On the Peculiar Experience of Watching Someone Else's Central Bank Do Its Job

Six of seven Bank of Korea board members raised the base rate to 3% on Thursday. Second consecutive hike. First back-to-back tightening since the seven-straight-meeting slog of 2022-2023. Governor Shin Hyun-song called it "preemptive," which is a word nobody at the Eccles Building has used out loud in about eighteen months. The dissenter wanted a hold, not a cut. Let that sit for a second: the most dovish person in the room still wanted rates unchanged, not lower. Compare that to a Federal Reserve whose July meeting produced three regional presidents dissenting in favor of hiking against a chair who didn't hike, and you start to see the shape of the joke.

Here's the part that should actually bother you. Korea's hike isn't a response to some idiosyncratic domestic overheating story — a housing bubble, a wage-price spiral, a currency crisis. It's a direct read-through of the same AI capex supercycle that's simultaneously breaking Washington's ability to communicate anything coherent. The BOK upgraded its 2026 growth forecast to 3.3%, three times its own Q2 call, explicitly citing AI-driven semiconductor exports. Real GDP grew 0.6% quarter-on-quarter versus a 0.2% house forecast. Core inflation hit 2.6% in July, the highest since December 2023, precisely because Samsung and SK Hynix are printing money on HBM and the income effects are sloshing into Seoul property and services prices. The chip boom that's got Warsh paralyzed with indecision about whether inflation is "sticky" or "transitory" is the same boom that just gave a Korean central banker the clearest, least controversial hike of his tenure.

That's the actual story here, and it's not really about Korea. It's that the AI capex cycle is a single global shock landing on a dozen different monetary policy frameworks, and the frameworks are responding in completely incompatible ways because they sit at different points in the supply chain. Korea is upstream — fabs, foundries, HBM, HBM again — and upstream means income, exports, and a straightforward hawkish read. The US is simultaneously the demand source for all of that capex (hyperscaler spending, data center buildouts) and the place where a $40 trillion debt stock makes any actual rate response fiscally catastrophic. So Seoul hikes on the good news and Washington sits on its hands worrying about the bad news buried inside the same good news. Two central banks, one shock, opposite reaction functions, and nobody's framework was built to handle that kind of divergence cleanly.

Which brings us, unavoidably, to today. Warsh steps up in Jackson Hole this morning with a Treasury that's already had to double long-bond buybacks to $4 billion a week to keep the 30-year off 5.5%, a 9-3 split committee, and now a regional peer three time zones away demonstrating in real time what it looks like when a central bank actually acts on the inflation data sitting in front of it instead of talking about structural themes and payment rails. If Warsh comes out today still refusing to define a reaction function, the unflattering comparison writes itself by tomorrow morning's research notes: the AI supercycle is real enough to force South Korea's hand twice in two months, and somehow not real enough for the chair of the world's reserve-currency central bank to say a complete sentence about it.

There's a second-order trade lurking in here too, and it's not subtle. The won firmed into Thursday's decision. Capital that's spent 2026 chasing the AI-adjacent growth story — semiconductors, HBM suppliers, the whole downstream cybersecurity complex that just had Okta rip 28% and CrowdStrike gain over 20% on agentic-AI demand beats — now has a genuine yield argument to sit in Korean paper instead of parking in a US long bond that the Treasury itself is visibly trying to prop up through market intervention. That's not a market functioning normally. That's capital voting on which government it trusts to actually respond to the data, and it just voted for Seoul over Washington in the space of forty-eight hours.

None of this means the BOK has it figured out. A rate hike into a semiconductor boom that could roll over the moment Nvidia's guide disappoints anyone is its own kind of bet, and Korean household debt hasn't gone anywhere just because the board found room to tighten. But the contrast is the whole point. One central bank looked at AI-driven growth data and did the obvious, boring, credible thing with it. Another has spent four months turning the exact same category of data into an excuse for silence, forcing its own Treasury Secretary into bond-market plumbing that used to be unthinkable. Whatever gets said in Wyoming in the next few hours, it's getting graded against a scorecard Seoul already filled out this week.