Pouetchoune
LightNews LightNews
banner
manette2play.bsky.social
Pouetchoune
@manette2play.bsky.social
50 followers 130 following 340 posts
Mamie beauf. She/her/Elle
Posts Replies Media Videos
manette2play.bsky.social
Écoutez kazdoura !
December 13, 2025 at 7:41 AM
Reposted by Pouetchoune
pookleblinky.bsky.social
Sam Altman: hey I made a machine that stole literally all your stuff in order to make CSAM of your child actors, sometimes it encourages people to kill themselves

Disney: Holy shit here's a billion dollars

Altman: the machine loses more than that each week but thank you. Soon it will become god.
December 12, 2025 at 10:27 AM
Reposted by Pouetchoune
padrepio.bsky.social
🫠 Les Sarkozy interdisent à Libé et au Monde d'assister à une dédicace à Menton - seulement eux.

Réponse du Parr... du père.
Devant la Librairie de la presse, les communicants font rentrer les médias nationaux et internationaux. Deux étudiants de Sciences Po Menton, fuyant leurs révisions de partiels, décident d'interroger l'ex-chef de l'Etat sur la censure de la presse qu'il exerce en interdisant au Monde et à Libération l'entrée. Réponse de la communicante : « De quoi je me mêle ? » Nicolas Sarkozy, lui, s'agace : « Nous sommes en France, on peut encore décider qui rentre ou pas. » A côté de son père, Louis Sarkozy est resté, selon leurs dires, « en mode figurant».
December 13, 2025 at 6:28 AM
Reposted by Pouetchoune
padrepio.bsky.social
Ça aura failli rester une histoire cool et sympa pendant deux mn avant que tout le monde décide d'en tirer le plus de pognon possible aussi vite que possible.
leparisien.fr Le Parisien @leparisien.fr · 2d
Thierry Cotillard, président du groupe Les Mousquetaires, a annoncé la production d’une peluche à l’effigie du loup de la pub qui a fait sensation sur les réseaux sociaux.
« On travaille dessus » : la peluche du loup « Mal aimé » de la pub Intermarché bientôt en vente
l.leparisien.fr
December 12, 2025 at 8:01 PM
Reposted by Pouetchoune
mazalbof.bsky.social
Heureusement, décision a été prise de *vérifie ses notes* supprimer 4.000 postes 🤓
December 12, 2025 at 12:51 PM
Reposted by Pouetchoune
arielries.bsky.social
"people with renaissance-level art skills don't exist anymore" yes they do their username is something like 03748218_ and they spend all their time drawing pornography so deranged it'd kill hugh hefner a second time
December 12, 2025 at 12:06 PM
Reposted by Pouetchoune
davesheik.com
Nouvelle vidéo : comment la Chine met en place des stratégies pour revendiquer Taïwan

Un ptit partage peut accomplir beaucoup si ça vous embête pas 👉🏾👈🏾

youtu.be/xi3-a_lj65E
Le plan secret de la Chine pour récupérer Taïwan
YouTube video by Dave Sheik
youtu.be
December 11, 2025 at 5:12 PM
Reposted by Pouetchoune
edzitron.com
Wow $1bn? That should cover a month of OpenAI’s inference costs
moreperfectunion.bsky.social More Perfect Union @moreperfectunion.bsky.social · 3d
BREAKING: Disney is making a $1 billion investment in OpenAI, and will allow its iconic characters to be on the company's Sora AI video generator.
December 11, 2025 at 3:59 PM
Reposted by Pouetchoune
marmelade.bsky.social
La préfecture du Rhône interdit les concerts du festival antifa de Lyon,dont celui du groupe punk Poésie Zéro.

Elle justifie en citant "des textes on très violent, hostiles au système, à la police et la société de surveillance."

Le fascisme est là.

www.leprogres.fr/politique/20...
La Préfecture prend un arrêté pour interdire les concerts de l'Antifafest Le festival Antifafest doit accueillir des conférences sur plusieurs thématiques mais aussi des concerts, du 11 au 13 décembre 2025, à la salle de la Rayonne et à l'Île Egalité à Villeurbanne. Mais la préfecture a pris un arrêté d'interdiction de la tenue de l'ensemble de ces concerts. Cet arrêté cite plusieurs paroles des chansons des artistes invités à se produire : « Considérant que le groupe de punk rock ska parisien Poésie Zéro est connu pour ses textes au ton très violent, hostiles au système, à la police et la société de surveillance » [...] « considérant que le rappeur lyonnais L'Allemand incite également à la haine contre l'institution policière », peut-on notamment lire.
December 10, 2025 at 4:16 PM
Reposted by Pouetchoune
marmelade.bsky.social
Dans le détail, voilà les paroles qui appuient la censure préfectorale.
CONSIDÉRANT que la programmation musicale des concerts des 11, 12 et 13 décembre comprend plusieurs groupes de punk rock ou de rap, coutumiers de propos « anti-système » et « anti-police » tels que Poésie Zéro, Costa, L’Allemand, ou Da Uzi ; CONSIDÉRANT que le groupe de punk rock ska parisien Poésie Zéro est connu pour ses textes au ton très violent, hostiles au système, à la police et la société de surveillance ; que parmi leurs titres incitant à la violence, peuvent être cités : • « 1312 » - ou de manière décodée ACAB (All Cops Are Bastards) – dénonce la violence policière et systémique ; que la pochette du titre présente un véhicule de police en feu ; que les paroles appellent à la violence de manière univoque : « tous les jours le 13 décembre, tous les jours du feu des cendres, au milieu des cortèges, des cocktails pour se détendre » ; • « Technoflic » dénonce la surveillance de masse et l’oppression policière dans une société hyperconnectée en ces termes : « des caméras qui te matent dans des rues connectées, qui comptent ton nombre de pas, qui savent ce que t’as bouffé, des flics qui te contrôlent avec des algorithmes, qui décident si t’es un bon ou un mauvais citoyen » ; • « CRS » s’attaque directement aux membres des Compagnies Républicaines de Sécurité et à leur rôle dans la répression sociale. Le titre se termine par « CRS SS » ; • « Il faut voler » incite à commettre des infractions au nom de l’inégalité sociale ; CONSIDÉRANT que le rappeur parisien Costa revendique son « appartenance au mouvement antifa ainsi que sa proximité avec l’univers des ultras » ; qu’il exprime régulièrement des propos incitant à la haine et à la violence contre les forces de l’ordre ; que dans le morceau « Coca Cherry », il indique avoir « la haine contre les schmitts » ;
December 10, 2025 at 5:00 PM
Reposted by Pouetchoune
vincentgranier.bsky.social
Les députés approuvent en commission, en 10 mn, la possibilité d'imposer à un professionnel de santé libéral la procédure de MSO.
La procédure vise essentiellement à contraindre les médecins à réduire leurs prescriptions d'arrêts de travail.

Dans un texte visant à "lutter contre les fraudes".
vincentgranier.bsky.social Vincent Granier @vincentgranier.bsky.social · Oct 15
#Médecins libéraux - Rappel que le PJL de lutte contre les fraudes fiscales et sociales, présenté hier en conseil des ministres, intègre une déclaration de guerre contre les médecins traitants:
-> la mise sous objectifs (MSO) ne pourra plus être refusée
December 10, 2025 at 6:34 PM
manette2play.bsky.social
J’ai fais une carte d’anniversaire pour une amie, ça faisait loooongtemps que j’avais pas dessiné! Ça fait du bien en vrai!
December 10, 2025 at 8:40 PM
Reposted by Pouetchoune
nhervieu.bsky.social
« Travailler plus pour gagner pl… Ah non en fait »

« La valeur "travail" est primordi… Oups. Raté aussi »

Quand tous les slogans politiques dominants sont contredits par les faits.

Une tartuferie au détriment des classes moyennes (dont la colère est évidemment réorientée vers les plus pauvres).
Graphique « La part de revenus allant au capital augmente, celle du travail baisse » 61% pour les revenus du travail en 1980 & 39% pour le capital. En 2025, c’est 53 & 47 % respectivement Politiquement, cette tendance est explosive. Cet effritement des classes moyennes, ces difficultés croissantes à boucler les fins de mois alimentent un mécontentement social sur lequel prospèrent les populismes, en Europe comme aux Etats-Unis.
December 9, 2025 at 7:53 AM
Reposted by Pouetchoune
culturecrave.co
Miles Morales co-creator Sara Pichelli says she does not get any royalties from the ‘Spider-Verse’ films, Insomniac games, or merch

"I don’t get anything … that's the saddest part of my life"

(via SiteJamesons | TW)
December 7, 2025 at 2:47 AM
Reposted by Pouetchoune
edzitron.com
Every AI startup is unprofitable, and a scenario that will burst the bubble would be a notable AI startup struggling to raise money, leading to others rushing to raise, leading to others rushing to sell their companies, leading to industry panic and the end of AI VC.
wheresyoured.at/premium-the-...
A Startup Rattles The Venture Part of the Chain of Pain We are currently in the "fucking stupid" part of the AI bubble, as evidenced by the New York Times' piece about how "A.I. Deal Making Is Getting Faster and Faster": Across Silicon Valley, investors have been doing anything they can to get into the hottest A.I. start-ups. One venture capital firm flew the 22-year-old founders of Mercor, an A.I. company valued at $10 billion, on a private jet to Las Vegas to race Ferraris. Another gave college students money to start companies rather than do internships. Others are acting as if they work for young start-up founders, making customer introductions and recruiting employees. All of them are moving at a breakneck pace. ... Investors argue that the A.I. opportunity is so enormous and that potential winners will be so successful that deal prices, especially when the companies are just starting out, do not matter. What matters is getting into the best start-ups, they said. If A.I. has boosted the valuations of Nvidia, Microsoft and Alphabet into the trillions, perhaps several more trillion-dollar companies are getting started today. This is, of course, extremely stupid logic. And things get even more stupider: Vibecode, a start-up that uses A.I. to let people build apps, set out to raise $3 million last spring. It had not released its app, but investors were so eager that the company garnered $15 million in offers in a week. Vibecode wound up amassing $9 million at double the valuation it initially sought. In August, two months after releasing its app, it rocketed to $1 million in annualized monthly revenue. You may wonder what Vibecode does, and the answer is "create native apps in seconds with AI." Its $50-a-month and $200-a-month plans offer $55 and $220 worth of credits a month, and it adds at the bottom in small writing that its "pricing tiers are experimental and may evolve" and that it is "...still trying to figure out the exact way to price this product, and… Yet this quote really explains the moment: Gaurav Jain at Afore Capital has adjusted his investment criteria. He previously sought out great founders with great ideas. Now, he said, he is willing to back great founders with not-so-great ideas — or no ideas at all. The strategy is risky, but Mr. Jain’s firm helps entrepreneurs develop their ideas.“We cannot wait for the companies to have an idea and take off, because then it’s too late,” he said. This is what it looks like when you're in a bubble: irrational, ridiculous, clownish investment plans that appear to involve somebody having a pulse and saying "AI." As a result, startup funding continues to grow at a worrying pace, with AI taking up increasingly-large swaths of funds and attention. Venture capitalist Jon Sakota estimated back in August that US venture capital would run out of money by the end of 2026 if it continues at its current pace, which it is going to have to to support this industry. According to Pitchbook in September, the picture is grimmer, projecting around $100 billion of venture capital dry powder available at the end of the year. At this point, the scenario begins: In early 2026, a VC-backed startup needs another $200 million to keep shoving it into a furnace with "GPU" written on it, and investors balk at the valuation, the cost, or both. This story spreads like wildfire. The vibes-based community of venture investing starts to get nervous, and the startup in question starts meeting with hyperscalers, OpenAI, and Anthropic in the hopes that it can be swept off the market. The problem isn't so much this company, but the effect that it has on the rest of the industry. AI firms, nervous about their future ability to do business, start trying to raise further rounds as a means of staying alive. Rumblings around data center deals falling apart begin to rankle venture capitalists further, and a few of them start making calls suggesting it's time for portfolio companies to exit. The moment this begins, panic sets in, as because AI companies lack any path to profitability, they effectively have three choices: get acquired, go public or die. Selling is the only option, and calls start being made. The problem becomes obvious: AI startups are, for the most part, doing fancy "prompt engineering," connecting to models from either Anthropic or OpenAI to provide their services. A few quick, hasty acquisitions happen — such as the $400 million all-stock acquisition of Neptune by OpenAI — but it creates a veil of uncertainty and anxiety. Eventually, a major funding round falls apart, creating a full-blown panic attack in venture, and VCs begin refusing AI deals, afraid of over-exposure. The big "oh shit" moment comes as a major player decides to make the ultimate mistake: instead of dying with dignity or selling to somebody else, they choose instead to go public.
December 5, 2025 at 5:36 PM
Reposted by Pouetchoune
edzitron.com
AI is funded by debt to buy GPUs and build data centers, which are rented to AI startups funded by debt and venture capital, who then rent their models to other AI startups funded by debt and venture capital. This "chain of pain" breaking will burst the bubble.
www.wheresyoured.at/premium-the-...
Part 3: The Chain of Pain While many people talk about how circular the Al bubble may or may not be, the reality is that it's far more like a chain — a deeply vulnerable one held together by debt and venture capital. A company buys GPUs from NVIDIA, at which point nobody is making any profit anymore. These GPUs are purchased, for the most part, using debt provided by banks or financial institutions. While hyperscalers can and do fund GPUs using cashflow, even they have started to turn to debt. At that point, the company that bought the GPUs sinks hundreds of millions of dollars to build a data center, and once it turns on, provides compute to a model provider, which then begins losing money selling access to those GPUs. For example, both OpenAl and Anthropic lose billions of dollars, and both rely on venture capital to fund their ability to continue paying for accessing those GPUs. At that point, OpenAl and Anthropic offer either subscriptions - which cost far more to offer than the revenue they provide — or API access to their models on a per-million-token basis. Al startups pay to access these models to run their services, which end up costing more than the revenue they make, which means they have to raise venture capital to continue paying to access those models. Outside of hyperscalers paying NVIDIA for GPUs out of cashflow, none of the Al industry is fueled by revenue. Every single part of the industry is fueled by some kind of subsidy. As a result, the Al bubble is really a stress test of the global venture capital, private equity, private credit, institutional and banking system, and its willingness to fund all of this forever, because there isn't a single generative Al company that's got a path to profitability. Our stock market is held up by NVIDIA's continued ability to sell GPUs, which is, for the most part, held up by its customers continued ability to raise debt, which is held up by its customers' ability to find other customers to pay to rent those GPU… To quote my piece coming out Monday: Who, after all, are the biggest customers renting Al compute? That's right, Al startups, all of which are deeply unprofitable. Cursor — Anthropic's largest customer and now its biggest competitor in the Al coding sphere — raised $2.3 billion in November after raising $900 million in June. Perplexity, one of the most "popular" Al companies, raised $200 million in September after raising $100 million in July after seeming to fail to raise $500 million in May (I've not seen any proof this round closed) after raising_$500 million in December 2024. Cognition raised $400 million in September after raising $300 million in March. Cohere raised $100 million in September a month after it raised $500 million. The Al industry is venture capital feeding money to OpenAl and Anthropic, who then feed that money to either hyperscalers or neoclouds, who then feed that money to either pay for power or buy more GPUs from NVIDIA. Anything that destabilizes a link in the chain could break the entire thing. There are many ways this scenario can go, so I'm going to lay out a few options.
December 5, 2025 at 5:36 PM
Reposted by Pouetchoune
edzitron.com
Yet the real scenario I expect is that CoreWeave - a deeply unprofitable neocloud with $14bn in debt that must keep raising to build more capacity - fails to deliver compute in time, defaulting on its contracts, then defaulting on its loans, then dying.
www.wheresyoured.at/premium-the-...
CoreWeave: The Great Bellweather CoreWeave's earnings are next week, so I'm going to put together a scenario based on things looking okay, but not great for yet another New Jersey-based monstrosity. Right now, CoreWeave has about $14 billion of debt, and already defaulted on part of its loan with Blackstone by moving assets out of the country earlier in the year. CoreWeave has a giant, ugly backlog of $55.6 billion — which may seem impressive, but it is mostly made up of $22.4 billion in OpenAI contracts, $14 billion in Meta contracts, and whatever Microsoft has agreed to previously. Furthermore, it's not obvious how CoreWeave accesses that revenue, as Core Scientific has yet to finish the capacity so that CoreWeave can get the money. Eventually, CoreWeave needs to start paying off those big, nasty loans. CoreWeave already had to spend $310 million in Q3 2025 just on interest from its loans, up from $266 million in Q2 2025. CoreWeave's current debt situation has it paying billions of dollars a year in principal payments on top of the interest on these loans, on top of operating expenses that effectively nullify any and all revenue coming into the company. Such a huge backlog means that CoreWeave, like Nebius, is likely to start running up against defaults on its contracts. Sources familiar with the matter tell me that Microsoft had been, at least in June 2025, propping up auxiliary capacity to cover CoreWeave's incomplete Denton facility. CoreWeave has, at this point, had to raise money at IPO ($1.5 billion) in March, a loan in May ($2 billion) and a loan in July ($2.6 billion), then added a further $3 billion in October to an already-existent line of credit. Here's why CoreWeave needs to keep raising money: Its operating expenses are higher than its revenues. Its revenues and contracts require it to constantly expand, either through signing leases with companies like Applied Digital and Core Scientific (which it tried and failed to buy), or building its own capaci… Eventually, something snaps with CoreWeave, and it needs more money, only to find the banks are unwilling to keep extending lifelines. Continued delays make it unable to continue increasing revenues, because...well...you can only charge for capacity that exists. Here are some scenarios about what could go wrong! CoreWeave defaults on its OpenAI and Meta contracts: Unable to build the capacity in time, Meta and/or OpenAI action a clause in its contract to cut CoreWeave loose, savaging its backlog. The bad news sends the stock into a tailspin. OpenAI defaults on its contract: Swagged-up white boy Sam Altman's pity party comes to a close, and OpenAI simply can't make the payments for the capacity it's already agreed to. The news of this leak spooks the markets, and investors start dumping CoreWeave. CoreWeave's financial condition decays and it cannot pay its loans: While I expect CoreWeave is likely on great terms with funders like Blackstone and Magnetar Capital, eventually somebody is gonna need some money. CoreWeave really has no path to success. Its costs rise with revenues, it effectively lives on debt, and at some point it will need to make more money than it spends, so much more that it actually starts paying off these loans rather than riding out the interest.
December 5, 2025 at 5:36 PM
Reposted by Pouetchoune
edzitron.com
The next AI bubble burst scenario is the collapse of neoclouds - specialist AI data center companies that have tons of debt entirely for buying GPUs and are deeply unprofitable, like Nebius, CoreWeave and Lambda.
www.wheresyoured.at/premium-the-...
Part 2: The Great Neocloud Insolvency So, remember how Nebius effectively used its Microsoft contract as collateral in that $4.2 billion loan? Well, it's really important to read the fine print on stuff, such as on Microsoft's Form 6-K it filed around its Nebius deal: The GPU Services will be deployed in several tranches during 2025 and 2026. Subject to the satisfaction of the deployment and availability of the GPU Services, the total contract value is about $17.4 billion through 2031. Microsoft may also acquire additional services and/or capacity under the Agreement, which would increase the total contract value to about $19.4 billion. Cash flow coming from the Agreement will be utilized to finance part of the capital expenditure associated with the Agreement. In the event that, following a grace period, Company fails to meet agreed delivery dates for a GPU Service and the Company cannot provide alternative capacity, Microsoft has the right to terminate that GPU Service. I'm a curious little critter, so I went and spoke with a source at Microsoft to find out what the timelines were on Nebius' supposed deliveries. According to my source, Nebius has convinced Microsoft it would have 8,640 GPUs (in NVIDIA GB200 racks) available in its Vineland, New Jersey facility by November 3 2025, and another 9,216 available by February 2 2026. Based on discussions with sources familiar with the matter, Nebius data center partner DataOne has only finished one out of the three buildings of the Vineland development, and while Nebius "committed to 100MW of installed capacity by the end of 2025" at the end of March, I see no proof that that's happened either. In fact, I'm having a lot of trouble even establishing what DataOne is. According to DataCenter Dynamics, it was "newly formed" in March 2025, when it received a funding round of indeterminate size from Northeast Precast, a concrete company. DataOne CEO Charles-Antoine Beyney — whose LinkedIn location is "United Arab Emirates," nearly 7000 miles from New Jersey — is also the founder of BSO, an IT infrastructure company that funded DataOne back in December 2024 to, and I quote, launch "Europe's first giga-scale AI hosting data centre located in France," saying it intended to, by 2028, "deliver a 400MW AI-compatible infrastructure," with the rollout scheduled to begin in April 25 "with an 80MW launch." I can find no proof that anything has been built. In February, Abu Dhabi based "technology group" G42 said it was working on an "AI data center in France" powered by AMD hardware." A month later in March 2025, DataOne and Nebius announced the 300MW AI data center plan in Vineland New Jersey. The reason I took such great pains to lay this out is that there isn't a chance in Hell that DataOne completes this data center on any kind of timescale that matches Microsoft's timelines. It took Microsoft over a year to even open its Fairwater data center in Wisconsin, which is "on track to complete construction and come online in early 2026," and while it's not obvious how many GPUs are in Fairwater, I'd wager it's likely where the 4600 GPUs inside its GB300 NVL72 racks are sitting. Point is: Nebius is relying on DataOne, an unproven partner backed by a concrete firm, to build a massive data center at a scale it has never achieved before. Phase one is apparently complete, with Phase Two meant to be "completed soon" according to notes from a Vineland New Jersey City Council meeting — though, to be clear, "construction" and "putting GPUs in" are two very different things entirely, especially considering it was meant to have over 17,800 GPUs in operation by the end of February. B…
December 5, 2025 at 5:36 PM
Reposted by Pouetchoune
edzitron.com
Another sign that the bubble is bursting would be the value of data center Real Estate Investment Trusts (REITs) - companies that own a bunch of data centers traded like stocks. Shares have had a great run...but are already showing signs of distress.
www.wheresyoured.at/premium-the-...
Part 1.5 — The REIT Awakening Actually, I’ll tell you what it leads to in a second, but there’s one other piece of the bank financing puzzle I need to mention — REITs. REITs is an acronym for Real Estate Investment Trust. These are, essentially, companies that own a bunch of physical assets (like houses, offices, hotels, and yes, data centers) that are traded as a normal company on the stock market. They’re essentially a way for investors to benefit from a particular type of real estate asset (and often in a specific location) without having to actually buy, own, and manage that asset. So, if I wanted to benefit from, say, commercial real estate in the US, I’d find a REIT that specializes in that, and I’d buy myself some shares. The payback comes in the form of the price appreciation of the asset (reminder: these are traded just like normal public companies on the stock market), as well as the proceeds of rent payments, which are passed back to investors in the form of quarterly dividends. As mentioned, there are data center REITs, and they’ve been having a great year. So much so, there are even more of them popping up. In October, Rick Perry — yes, that Rick Perry — launched a data center REIT called Fermi with a plan to raise $550m at a valuation of $13.16bn. Its IPO ultimately did much better, raising $682.5bn, albeit at a lower IPO value of $12.46bn. Although shares in the REIT skyrocketed by 55% in the first day of trading, they are now underwater, and Fermi’s current market cap is now $9.53bn. Not good! REITs get capital either from traditional equity investments, or from their IPOs, and many of those investors are the usual suspects — banks, hedge funds, private equity funds, and so on. The big players with the capital to move markets, and move them far. I mention this to point out that there isn’t just one way that banks can finance data centers — directly, by purchasing bonds. There are roundabout ways, and the advantage of REITs is in the fact th…
December 5, 2025 at 5:36 PM
Reposted by Pouetchoune
edzitron.com
The first scenario I have for the bubble bursting would be the end of data center debt financing, starting with banks selling off their data center debt (which has begun), dealflow slowing, deal terms getting worse, deals falling apart, then stopping entirely.
www.wheresyoured.at/premium-the-...
Part 1: The Banks Stop Funding Data Centers The entire data center debt pipeline is held up by gormless bankers and hype — and, to be specific, that the parties being loaned the money will be able to pay off the loans in question. This means that they do not believe in the long-term stability or health of these companies. At some point, these banks are going to decide that there's no reason to keep funding these things, and what will cause them to do that will be vibes plus the health of the underlying businesses. Gone are the days where one can simply fund every single data center deal imaginable and get the big thumbs up from your boss. Bloomberg's story on Applied Digital's Debt Deal, "CoreWeave-Tied Data Center Debt Tests Strength of Flooded Market," has a distinct layer of anxiety: Applied Digital’s debt sale is seen by investors as riskier than those from TeraWulf and Cipher because of its reliance on speculative-grade CoreWeave Inc. as the main tenant, the people said, asking not to be named discussing a private transaction. The offering is testing the strength of a market being flooded with debt to fund the compute capacity needed to support the artificial intelligence frenzy. Building out data center capacity is estimated to cost at least $5 trillion and will require hyperscalers to turn to every corner of capital markets, according to a recent report by JPMorgan Chase & Co. Now, I will add that this story notes that TeraWulf and Cipher Mining's bond sales both attracted significantly more interest ($10 billion on a $3.2 billion for TeraWulf and $7 billion on a $1.4 billion for Cipher), but sources tell me that, much like IPOs, there are always over-inflated order books behind any "hot" deal, and this is no indicator of any particular health in the market. In any case, this is a scenario, and so we're going to scenario hard. In this case, it doesn't even need to be one event — such as a customer default — that creates the panic. These stories about SRTs have the potential to start a panic where, say, large investors are nervous about the underlying risk of the debt being sold. This panic doesn't even need to escape into the public markets. As we speak, investing in data centers is still considered a good idea. If investment banks and hedge funds begin to get nervous about the prospect of debtors not paying, these deals will sour and then stop happening entirely. When I say "sour," I mean the terms on the loans will begin to get worse, or, as bankers get nervous, hyperscalers will backstop these deals to push them through, as happened with Google's backstop on the TeraWulf deal, where Google will (to quote Bloomberg) "step in to repay creditors if Fluidstack (the company leasing it) defaults or goes bankrupt, once the data center is operational and Fluidstack begins its lease." Do note that phrase "once the data center is operational." TeraWulf's own announcement of the debt offering said that "WULF Compute intends to use the net proceeds from the offering to finance a portion of its data center expansion at its Lake Mariner campus in Barker, New York (the “Data Center Expansion”)." Considering engineering firm Ramboll was only retained in October 2025 to redevelop the Lake Mariner/Barker campus, one has to wonder if that happens. If banks become worried about funding data center development, this creates two problems: New data center projects will not be able to get funded. Current data center projects that new capital won't be able to get a lifeline, and if there's one rule in construction, it's that things regularly go over budget. These problems accelerate due to the brittle nature of neocloud businesses. Last quarter, CoreWeave had to lower its guidance (IE: how much it would make) in 2025 because of a delay in … The Signs This Is Happening Banks actually sell off data center debt: While we've already seen CoreWeave IPO backer Magentar sell off $500 million in stock, what we're actually looking for is banks like Morgan Stanley and Goldman Sachs to start trying to shift the underlying billions of debt. Data center deals slow: At first it'll be subtle. We'll see less deals happening, and it'll be kind of hard to gauge when it's slowing until we're months in. Data center deal terms get worse: we're already in the era of data center junk bonds, bonds (think: loans) that have higher payoffs but more risk that the party in question won't pay. If the terms around these deals get worse — think 9% to 11% rather than the 7-8% we're currently seeing — it's a sign that both banks and investors are a little worried. Data center debt deals close at a lower amount than originally planned: In this theoretical scenario, a data center deal trying to raise $3 billion only manages to snag $2 billion, or $2.5 billion. This will be a sign that investors are no longer interested in taking on this risk. Data center deals begin to fall apart: as data center developers (either new or existing) attempt to build more capacity, they find that bankers can't pull together a successful debt offering, or worse still that banks aren't willing to put one together at all due to the risky nature of the deal. Data center deals stop entirely: Eventually, it becomes impossible to raise money to build an AI data center, even with a supposed tenant. Why This Would Be Bad Once this begins to happen, it will put pressure on hyperscalers to stop A) building data centers and B) signing cloud compute deals with neoclouds like Nebius, TeraWulf (providing compute to Fluidstack), CoreWeave and Lambda. This pressure will begin making hyperscalers wary of further extensive hyperscaler capital expenditures.
December 5, 2025 at 5:36 PM
Reposted by Pouetchoune
edzitron.com
Two weeks ago I said that there were likely millions of uninstalled Blackwell GPUs sitting in warehouses as there is neither the data center capacity nor the power to turn them on. Michael Burry is looking into it and it looks like he’s got evidence I’m right.

www.wheresyoured.at/the-haters-g...
Cassandra Unchained & @michaeljburry X.com Calling for pics and proof of Nvidia GPUs being warehoused in mass quantities In the US but also overseas. Some have reached out to me already. And it is getting interesting. But I need more. Kakashii @kakashiii111• 8h To give a little more context ahead of my post next week, here is some rough simple math: Let's assume everything Jensen says is true (including reading thousands of emails a day): Jensen said 6 million Blackwell GPUs have shipped. Since Blackwell is out, Nvidia reported 111B x.com/kakashiii111/s... Show more w Generation In-Service (New Build and Expansi ally Strong Demand for e Blackwell NVL72 en by Two Exponentials more than half a trillion dollars in Blackwell and Rubin revenue ough CY2026 (so far) 5X Hopper's lifetime revenue $0.5T+ 20M GPUs Shipped 6M GPUS August 2025 January - August 2025 Cumulative of ts Installed Capacity (MW) No. of Installed Units Capacity (MW) 0 3 -40 888 61 3,095 0 0 3 20 0 1 4 474 27 3,775 0 1 3
December 7, 2025 at 8:57 PM
Reposted by Pouetchoune
edzitron.com
GPUs are expensive to install, expensive to run...and the AI services that run on them *don't make much money from consumers* and lose incredible amounts of money due to the inconsistent and hard-to-predict costs of LLMs. There's no fixing this problem.
www.wheresyoured.at/the-haters-g...
AI Software Does Not Make Enough Money To Cover The Cost of Renting GPUs To Run AI Services, And Costs Vary Dramatically Buddy, I hear you yelling that I should sing Enter Sandman, and here we go! I’ve already talked about this so many times, but there really is just not that much money in AI services, in part because the services aren’t worth paying more for, and in part because costs are difficult to control. Large language models are a uniquely tough economic situation for startups — or any business, for that matter. As a business, you do not “pay to access an LLM” as a one-off cost, but pay on a per-million-tokens basis, either for input tokens (what a user loads into the model, such as text or files) or output tokens (what the model outputs through a process called “inference,” which is when it creates said output based on what a user asks). For example, OpenAI charges $1.25 per million input and $10 per million output tokens for GPT-5.1. A good rule of thumb is that one token is about ¾ of a word, and it’s important to remember that generating code is also generating characters (so letters or numbers), and one token equals one character. I also should note that the amounts that OpenAI charges for its models are not necessarily indicative of the costs of running them. In other kinds of software, you could charge a user a monthly subscription fee — $20-a-month, for example — and know roughly what a user might cost, because the unit costs are consistent. The cost of Netflix showing a 4K movie to a person in Idaho, or Meta loading Instagram for somebody in Sweden, for example. While there is some regional variation, it’s not particularly massive (at least, compared to the inconsistencies of genAI), and it’s something you can anticipate ahead of time. The “cost of generating an answer from a Large Language Model” is variable every single time based on multiple criteria: The user’s query, which might be simple, such as “what if Scooby Doo lay eggs,” or as complex as “generate the code for a piece of software that will help me keep track of The Infidel Marmaduke The Dog, Inshallah.” The model’s interpretation of the user’s input, which is when hallucinations become a problem. In the Scooby Doo egg example, a model might decide it needs to use a “web search tool” to find sources that might explain whether a dog can lay an egg, or whether Scooby Doo has, in previous canon, laid eggs. I sound like I’m joking around, but this example would lead the model on a wild goose chase, with each “tool access” and piece of information generated requiring you to generate more output tokens. In the Marmaduke Software example, remember that code is extremely token-intensive, and this is not a particularly detailed prompt. It does not specify what kind of software would track Marmaduke, and has numerous additions (infidel, Inshallah) that the Large Language Model might choose to interpret and build something that the user doesn’t want. As a result, the user might receive software that doesn’t do what the user wants, causing the user to go back and demand more software. The user’s response to the output it receives. If you’re asking a model a question, you’re likely to ask a follow-up question. And based on what the model says, you may have more follow-up questions. Every follow-up question is an opportunity for the user to demand more output from the model, and for the model to decide, based on maths, to generate more answers. If a model doesn’t do what a user wants, the user is also likely to keep asking the model stuff until it either gets it right or the user gets annoyed. This isn’t necessarily the LLM’s fault. Like any bad customer, the user hasn’t been clear about what exactly the software should do, but might be mad tha… Services try to ease this burden by imposing limits (like ChatGPT has), such as the amount of requests a user can make within a five hour period, the amount of information that can make up an input from a user, the amount of information from the conversation that a model can consider at once (called a context window), and what information can be loaded in (text, images, PDFs, and so on). The problem is that even with those limits, every single AI service is effectively intricately prompting the model on the back-end to deliver the service itself using something called a “service prompt” — think of it as a kind of primer on what kind of service you want it to perform, and also what limitations you have on said service. There is no way, it appears, to make a model be more cost-conscious, other than to tell it to not do something or give shorter answers. While you can tell a coding model to “write efficient code,” and even define specifically what efficiency is and looks like, said model is still, every single time probabilistically working out A) what the user wants and B) what the service wants the model to do as a result. In practice, this means that there’s very few ways to prevent a user from costing you far more than they’re paying you each month without directly charging them for the cost of using the models, which is far more than users are used to paying for. This means that most AI software companies are charging a monthly fee with amounts of “requests” or “credits” that a user can make. This pricing structure obfuscates the true cost of the models, all while allowing users to do whatever they want within those boundaries. For example, AI coding platform Augment Code had one user paying $250 a month that cost it $15,000 a month, and even with further changes to the product (switching from letting users make “requests” to giving them “credits), the company does not appear to be profitable. You may think that these companies will simply “raise their prices…
November 24, 2025 at 5:04 PM
Reposted by Pouetchoune
edzitron.com
A dirty little secret of the AI boom is that it appears everybody loses money on buying AI GPUs, as the incredible upfront costs (billions) put every data center in a deep, dark hole - and because GPUs are so power-hungry, it's difficult to make any margin at all.
www.wheresyoured.at/the-haters-g...
Why It’s So Unprofitable To Run AI Services Using NVIDIA’s (or anyone else’s) GPUs Before we go any further, it’s important to remember that the rest of the cloud computing industry is relatively high-margin, and that’s because the stuff it sells is, by and large, nowhere near as expensive as running GPUs. That’s because you really don’t need that much compute to serve a normal website, or a normal web application. Cloud providers can take a resource — like physical memory, or storage, or a single virtual CPU core — and sell it at a pretty steep mark-up relative to the cost of actually providing it. Ordinary computing resources are cheaper, and they can be divided across more users. As a result, the margins are better and the upfront costs are lower. By contrast, a single AI application — like one that generates a video, or an image, or that processes a large codebase — may use multiple GPUs simultaneously. The whole division of resources that makes cloud computing so efficient (and so lucrative) simply doesn’t apply here. And that’s without mentioning the cost of ownership, or indeed, the additional power costs that come from GPUs, or the cost of keeping them cool. I could keep on going. By virtue of their scale, AI data centers often require additional infrastructure that doesn’t really apply with conventional computing gear. This may involve a greater number of transformers and substations to deal with the insane power demands (all of which are extremely expensive, and have painfully long lead times), or on-site power generation infrastructure. The Stargate Abilene facility, as well as xAI’s Colossus facility in Memphis, Tennessee, all have (or will have) on-site natural gas turbines, as well as pipelines that provide the gas they burn, and all of this costs money. Again, these turbines have an incredible lead time — which is why we’re seeing creative alternatives spring up, like former jet engines being repurposed into stationary generators. Sidenote: I have no idea how efficient those converted engines actually are, but the Interesting Engineering article linked above notes that the engines being used by one such remanufacturer are the General Electric CF6-80C2 — an 80s vintage design that flew with the McDonnell Douglas MD-11 (which first flew in 1988) and the Boeing 747-400 (an aircraft that debuted the same year, and that has largely left passenger service because it’s not a particularly fuel-efficient plane to run). Obviously, there are variables that affect fuel efficiency in aviation that don’t apply to a stationary, ground-based power generator. Still, I’m curious about the economics of this, and whether these repurposed turbines have similar fuel consumption habits to, say, a purpose-built natural gas turbine! As I explained before the premium break, one does not simply “plug in” a GPU, nor does one just buy “a” GPU. You usually buy thousands, tens of thousands, or hundreds of thousands of these things, and the more you buy, the more likely it is you have to build a new data center to house them. While previous generations — Ampere (A100) and Hopper (H100, H200) — could be integrated into older data centers, new generations like Blackwell (2024) and Vera Rubin (coming next year) are so power-hungry and generate so much heat that they require specialized liquid cooling systems (which means using sealed tubes of specific coolant liquid to cool something), which in turn requires bigger, more power-intensive data center architectures to handle them. Oh, and the chips are tens of thousands of dollars more expensive than the previous generation, while also having higher electricity and cooling costs. Important detail to add: remember all that stuff I said about parallel processing and doing a bunch of maths at once? Doing all that maths is very intensive on these GPUs, meaning they’re running at full tilt a lot of the time. That’s a lot of heat. I get this all sounds super complex and confusi… Because you can’t reuse the old servers and cooling, you have to both buy a bunch of new stuff and build a bunch of new stuff. You can’t just use the old cooling either, so you need a bigger data center, meaning more construction (unless you’re just building a new one). Construction is expensive and takes time, meaning that while you may have bought all those new GPUs, you have to wait to get them all installed. When your construction is done, you now need more power than before. Once your data center is built, you have spent an incredibly large amount of money and taken a whole bunch of time. When you turn on the GPU, it immediately starts costing you money. This means that every GPU purchase starts in a deep, dark hole. While a GB200 rack containing 72 GPUs might cost $3 million, its costs balloon even further when you factor in all the stuff you had to build around it so that it A) turns on, B) functions properly and C) doesn’t overheat and turn off. It’s difficult to say how much more because things like power and construction vary from place to place, and something as simple as, say, permitting can have an outsized effect on the price of construction, but my napkin maths says there are tens of thousands of dollars of costs at minimum for each GB200 rack. So, yeah, when you finally turn a GPU on, you’re tens of thousands of dollars in the hole already. And then we get to the muddy part — how much do these fucking things cost to run? The answer is “nobody seems to want to tell us,” and also “it’s complicated.” It’s hard to tell exactly how much one GPU costs, despite many companies (like Microsoft, Google or Amazon, or AI-specialist companies like CoreWeave) selling hourly access to GPUs, because data centers are full of them, and because a GPU isn’t just a GPU, it’s also connected to a CPU, a server, storage, and many other pieces that need to exist for it to do anything.
November 24, 2025 at 5:04 PM
Reposted by Pouetchoune
edzitron.com
Every year, NVIDIA makes a new GPU that's more power hungry (requiring even more power + data centers) and even more expensive, to the point that this company can no longer rely on customers cash flow. NVIDIA is fueled by private debt that must eventually run out.
www.wheresyoured.at/the-haters-g...
NVIDIA's Future Success Is Predicated On The Ability For Everybody To Keep Feeding It Tens Of Billions Of Dollars Every Quarter, Requiring Endless Debt, Space, and Actual AI Demand To Exist Here are some thoughts I have had: A 25MW data center costs about $1 billion, with $600 million of that being GPUs — 200 GB200 racks, to be specific. It needs about 20 acres — 100,000 square feet for the data center, roughly. NVIDIA sells about $50 billion of GPUs and associated hardware in a quarter, so let's say that $40 billion of that is just the GPUs and $10 billion is everything else (primarily networking gear), so around 13,333 GB200 racks. I realize that NVIDIA sells far more than that (GB300 racks, singular GPUs, and so on). Deep-pocketed hyperscalers like Microsoft, Google, Meta and Amazon representing 41.32% of NVIDIA's revenue in the middle of 2025, funneling free cash flow directly into Jensen Huang's pockets... ...for now. Amazon ($15 billion), Google ($25 billion), Meta ($30 billion) and Oracle ($18 billion) have all had to raise massive amounts of debt to continue to fund AI-focused capital expenditures, with more than half of that (per Rubenstein) spent on GPUs. Otherwise, basically anybody buying GPUs at any scale has to fund doing so with either venture capital (money raised in exchange for part of the company) or debt. NVIDIA, at this point, is around 8% of the value of the S&P 500 (the 500 leading (meaning they meet certain criteria of size, liquidity (cash availability) and profitability) companies on the US stock market). Its continued health — and representative value as a stock, which is not necessarily based on its actual numbers or health, but in this case kind of is? — has led the stock market to remarkable gains. It is not enough for NVIDIA to simply be a profitable company. It must continue beating the last quarter's revenue, again and again and again and again, forever. If that sounds dramatic, I assure you it is the truth. NVIDIA's continued success — and its ability to continue delivering outsized beats of Wall Street's revenue estimates — depends on: The willingness of a few very large, cash-rich companies (Microsoft, Meta, Amazon and Google) to continue buying successive generations of NVIDIA GPUs forever. The ability of said companies to continue buying successive generations of GPUs forever. The ability of other, less-cash-rich companies like Oracle to continue being able to raise debt to buy massive amounts of GPUs — such as the $40 billion of GPUs that Oracle is buying for Stargate Abilene forever. This is becoming a problem. The ability of unprofitable, debt-ridden companies like CoreWeave, AI "neoclouds" that use the GPUs they purchase from NVIDIA as collateral for loans to buy more GPUs, to to continue raising that debt to buy more GPUs. The ability of anybody who buys these GPUs to actually install them and use them, which requires massive amounts of construction...and more power than is currently available, even to the most well-funded and conspicuous projects. In simple terms, its success depends on the debt markets to continue propping up its revenues, because there is not really enough free cash in the world to continue pumping it into NVIDIA at this rate. And after all of this, large language models, the only way to make any real money on any of these GPUs, must prove they can actually produce a profit. Per my article from September, I can find no compelling evidence (outside of boosters speciously claiming otherwise) that it's profitable to sell access to GPUs. Based on my calculations, there's likely little more than $61 billion of actual AI revenue in 2025 across every single AI company and hyperscaler. Note that I said "revenue." Absolutely nobody is making a profit. The NVIDIA situation is one of the most insane things I've seen in my life. The single-largest, single-most-valuable, single-most-profitable company on the stock market has got there through selling u…
November 24, 2025 at 5:04 PM