one micro joule of that man’s pain. He has been on my “filty-dorty” list since the 80s. Sorry, I was hoping to share his pain in a sadistic, energy vampire sense. I should have been more effusive. ROFLMAO Of course I knew you were being sarcastic, Nat, and sorry if I was something of a buzzkill. It’s just that “Larry Ellison” is kind of my “Niagara Falls” of Stooges fame. See it’s just started my eye twitching again. . . “Slowly I turned…” is the line of that Shadow Stooge Bessent? The schadenfreude, it compensates… I hope OpenAI goes down in flames but people should know that Ed Zitron’s predictions for the most part haven’t aged well. This is a good wrap-up: https://danluu.com/zitron/ Building yesterday’s data center with tomorrow’s debt is applicable to every massive civil/electrical engineering project called data centers. Ed. Zitron does not spend much time on the difference between EBIT, and EBIT with depreciation…. These data centers will be sinks for cash to operate and same time stay current. Musk’s orbital chunks of compute will be upgraded with each node falling out of orbit when fuel runs out. The land data centers do not have increments of change. Not only is the cost revenue model for hyperscaler flawed but financing keeping up with the competition ( like 19th century RR over build) is not in sight of cash flow. Debt will be needed, until they can con the financial media to sell infinite shares…. Headline valuations…. Zitron a genius when the “easy” debt ends. Then new models shutdown LLM, and new chips that don’t need so much wattage….. This piece isn’t about Zitron’s predictions, it’s about his reporting which has been confirmed by the Financial Times. His prediction record is 100% irrelevant to the state of OpenAI’s business. Having read Dan Luu, I don’t think he understands Zitron’s points about “dying companies” he’s talking about them dying as “growth” investments and becoming “mature” companies with steady profits but no more exponential growth potential and losing their stock multiple. I’d have to say that Luu is either being deliberately obtuse or he didn’t understand Zitron’s arguments. Nor has he identified many “predictions” — he’s just taking statements Ed has made and turned them into “predictions” – saying “Meta is dying” is not a prediction with any kind of time frame. Zitron isn’t trying to sell stock tips or some crap like that. The same was being said of those who warned of the sub-prime real estate bubble right up until the collapse. Things along the lines of : “You claimed back in 2005 that we were heading for a catastrophe, it’s now 2007 and real estate is still booming. Seems your predictions aren’t worth much.” How did that end up working out ? I call these deniers Bubble-Beetlejuices … LOL Mention a bubble three times and they come out of a crack in the wall saying exactly this. They feel like the bubble is working on their timeline, and since it hasn’t shown up in X amount of time, it isn’t going to do so at all. They seldom have the stomach for details or historical precedence. Assuming you’ve read Karen Hao’s compendious history of Open AI, Empire of AI. It’s by far the best overview of the industry that I know. As with Khalidi’s The Hundred Years’ War on Palestine in the IP realm, one is tempted to assert that people not familiar with the terrain it surveys don’t have a comprehensive awareness of the issues at stake. You shouldn’t assume. I’ve skimmed Hao but never read it. Same with Khalidi. I need to read more books! Thanks for this post. Great post. For some reason or other this sounds like good news. :) And best of all, at long last it’s being reported, or starting to be reported in that most mainstream of MSM newspapers. (Maybe the FT editors read NC . / ;) I do hope to see the Ellisons go down with the shit ( Not a misspelling), although I expect Sam Altman to come out of this in relatively good shape. Knowing that Trump is once again backing a loser makes it sweeter and this will also cause a lot of people to start asking rude questions like “How do you plan to make a profit?” and “How are you going to pay for all of this?”. It’s going to be fun to watch, not as much fun as the time the hogs et Grandma, but close. I’m just hoping the answer isn’t “with an enormous pile of gifted taxpayer money, in exchange for favours to Donald to be named later, or possibly kept out of public view entirely.” I hope she was dead first. re: In the quarter before it hoped to list, OpenAI added a billion dollars of revenue and three billion dollars of losses. It lost roughly $1.84 for every dollar it took in — worse than the $1.22 Ed Zitron has been citing for a while. —— But, but.. they make it up on margin. (old joke) NVIDIA puts out a new generation of data center processors every 12 months… I wonder, how this foolish poursuit is even possible. Big Thanks for the article. The shadenfreude it helps us to cope with this crazy crazy world. related – interview Taxpayers Must Not Be on the Hook for Bailing Out the AI Industry Hundreds of billions of dollars are flooding the AI industry. Economist Gerald Epstein says this bubble could burst. Sept. 2 https://truthout.org/articles/taxpayers-must-not-be-on-the-hook-for-bailing-out-the-ai-industry/ Thanks for this link. Which banks, hospitals, telecoms, SAPs, airlines, insurers, militaries, governments, etc., are dependent upon Oracle databases/data centres, I wonder. And are they taking measures in light of Oracle about to go wonky. Those who invested in Oracle on the basis that it was a safe bet, not engaged in highly speculative and risky schemes, in other words. A lot of them are probably dependent on Oracle and moving critical IT infrastructure is a large project which isn’t undertaken lightly. In theory it should be easy to copy from one set of servers to another, but practice is the difference between theory and practice. I think that when the bubble burst, most likely the traditional non-AI data centers will continue as before. If Oracle goes belly up, this is the profitable core that a bankruptcy manager would try to restructure around or sell of to pay down debts. This is not financial or IT structure advice. If Oracle has started firing lawyers, it means their end is near. Superlative breakdown, Nat!! Thanks as always!