I keep trying to think of a cool or interesting introduction to this newsletter, and keep coming back to how fucking weird everything is getting.
Two days ago, cloud stalwart Oracle crapped its pants in public, missing on analyst revenue estimates and revealing it spent (to quote Matt Zeitlin of Heatmap News) more than $4 billion more in that quarter than analysts expected on capital expenditures, for a total of $12 billion.
The "good" news? Oracle has remaining performance obligations (RPOs) of $523 billion. For those that aren’t fluent in financese, this is future contracted revenue that hasn’t been paid for, or even delivered:
"Remaining Performance Obligations (RPO) increased by $68 billion in Q2—up 15% sequentially to $523 billion—highlighted by new commitments from Meta, NVIDIA, and others," said Oracle Principal Financial Officer, Doug Kehring.
So we've got — per Kakashii on Twitter — $68 billion of new compute deals signed in the quarter, with $20 billion from Meta (announced in October), and a few other mystery clients that could include the ByteDance/TikTok deal.
But wait. Hold the fort — what was that? NVIDIA?
NVIDIA? The accelerated computing company? The largest company on the stock market? That NVIDIA? Why is NVIDIA buying cloud compute? The Information reported back in September that NVIDIA was "stepping back from its nascent cloud computing business," intending to use it "for its own researchers."
Well, I sure hope those researchers need compute! NVIDIA has, according to its November 10-Q, agreed to $26 billion in cloud compute deals, spending $6 billion in a year each in Fiscal Years 2027 and 2028, $5 billion in FY2029, $4 billion in 2030, and $4 billion in 2031.
AI boosters damn near ripped their jorts jumping for joy at the sight of this burst of new performance obligations, yet it seems that the reason that NVIDIA CEO Jensen Huang said back in October that AI compute demand had gone up "substantially" in the last six months was because NVIDIA had stepped in to increase it. It signed a deal to buy $6.3 billion of unused capacity from CoreWeave, another to buy $1.5 billion from Lambda, and now apparently needs to buy even more compute from Oracle, despite Huang saying in November that cloud GPUs are "sold out", which traditionally means you "can't rent them."
We are in the dynasty of bullshit, a deceptive epoch where analysts and journalists who are ostensibly burdened with telling the truth feel the need to continue pushing the Gospel According To Jensen. When all of this collapses there must be a reckoning with how little effort was made to truly investigate the things that executives are saying on the television, in press releases, in earnings filings and even on social media, all because the market consensus demanded that The Number Must Continue Going Up.
The AI era is one of mythology, where billions in GPUs are bought to create supply for imaginary demand, where software is sold based on things it cannot reliably do, where companies that burn billions of dollars are rewarded with glitzy headlines and not an ounce of cynicism, and where those that have pushed back against it have been treated with more skepticism and ire than those who would benefit the most from the propagation of propaganda and outright lies.
So today I'm giving you Mythbusters — AI Edition. This is the spiritual successor to How To Argue With An AI Booster, where I address the technical, financial and philosophical myths that underpin the endless sales of GPUs and ever-increasing valuation of OpenAI.
This is going to be fun, because I truly believe that both the financial and tech press take this all a little too seriously, in the sense that everything is so dull. With a handful of exceptions (The Register being the best example), most publications treat financial reporting as something that must be inherently separate from any kind of analysis or criticism. And so, that’s why, if a publication calls bullshit on something insane, that call is almost always segmented away in its own little piece.
If you asked me why I thought this is the case, I’d say it’s probably because (excluding those cases of genuine malfeasance and fraud, like Enron and Worldcom and Nortel) we haven’t seen anything as egregiously offensive or dishonest as what’s emerged from the AI bubble. And so, reporters are accustomed to a lack of civility that, frankly, isn’t warranted.
I also think the total lack of levity or self-awareness leads to less-effective analysis, too.
For example, lots of people are freaking out about Disney investing $1 billion for an equity stake in OpenAI, all while licensing its characters to be used in Sora, and I really think you can simmer the deal down to two points:
- Wow, $1 billion? That's going to pay for a whole month of OpenAI's inference!
- Regardless of how many guardrails OpenAI puts on Sora (which is currently 21 on free apps on the App Store), there is nothing that will stop degenerates from making a video of Goofy flying a plane into a building, or Donald Duck recreating Frank's entrance from Blue Velvet, or Darth Vader saying every slur imaginable, which already happened when Disney launched a generative AI Vader in Fortnite.
Oh, and while I'm here, let's talk about TIME naming the "Architects of AI" its person (people) of the year. Who fuckin' cares! Marc Benioff, one of the biggest AI boosters in the world, owns TIME, and has already run no less than three other pieces of booster propaganda, including everything from "researchers finding that AIs can scheme, deceive or blackmail," to the supposed existence of an "AI arms race" to "coding tools like Cursor and Claude code becoming so powerful that engineers across top AI companies are using them for virtually every aspect of their work."
Are any of these points true? No! But that doesn't stop them being printed! Number must go up! AI bubble must inflate! No fact check! No investigation! Just print! Print AI Now! Make AI Go Big Now! Jensen Sell GPU! Ahhhhhhhhhhh!
Okay, alright, let's go into it. Let's bust some myths.
That sounded better in my head.