Hello and welcome to the final premium edition of Where's Your Ed At for the year. Since kicking off premium, we've had some incredible bangers that I recommend you revisit (or subscribe and read in the meantime!):
- The Hater's Guide To NVIDIA, a comprehensive guide to the largest and weirdest company on the stock market, which was several weeks ahead of most on the "GPUs in warehouses" story.
- Big Tech Needs $2 Trillion In AI Revenue By 2030 or They Wasted Their Capex, a mathematical breakdown of how big tech has to make so much money before 2030 or it will have wasted every penny building AI data centers.
- Oracle and OpenAI Are Full Of Crap, where I broke down how Oracle doesn't have the capacity and OpenAI doesn't have the money to pay for their $300 billion compute deal, predicting the current state of affairs with Oracle's data centers months in advance.
- The Ways The AI Bubble Will Burst, a detailed piece about how the collapse of AI data center funding will eventually lead to the collapse of AI startup funding, creating a "chain of pain" that eventually leads to nobody buying GPUs and the end of this era.
I pride myself on providing a ton of value in these pieces, and I really hope if you're on the fence about subscribing you'll give me a look.
Last week has been a remarkably grim one for the AI industry, resplendent with some terrible news and "positive stories" that still leave investors with a vile taste in their mouth.
Let's recount:
- Disney is investing $1 billion in OpenAI in a deal where OpenAI will "bring beloved characters from Disney's brands to Sora," including a three-year licensing deal. One might think that a licensing deal is weird, given that Disney is investing, and one would be right! Apparently OpenAI is "paying" to license Disney's characters entirely in stock warrants, and Disney has the opportunity to buy an undisclosed amount of future stock.
- Amazon is in discussions to invest $10 billion in OpenAI at a valuation of over $500 billion, per The Information, and plans to use Amazon's Trainium AI server chips (its in-house competitor to NVIDIA's GPUs that some startups, per Business Insider, claim have "performance challenges" and "underperformed" NVIDIA's years-old H100 chips), apparently. Any excitement you might have over this deal should be tempered by the fact that OpenAI and Amazon Web Services signed a $38 billion deal back in November, meaning that this is likely a situation where Amazon would hand money to OpenAI, which would then hand the money right back to Amazon, and that's assuming any real money actually changes hands.
- Though this is just one source, I've heard tell that Amazon, at times, sells Trainium at a loss to get customers. Then again, I think this might be the case with all AI compute.
- Bloomberg reported that Oracle has pushed back the completion date of multiple data centers being built for OpenAI, "largely due to labor and material shortages." Oracle responded, saying that "there have been no delays to any sites required to meet our contractual commitments, and all milestones remain on track."
- It isn't clear what data centers these are, but a clue might be...
- ...that Blue Owl has pulled out of funding a $10 billion deal for a data center for Oracle/OpenAI in Michigan, per The Financial Times. This is a very, very, very bad sign. Blue Owl is arguably the loosest, friendliest lender in the data center space, and while Oracle claims another partner is allegedly talking to Blackstone, one has to wonder whether Blackstone is lining up to fund "the deal that Blue Owl couldn't handle."
- Blue Owl is the pre-eminent lender in data center financing. It backed Meta's $30 billion Hyperion data center project with $3 billion of its own capital, it sunk $3 billion into OpenAI's Stargate New Mexico deal, and an indeterminate amount in Stargate Abilene, likely somewhere between $2.5 billion and $5 billion, on top of a $7.1 billion loan provided to Blue Owl and developer Crusoe to finish the project, on top of another $5 billion joint venture with Chrisa and Powerhouse to build a data center for rickety, nasty AI compute company CoreWeave.
- So why did this deal fall apart? Well, according to the Financial Times, "lenders pushed for stricter leasing and debt terms amid shifting market sentiment around enormous AI spending including Oracle's own commitments and rising debt levels." If only somebody could have warned them, somehow.
- Though I'll get into more detail after the premium break, both Oracle and Broadcom reported earnings, and both saw their stocks get dumped like a deadbeat boyfriend with a bad attitude and credit card debt.
- In Oracle's case it was the same old story -- lots of debt, decaying margins and negative cash flow, along with a bunch of commitments.
- Did I mention that Oracle has $248 billion in upcoming data center lease commitments? More than double those made by Microsoft?
- In Broadcom's case, things were a little weirder. While it beat on estimates, it partly did so, per The Coastal Journal, by playing funny non-GAAP (generally accepted accounting practices) games with things like how it handles stock compensation and the amortizations to raise its "adjusted" earnings per share, boosting non-GAAP revenues by $4.4 billion.
- The other problem was related to OpenAI. Back in October, Broadcom and OpenAI announced a "strategic collaboration" for "10 gigawatts of customer AI accelerators," with "Broadcom to deploy racks of AI accelerator and network systems targeted to start in the second half of 2026, to complete by 2029." I'll get into the nitty gritty later, but CEO Hock Tan said that Broadcom "did not expect much [revenue]" in 2026 from the deal.
- CoreWeave's Denton Data Center has become a nightmare, with, per the Wall Street Journal, heavy rains and winds causing "a roughly 60-day delay" that prevented contractors from pouring concrete for the data center, pushing the completion date back by "several months" on top of "additional delays caused by revisions to design" for a data center specifically built to lease to OpenAI.
There are a few common threads between all of these stories:
- OpenAI doesn't have cash.
- The Disney licensing deal? Paid for in stock.
- The AWS contract? Amazon has to give OpenAI $10 billion to pay for it, because OpenAI doesn't have the cash.
- Broadcom's deal with OpenAI? "not much" revenue in 2026, probably because OpenAI doesn't have the cash.
- The Money For Data Centers Is Running Out.
- Blue Owl is the loosest lender in the universe, and if it's having trouble raising money, everybody will very soon.
- Investors are aggressively dumping Oracle because it keeps trying to build more data centers for OpenAI, a company that does not have the money to pay for its compute.
- AI Is Wearing Out Its Welcome, and the AI Bubble Narrative Is Impossible To Ignore
- It used to be (back in September, at least) that you could announce a big, stupid deal with OpenAI and see a 40% stock bump. Now the markets are suddenly thinking "huh, how is it gonna pay that?"
- Oracle's stock also got dumped because it increased capital expenditures in its latest quarter to $12 billion, on analyst expectations of $8.4 billion.
And the other key thread is the year 2026.
Next year is meant to be the year that everything changes. It was meant to be the year that OpenAI had a gigawatt of data centers built with Broadcom and AMD, and when Stargate Abilene's 8 buildings were fully built and energized. 2026 is meant to be the year that OpenAI opened Stargate UAE, too.
Here in reality, absolutely none of this is happening, and I believe that 2026 is the year when everything begins to collapse.
In today's piece, I'm going to line up the sharp objects sitting right next to an increasingly-wobbling AI bubble, and why everything hinges on a looming cash crunch for OpenAI, AI data centers, those funding AI data centers, and venture capital itself.