A monologue, approximately 6:47am, somewhere with bad coffee
Okay. Let's try to hold this in our heads simultaneously.
Nvidia just printed $81.6 billion in revenue. Eighty-one-point-six. Up 85% year-over-year. Data center at $75.2 billion, nearly double what it was twelve months ago. Q2 guide: $91 billion, which is to say another $10 billion higher than a quarter most companies would tattoo on their chest and retire. The board authorized $80 billion in buybacks. They raised the quarterly dividend from one cent to twenty-five cents — a 25x increase, not a typo. Jensen Huang said, with no particular sense of irony, that the AI factory buildout is "the largest infrastructure expansion in human history."
And the stock slid 1.5% in after-hours.
Sit with that. The largest company on earth by market cap delivered the most profitable quarter any semiconductor firm has ever printed, guided $4 billion above consensus for the next quarter, confirmed that $725 billion in hyperscaler capex is flowing directly through its order book — and the market shrugged, shuffled its feet, and drifted lower.
Steve Sosnick at Interactive Brokers called it perfectly: how much good news is already baked in. There is no upside surprise left in the concept of Nvidia beating. Beating is the base case. Beating is table stakes. The market is asking what's next after the beat, and the honest answer is: probably another beat, which isn't enough anymore.
This is what happens when a stock climbs 21% year-to-date before reporting. You've already been paid for the news.
So fine. Grind that out of your system. The NVDA reaction is comprehensible, if annoying.
What is not comprehensible — what I keep turning over, getting nowhere — is everything surrounding the Nvidia number like a burning ring.
Outside the earnings call, at approximately the same moment Jensen Huang was telling analysts that "agentic AI has arrived," the 30-year Treasury was sitting at 5.20%. The highest print since 2007. Oil is at $99 on WTI, having been at $108 last week when the UAE said it intercepted Iranian missiles for the first time since the ceasefire. That ceasefire is doing quite a lot of work as a word given that it apparently involves intercepting missiles. Hyperscalers are planning to spend $725 billion on AI this year — up 77% from 2024 — and simultaneously the bond market is repricing every long-duration asset on the planet because energy costs are reigniting inflation that wasn't dead to begin with.
April CPI: 3.8%. The Fed's target: 2%. The gap: intractable.
So here is the contradiction I cannot solve over bad coffee: the AI capex supercycle is real. The numbers are not fake. $75.2 billion in data center revenue is not accounting fog. Microsoft, Meta, Google, Amazon — these companies are not stupid, and they are collectively committing sums that make the interstate highway system look like a patio renovation project. But they are borrowing against a yield curve that is repricing their cost of capital upward every week. Their equity valuations, to the extent they rest on discounted future cash flows, are being mechanically compressed by every basis point the 30-year moves.
Nvidia trades at roughly 35x forward earnings. Before you reach for the "but growth justifies the multiple" argument, ask yourself what that multiple assumes about the discount rate. It assumes something that looks like the last ten years — which is not what 5.20% on the 30-year represents.
And yet here's the other half: if the AI capex cycle is genuinely structural — if Jensen is right that this isn't a bubble but a once-in-a-generation rewiring of global compute — then the investment can't stop. The hyperscalers can't pull back. They're in a race where slowing down is losing. Alphabet can't tell Meta it's going to pause Blackwell orders because the long end of the curve is uncomfortable. The capex is committed. The supply chain is locked. The revenue is accruing to Nvidia quarter after quarter with the steadiness of a utility.
Which makes Nvidia something strange and new: a growth company that behaves, structurally, like critical infrastructure. Tollbooth, not cyclical. The same yield-curve logic that crushes speculative tech should, in theory, treat Nvidia differently. Except the market is not making that distinction cleanly, hence the 1.5% dip on a report that would have sent the stock up 15% in 2023.
Meanwhile. Walmart reports before the open this morning. Analysts want $0.65 EPS on $172.5 billion in revenue. The number matters less than the language — what Walmart's management says about consumer behavior under $3.50 gasoline and a mortgage market that hasn't been this punishing since before the financial crisis. Walmart is the most honest macro instrument in American commerce. The quarterly call is, functionally, a field report from the 60% of the country that does not own meaningful equity. If Doug McMillon describes any softness in discretionary categories, any trade-down in basket composition, any deceleration in store traffic — that data point will matter more for the real economy than another Nvidia beat.
The flash PMIs dropped this morning too. S&P Global's composite figures are landing into a market that is already pricing in stagflation as a non-trivial scenario. Manufacturing has been printing above 50 partly because companies have been building precautionary inventory ahead of expected supply disruptions — which means the headline number has been flattering a demand picture that is actually decelerating. When that inventory cycle unwinds, the real PMI will arrive. Services are holding, but the Atlanta Fed's business inflation expectations ticked up to 2.4% this week. Quietly, incrementally, everything that was supposed to come down is going sideways or up.
Kevin Warsh chairs the FOMC for the first time on June 17. He comes to that meeting with a mandate he cannot fulfil — a president who wants cuts, a bond market demanding the opposite, and an inflation print that argues for neither patience nor aggression but for the kind of honest helplessness that central bankers are constitutionally forbidden from expressing publicly.
Bitcoin is at $80,000 and doing nothing interesting. The CLARITY Act markup hit its deadline today. The crypto industry's new regulatory framework is threading through a Congress that has exactly as much bandwidth for digital asset legislation as it does for everything else, which is to say: limited, harried, and intermittently attentive. The Warsh crypto enthusiasm feels very far away when the 30-year yield is at 5.20% and nobody cares.
The largest infrastructure expansion in human history is happening right now, quarter by quarter, GPU by GPU.
The bond market is breaking.
Both things are true.
I genuinely do not know which one wins on a 24-month view. Neither does anyone else, no matter what they tell you in the second paragraph of their morning note.
May 21, 2026
