[Week 8 of 2026] Memory, Megabonds, Minutes

Welcome back to Price and Prejudice with a few musings from Week 8 of 2026. Now that the Capital Markets & Investments class at Columbia is in progress, the blog will also reference to some contents from class as well.

The Memory Nobody Wanted

In textbook capital budgeting, we often assume that the return to investment is tangible and can be easily calculated. You estimate cash flows, pick a discount rate, and the NPV either clears the hurdle or it doesn't. But in practice, this is messier. This FT article discusses the rise of SK Hynix, which spent over a decade increasing R&D by 14 percent a year on a chip that had no obvious market.

The first real use case for their core product – high-bandwidth memory chips – was a prohibitively expensive gaming card, which was not a viable market. Then came the AI boom, and suddenly Hynix was the only company with the manufacturing knowhow, the supplier relationships, and the accumulated process knowledge to meet Nvidia's demand at scale. The market cap is up 340 percent in a year, and this is exactly what a "real option" looks like when it lands in the money.

From an investor's perspective, the harder question is whether you can spot these options before they pay off. With AI, the key is identifying which companies can play an important role in the supply chain. There are recent examples from Japan: Toto, a bidet company, which makes electrostatic chucks used in NAND chip production, and Ajinomoto, which produces MSG, producing an insulating film used in advanced semiconductor packaging. This is a useful reminder that the AI buildout has a very long tail, and the winners won't all have the word "semiconductor" in their name.

Borrowing Like a Utility

We briefly discussed the issuance of century bonds by Google, and this article shows that bond issuance is generally high for all "hyperscalers." In just the first two months of 2026, the five largest hyperscalers issued $45 billion in bonds in the US alone, nearly half their total for all of 2025. And bond investors are absorbing all of it without much protest: spreads on Alphabet's debt are less than one percentage point over Treasuries, pricing it roughly like sovereign debt.

The deeper story is what this says about what these companies are becoming. Companies like Google and Meta spent two decades being valued as asset-light, high-margin software businesses, but now their capital expenditure reaches half of their revenue. If the AI buildout pays off, the debt looks cheap in retrospect. If it doesn't, the market will eventually have to decide whether to value these companies like the software businesses they were or the infrastructure companies they're becoming. And those are very different multiples.

The Fed Reads the Room, Then the Room Changes

The minutes from the Fed's January meeting dropped Wednesday after class, and the WSJ had a full breakdown published roughly 45 minutes after release. This turnaround time tells you something about how closely these documents are watched. The headline was pretty straightforward: most officials want to see more inflation progress before cutting again, and the bar for the next cut is higher than it was in the fall. One interesting detail that WSJ highlights is that several officials saying they would have supported "two-sided" language in the post-meeting statement, explicitly flagging the possibility of rate increases if inflation stays above target.

How informative those minutes are for what actually happens next depends not only on what the officials said, but on what has occurred since the meeting. And this week was eventful, since the Supreme Court struck down Trump's IEEPA tariffs 6-3 on Friday, ruling that the 1977 emergency powers law never actually authorized the president to impose them. Trump responded within hours by announcing a 10% global tariff under a different statute, the Trade Act of 1974, which carries a 150-day ceiling without congressional extension. The practical upshot of this is that one large source of inflationary pressure has been replaced by a smaller and time-limited one, and the Fed officials who were most worried about tariff pass-through now have a somewhat different picture to work with.

One issue, from a rate watcher's perspective, is that nobody is sure what this means for the path of rates. The tariff ruling reduces one tail risk but doesn't eliminate uncertainty. What the ruling does change is the policy environment the Fed is navigating: two weeks ago, the two-sided language looked hawkish, but today it looks more like a committee that was prudently hedging against a scenario that just got a little less likely.

Podcasts

• This is an interesting podcast conversation on how industry participants think about the growing role of alternatives. Turns out that it doesn't require that much of a different framework.