the Week the Market Ate Itself
Notes from the Week the Market Ate Itself
February 22, 2026
Somewhere in a glass-walled office, a portfolio manager is staring at a silver position that was worth $122 an ounce three weeks ago and is now gasping at $83, trying to explain to a risk committee that no, this wasn't leverage, this was a conviction trade, this was the debasement thesis, this was the multi-year case for hard assets in a world of debauched fiat currency, and yes he's aware of what happened in 1980 with the Hunt Brothers but that's completely different—
Except it isn't. It never is.
Here's what February 2026 actually was, stripped of the narrative confetti: a $15 trillion collective bet on the slow death of the dollar got rear-ended by a single personnel announcement. One name — Kevin Warsh — dropped on a Friday afternoon, and the gold-to-silver-to-Bitcoin-to-Korean-tech daisy chain that had been the dominant trade of the past eighteen months unwound with the kind of velocity that makes you reconsider whether financial markets are sophisticated price-discovery mechanisms or just very expensive games of hot potato.
Gold was above $5,500 an ounce in late January. Let that sit for a second. Five thousand, five hundred dollars. Per ounce. The entire edifice was built on one structural bet: that the Fed would be kept on a short leash, that rates would keep falling regardless of what the inflation data said, that whoever Trump installed at the Eccles Building would be a monetary yes-man who'd print his way through any inconvenience. That was the trade. That was the entire trade.
Warsh killed it with a résumé.
The mechanics are worth understanding because they reveal something embarrassing about how markets actually work. The "debasement trade" — anti-fiat assets, hard money proxies, long gold, long silver, long anything that isn't a dollar — had become so crowded by January that it had effectively decoupled from any rational valuation anchor. This is what Paul Donovan at UBS called, with characteristic British understatement, a "fear of missing out" trade becoming exhausted. What he meant was: the marginal buyer had run out. Everyone who was going to buy gold at $4,000 had bought it at $4,000, and then at $4,500, and then at $5,000, and then the positions were stacked with enough leverage that all you needed was a breeze in the wrong direction.
Warsh was a gale-force wind.
Spot silver dropped 31.4% in a single day — its worst session since March 1980. Gold fell nearly 9%. Bitcoin, which had been quietly playing along with the anti-fiat theme, tumbled from above $83,000 to a low of $74,570 and hit its lowest level since April. The Korean Kospi, which had surged 76% in 2025 on AI chip euphoria and had somehow become a satellite asset in the global momentum trade, sank 5.26% — its worst day since April.
Everything connected. Everything fell together. Which tells you they were all one trade wearing different clothes.
Now, here's where it gets philosophically interesting, because the Warsh story has a twist in it that markets have been quietly chewing on ever since.
Warsh is actually expected to push for more rate cuts this year than the 50 basis points currently priced by markets, based on his recent remarks. He's shifted dovish on short-term rates. He's talked about AI-driven productivity suppressing long-term inflation. He has, in his recent incarnation, been closer to Trump's rate-cut instincts than to the austere hawk the precious metals market apparently decided he was.
So why did gold crater? Because the markets weren't trading Warsh's actual policy views. They were trading the signal: an institutionalist at the Fed, someone with standing and credibility, someone who wouldn't torch Fed independence to get Trump his rate cuts by Thursday. The debasement trade wasn't about rates, not really. It was about the perception of institutional decay. Warsh — paradoxically, whatever his actual policy bent — looked like an adult in the room. And the adult-in-the-room trade is very bad for gold.
As Thierry Wizman at Macquarie put it: "Warsh is not the Fed's guy, he is Trump's guy, and has shadowed Trump on monetary policy almost every step of the way since 2009." Make of that what you will. The market made of it a 9% single-day collapse in the asset the whole world had been treating as civilization insurance.
Meanwhile, the software sector has been quietly having its own crisis, and nobody is paying enough attention because the headlines are all precious metals and Fed chairs.
The iShares Software ETF is down 20% year-to-date. ServiceNow, Salesforce — both down close to 7% in a single session earlier this month. The private credit crowd has been feeling it too: Blue Owl, TPG, Ares, and KKR all dropped by double-digit percentages on a single Tuesday, as fears spread about exposure to software-adjacent lending. Apollo was off 7%. BlackRock shed 5%.
The mechanism here is AI eating its own ecosystem. The same AI buildout that has made Nvidia a $3 trillion company is now generating tools capable of replacing the software that enterprise companies buy from Salesforce, ServiceNow, Workday, and the rest. Companies are using Claude Code, Cursor, and a dozen other agent-based development tools to build their own software instead of buying it. The SaaS model — recurring revenue, sticky customers, high margins — was premised on software being hard to build. It's getting less hard by the month.
The Magnificent 7 is expected to generate almost $1 trillion in operating cash flow this year while spending $700 billion of it on AI capex — meaning their collective free cash flow will be down roughly 25% from 2025. Investors spent a year cheering the capex numbers. They're now asking what the return on it looks like, and the answer, so far, is: unclear, but it definitely involves killing someone else's business model first.
Which brings us to Wednesday. Nvidia reports Q4 fiscal 2026 after the close, and if this week has a spine, that's it.
Wall Street is expecting roughly $65.5 billion in revenue, with EPS of around $1.52. Polymarket has traders pricing a 94.5% probability that Nvidia beats the consensus. The hyperscalers have been publicly and expensively committed: Amazon guides $200 billion in capex for 2026, Alphabet $175–185 billion, Meta $115–135 billion. The GPU demand has to go somewhere. It's been going to Nvidia.
Nvidia's stock has rallied 143% from its April 2025 low of $86.62 to a peak of $212.19 in late October, before sliding into a correction that has now persisted for three months. The stock is essentially flat for 2026. In any normal world, a company guiding $65 billion in quarterly revenue and growing 60%+ year-over-year doesn't trade flat for three months. But this isn't a normal world. This is a world where the market has started asking whether the Blackwell supercycle is a durable infrastructure buildout or a very expensive arms race with a murky ROI.
The key question isn't whether Nvidia beats estimates. It will beat estimates. It has beaten the analyst consensus in 20 of the past 22 quarters. The question is whether the guidance for Q1 FY2027 and the tone around China sales — which have been essentially zero since export restrictions bit, with H20 chip revenue coming in around just $50 million last quarter after large purchase orders evaporated — suggests an acceleration or a plateau.
If the Blackwell cycle is real and the Rubin architecture transition looks smooth, tech has a reason to rally. If guidance is softer than the hyperscaler capex promises imply, you'll spend Thursday morning watching Nasdaq futures try to find a floor.
Everything this week points toward Wednesday evening. The precious metals saga, the software collapse, the Fed transition, the tariff litigation loop — all of it is noise until Jensen Huang's face appears on a screen and he either says "demand is insane" or he doesn't.
One more thing, because it would be irresponsible to ignore it: Japan.
Thirty- and forty-year Japanese government bond yields hit multi-decade highs above 4%, with the 30-year yield climbing more than 25 basis points in a single day — around eight times the average daily move of the previous five years. The Bank of Japan is tapering bond purchases. The yen carry trade — which for decades has quietly funded global risk appetite by allowing investors to borrow cheaply in yen and deploy that capital everywhere else — is becoming more expensive by the week. Japanese institutions have signaled plans to repatriate capital and rebuild domestic JGB holdings.
Nobody is talking about this enough. The carry trade unwind of August 2024 sent a 10% shock through global equities in 72 hours. The current moves in JGB yields are slower but structurally more significant, because they're driven by Japan's actual fiscal trajectory — government debt above 230% of GDP, the highest among G7 economies — rather than just positioning.
The bill is coming for a forty-year experiment in infinite debt tolerance. The question is whether it arrives as a controlled normalization or a disorderly repricing.
Markets are not priced for the latter. Markets are never priced for the latter.
Nothing here is investment advice. All views are the author's own, offered freely and revised without apology when the data demands it.