[Week 22 of 2026] Banks, Bets, and Brevity

Welcome back to Price and Prejudice with a few musings from Week 22 of 2026.

Free Banking, Stablecoin Edition

This WSJ piece by Greg Ip provides what I think is a pretty good summary of the standard academic concern. The argument is that stablecoins are essentially a return to the free banking era from 1837 to 1863, when banks issued their own private notes and the system was fragmented. The core issue is what people call "singleness" – a dollar must always equal a dollar regardless of when, where, or with whom it is used. Bank deposits achieve this because banks can borrow from the Fed to honor redemptions and because dollars move between banks via the Fed's settlement rails. Stablecoins, despite being marketed as dollar-pegged, drift around a dollar and move through proprietary, fragmented infrastructures. They import the credibility of public money without inheriting the plumbing that makes singleness work.

One angle the article doesn't push is what stablecoins do to the tax base. The 84% figure on illicit crypto activity gets a passing mention as a money laundering issue, but the more interesting version of this concern is what happens once synthetic equity-like products start riding stablecoin rails at scale. As an example, a tokenized S&P 500 product traded on an offshore platform and settled in stablecoin looks economically identical to SPY, but lives entirely outside the standard regulation around capital gains taxation e.g. wash sale rules, the qualified dividend rules, and the cost basis reporting infrastructure. The equity tax system was designed assuming intermediaries report transactions to the IRS, and the smart money is probably the cohort with the most to gain from migrating to rails where those intermediaries don't exist.

Korea's Triple Long

It would be remiss of me as a Korean finance professor not to cover how South Korea's KOSPI has crossed 8,000 for the first time, up 207% year-over-year, with most of the move driven by SK Hynix and Samsung. Margin loans are at a record high, and middle-aged and older Koreans are now borrowing to buy in. I also had to talk my own mom out of adding to her Samsung position last week, so I can vouch for the FOMO from a sample of one.

The reason this matters more than the average retail-mania story is that Korea's poverty rate for older adults is already 40%, and the marginal margin borrower here is putting retirement money on the line. In fact, the Korean households are now exposed to the same AI cycle through three correlated channels: (1) their jobs and the broader export economy (which runs through chipmakers and their suppliers), (2) the Korean won and the capital account (which is increasingly driven by foreign inflows into the same AI plays), and (3) leveraged equity position. This is the textbook violation of diversification, concentrating idiosyncratic risk in the asset you cannot diversify because you also work in the same sector.

It is indeed a very fun time to trade Korean stocks, since Korean equities have historically traded at a "Korea discount" tied to chaebol governance, geopolitical risk, and dual-class structures. I think the bigger question is whether the structural reasons for the discount actually went away, or whether the AI cycle is just papering over them. My guess is that the structural issues haven't gone anywhere, which means the discount probably comes back when the cycle dies out (or perhaps the cycle never ends).

Paying by the Word

This WSJ piece reports that corporate America is starting to ration AI as token costs explode. Some companies have hit their annual AI budgets in three months, while others are seeing bills double or triple. The "tokenmaxxing" culture (employees burning premium-tier compute on small talk and trivial queries) is starting to get policed, and many large tech companies are all reining in usage, building cheaper internal tools, or tracking token-to-outcome ratios. The most striking statistic in the piece is that only 18% of corporate spending on AI coding tokens translates into shipped product.

If you think about it, the "token" is a strange unit of account compared to the ones that preceded it. Software was historically priced in seats, a fixed cost per user with no marginal incentive to economize on intensity. Cloud compute introduced pay-as-you-go in time units like compute-hours, which were at least intuitive to budget. APIs charge per call, and each call is roughly atomic. Tokens are different because they price language proportionally: the bill scales with how much you say to the model and how much the model says back. So verbosity is bad, and so is politeness. Careful exposition also costs more.

You can imagine in the future where this ends up. The natural successor to McKinsey's headcount-optimization playbook is a token-optimization playbook i.e. "drop articles," "do not say hello to LLMs," "use abbreviations," "single-sentence prompts only." And regularly there will also be a debate on whether token-shaving even matters. If it's really true that only 18% of corporate token spend ships product, the gain from cutting pleasantries is rounding error on the gain.