The AI market works on bubble economics. It’s about stories and magic! Making financial sense, that’s for the old people now! Get with the program, granddad! But magic doesn’t happen. And it never happens with money. We know you all just want this thing to crash. Every time the Nasdaq hiccups, will…
Annons
Annons
The AI market works on bubble economics. It’s about stories and magic! Making financial sense, that’s for the old people now! Get with the program, granddad! But magic doesn’t happen. And it never happens with money. We know you all just want this thing to crash. Every time the Nasdaq hiccups, will this pass in a few days? Or … is this it? After the Democrats swept the US elections on Tuesday, the market went up a bit on Wednesday. But on Thursday stocks crashed hard, with big losses in the S&P, Dow, and Nasdaq. Large holders were dumping their tech stocks specifically — anything that touched AI. [Bloomberg, archive] There was some recovery this morning, after a good jobs report — and some traders deciding this wasn’t the big one and they could buy the dip. Cross fingers. The finance papers blame the crash on the threat that the incoming Democrats might regulate AI in any manner whatsoever and pull back on data center builds. Oh no! [FT, archive] One US portfolio manager called the Democrat win “a horrible day for America.” He happens to hold a lot of Anthropic. [FT, archive] But the political pushback was coming. Data centers were the issue in the election. The AI and crypto lobby groups — who are largely the same actual guys — blew almost all the $250 million they poured in. [Molly White] Banks and financiers don’t think data center loans are such a great bet anymore. CoreWeave bonds have now topped 16% yield. CoreWeave raised debt earlier in 2026 at 9%. You can’t build new data centers on 16% interest. [CNBC] Goldman Sachs and Apollo led SpaceX’s recent $40 billion bond offer to send Nvidia chips to space — and now they’re frantically trying to sell the debt off earlier than planned. This deal was written so SpaceX could not claim force majeure to pause payments — the way Oracle did in September. [Bloomberg] Rumors were circulating on Thursday that someone was dumping $500 million of data center loans. That turned out to be Dutch bank ING, which has put the loans up for auction — including a $25 million slice of OpenAI’s Stargate site in New Mexico, plus debt tied to big operators Vantage and CyrusOne. [Bloomberg, archive] The Bank of England is looking into loans for UK data center projects that don’t have an actual tenant lined up. These are the hot prospects for defaulting. As Kat Jones of APRS told us in October, UK hyperscale data center plans are mostly not coming from tech companies, but from local real estate speculators who don’t have a tenant lined up. [Bloomberg, archive] The European Central Bank is thinking about reviewing banks’ loans for data centers. [Bloomberg,] At the Financial Times Future of AI Summit, the AI vendors were shocked that nobody likes them. They don’t really understand the objections — one CEO compared data center protestors to protests against masks for COVID. [FT, archive] The market crash makes the big IPOs look less likely. Greg Brockman of OpenAI is already talking up 2027 for their IPO. Anthropic’s investors are suggesting the company reduce their IPO valuation to just (just!) $1 trillion — when it was originally slated to be $2 trillion, then $1.5 trillion. Anthropic still wants to get the IPO out the door this month. Good luck! [Reuters; WSJ] US and world markets are still hugely dependent on AI stocks and debt. Pension funds and hedge funds are highly exposed — because they only looked at the number going up, and they were careful never to ask tricky questions like: was there any substance to the number, or if was it just a hot air balloon? But if AI crashes, everything crashes. [Bloomberg, archive] FT Alphaville asks the obvious question about data centers: [Substack] Shouldn’t efficient markets have already priced in these states’ critical shortage of electricity capacity, along with the somewhat sketchy evidence that data farms were ever being built in the numbers promised? But that makes the mistake of assuming any of these people think further ahead than lunchtime. Video — Podcast