The stock market had a wild swing into the red on Thursday as the S&P 500 lost almost 4.4% of its value, and we’re back to levels not seen since… checks notes… July 2026.
Chip stocks had it far worse. The Philadelphia Semiconductor Index, which tracks the big chip companies, fell more than 12%. That’s its worst day since March 2020.
I thought this was an interesting one because of what investors are worried about. The bottleneck for compute (the computing power AI runs on) has shifted from energy, to chips, to capital… and now to politics.
A big part of the US midterm election promises was built on the anti-data center wave. And we can probably see why. Data centers are power hungry, so if one is being built in your backyard, there are concerns about energy shortages or prices rising. And it doesn’t help that the CEOs of the leading frontier labs have been a downright disaster when it comes to communications, especially the part about how AI is going to not just replace humans (i.e. the voters), perhaps it may even destroy humanity.
And part of the political promises being made right now is that governors are going to pause new data center permits until new rules are written. Texas is the clearest example. Its Republican governor froze permits in September, and the Democrat who just beat him campaigned on a wider freeze.
Fewer permits means fewer new data centers, and fewer new data centers means fewer chips get ordered, which is how an election ended up hitting chip stocks.
It was already tough enough for neoclouds to get access to capital. Neoclouds are companies like CoreWeave that borrow heavily to buy chips and then rent out the computing power. And all the AI players are essentially fighting for the same pool of capital, whether you’re a behemoth (like Microsoft, Amazon, Alphabet or Meta), a frontier lab (Anthropic or OpenAI), or a neocloud.
And in recent months, with Alphabet lining up almost $85 billion from selling new shares in June, Meta almost $98 billion in October, and Anthropic in the process of IPO-ing, there are a lot of hands reaching into the same pool of capital. But I don’t think the pool has run dry.
A week after that share sale, Meta came back to borrow $15 billion and drew $58 billion of orders. Microsoft did the same a few days earlier, with $115 billion of orders for a $25 billion bond sale. The money is there. It’s just become a lot pickier about who it goes to, and the first thing it looks at now is credit risk.
OpenAI, on the other hand, reportedly tried to raise $30 billion, had only about $11 billion committed, and cut its target to $20 billion.
You can see the pickiness most clearly in the bond market. CoreWeave, the biggest of the neoclouds, borrowed at around 9.75% interest. In early November its bonds were trading at about 77 cents on the dollar (in other words, a lender who paid $100 could only sell for $77), which means the market now wants roughly 16% to lend to CoreWeave.
Now, it’s not just capital that’s the constraint, but also permits.
I’ve mentioned my concerns about neoclouds on Chuang’s Do More podcast before, especially how stretched their balance sheets are. Players like this do well in a supply shortage, where prices for compute are high and customers are willing to pay up. But in the event of a road bump, they’ll be in a very fragile position. A permit freeze could perhaps be one of those road bumps. Neoclouds grow by building new data centers, which means new sites and new permits, while the hyperscalers already have plenty of sites they can repurpose or fill with newer chips.
And I think this will benefit the players who already have data centers running and a strong balance sheet. If new capacity gets harder to build, the capacity that already exists gets more valuable.
And that’s the part I’d take away as an investor. Being able to wait matters just as much for us. If I’m using borrowed money, a 20% or 30% drop can force me to sell at the worst possible time, and then I never get to see the recovery. To enjoy the long-term compounding, we first have to survive the short term, and that’s the first thing our game plan has to take care of.
Weeks like this are when a game plan earns its keep. If you don’t have one yet, I wrote The Lunch Break Investor to help you build one you can actually stick to.
Compound wisely,
Thomas