[Week 6 of 2026] Integration, Illiquidity, and Intelligence

Welcome back to Price and Prejudice with a few musings from Week 6 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 Vertical Integration of Everything

SpaceX is buying xAI, folding rockets, satellites, LLMs, and a social network into one giant blob. Part of the reason, at least according to Elon, is that this can power AI from space. The combined company is now valued north of a trillion dollars, which feels not too strange in 2026.

This actually maps pretty neatly into older economics about the boundary of the firm, most famously put forth by Ronald Coase. His argument was that firms exist when it's cheaper to do things internally than through markets, and that seems to be in line with Musks' complaints: regulators are annoying and contracts can be overturned in courts, so vertical integration reduces coordination risk and speeds up decision-making. And the push to go public and get into stock market indexes fast thus makes perfect sense because this whole structure needs a stable source of capital, and the stock market indexes effectively force passive funds to buy and they rarely ask follow-up questions.

When the Collateral Is the Point

I briefly mentioned asset backed finance or structured finance in class. Here's a nice primer from Guggenheim that provides you all the details you wanted to learn about this market and pitches it as reassuringly boring, which is always how finance sells itself before it gets interesting.

On a related note, this FT article talks about art-backed loans blowing up, with defaults rising and "loan-to-own" lenders (whose primary goal is to take control of the borrower's art) circling. I'm a big fan of Edward Hopper, but it never occurred to me that I wanted to own it, let alone seize it through a default. Perhaps this is what NFTs were trying to fix: art ownership without hassle or storage.

Benchmarking Finance Tasks

This document from Anthropic explains how its Claude Opus 4.6 model performs, including in finance. As with any testing environment, finance is treated less as a mysterious art and more as a clean testing environment: defined tasks, verifiable answers, professional standards. The interesting part is how disruptors think about finance here. The model is evaluated on things like digging through SEC filings, building spreadsheets, and producing slides and memos, which is (correctly) the stuff junior analysts actually do all day.

You could also perhaps argue that a lot of people in finance are not hired because they score well on benchmarks, but they’re hired because they’re interesting to sit next to at 11 p.m., conditional on knowing enough finance not to break anything. Models are getting good at the minimum competence part, but humans still dominate the “vibes under pressure” category (for now). Finance workplace runs on both systems and personalities, and only one of those fits neatly into a PDF.