Looking past all the RED

Today, the AI trade is going through some immense red light therapy. $CRWV, $NBIS, and $IREN, along with some of the hyperscalers, are getting murdered in the market. Alphabet dropped 12.6% on October 29, Micron dropped about 5%, and SK hynix has been caught in it too. Twitter doomsayers are predicting the end of the AI trade and the end of the world. Just because portfolios are red does not mean the trade is over.

But there are reasons to be worried. Google is looking at $195–205 billion in capex this year and says it’s going to spend significantly more next year. It just reported negative $9.1 billion in quarterly free cash flow. That is an insane amount of money going out, and Wall Street wants to know when it starts coming back. Google Cloud revenue more than doubled, so there is actual business coming from this. The question is how much cash is left after paying for all of it.

The neoclouds have their own problem: debt. Reuters put CoreWeave’s debt at roughly $50 billion on October 15. In the first half of this year, it brought in $3.66 billion in operating cash flow and spent $14.12 billion on property and equipment. That’s a $10.45 billion gap before you start talking about how amazing the future is going to be. Its second-quarter net interest expense alone was $640 million. The debt bill keeps coming regardless of how excited everyone is about AI.

Now, this doesn’t mean every neocloud is in the same position. At June 30, Nebius had about $8.55 billion in debt and $8.04 billion in cash and equivalents. IREN had about $7.59 billion in debt, $5.90 billion in cash and equivalents, and another $1.72 billion in restricted cash. That last part isn’t money it can freely spend on anything. Nebius also announced a new $3.5 billion Citi facility on October 22, although it matures in 364 days. So lenders haven’t completely shut the door, but you have to look at who still needs funding and how often they need to go back for more.

Also, let’s remember how much some of these names have already run. In 2025, Micron was up about 239%, Nebius 202%, and IREN 285%. CoreWeave finished the year about 79% above its IPO price. Through October 29 this year, Micron and Nebius were still up roughly 227% and 105%, respectively.

Mr. Market is allowed to start asking harder questions. I wouldn’t write this off as just a small correction before the midterms either. Historically, midterm years have had average S&P 500 drawdowns of about 18%, according to J.P. Morgan’s review of data from 1945 through 2025. That doesn’t mean the market always falls before an election, or that it has to bounce straight after this one.

The way I see it, the debt concerns and the capex concerns feed into each other. Hyperscalers are buying capacity from neoclouds, and neoclouds are spending huge amounts before that capacity starts earning revenue. If delivery gets delayed, the revenue gets delayed too, while the interest bill still needs to be paid. But I wouldn’t say the hyperscalers are all running out of cash. Microsoft just reported $18.6 billion in quarterly free cash flow and 48% Azure revenue growth in it’s latest earnings. Its capex was below expectations partly because third-party capacity deliveries were delayed. That is a different problem from there being no demand.

Now coming to the component manufacturers, like SK hynix, Micron, TSMC, and Sandisk. The concern is pretty straightforward: if the people buying the equipment slow down, the people selling it feel it too. Micron’s Cloud Memory and Core Data Center businesses made up about 63% of its latest quarterly revenue. TSMC’s high-performance computing business was 68% of revenue, although that includes more than AI. Sandisk’s datacenter business was about 43.5%. So yes, there is a lot of exposure here, but it isn’t accurate to say all these companies get most of their sales directly from hyperscalers and neoclouds.

This is still the part of the AI trade I’m more interested in. Micron ended its fiscal year on September 3 with about $73.5 billion in cash, investments, and restricted cash against roughly $5.2 billion in debt. It also reported $62.3 billion in adjusted free cash flow for the year. At September 30, SK hynix had ₩98.7 trillion in net cash. TSMC had about NT$4.02 trillion in cash and marketable securities and generated NT$476.6 billion in quarterly free cash flow after capex. Micron and TSMC are spending huge amounts themselves, but they’re also bringing in cash after doing it.

That is why I think the component manufacturers deserve more attention when people talk about buying the dip. They still depend on the AI buildout, but the balance sheets give them room if it takes longer than everyone expected. SK hynix also says long-term agreements cover more than half its production capacity for 2027 and 2028. That gives it some visibility beyond whatever people are saying on Twitter today.

And even after all this red light therapy, a stock being down doesn’t automatically make it cheap. Microsoft was still trading at roughly 28 times trailing earnings on October 29. CoreWeave’s valuation depends heavily on the cash people expect it to generate in the future. The same applies to the chipmakers: if you’re valuing them on record profits, you have to think about what happens when memory prices or demand cool off. TSMC has Taiwan and geopolitical risks too. A clean balance sheet doesn’t remove those problems.

If I’m buying the dip here, I’m looking at TSMC, Micron, and SK hynix first. I think they’re a better place to start than automatically buying every beaten-down neocloud. I wouldn’t put every hyperscaler in the same bucket either. The price still matters, but I’d rather own the suppliers with the cash to sit through a messier AI buildout.