Nadia covers the AI trade with a healthy suspicion of round numbers and a longer memory than most of the market. She has called two bubbles correctly and one early, and will remind you of the ratio.
The chipmaker's short-lived plan to bankroll the clouds that buy its chips is the most honest thing the AI trade has done all year — because it stopped. The precedent it's flirting with is one the market agreed to forget.
A confidential S-1 lets a company go public on a number it hasn't yet had to defend. On Wall Street, the gap between a run rate and an audited figure is where the money gets made and lost.
A new filing puts more than 80% of Nvidia's disclosed equity in two names — a $21 billion SpaceX stake and $30 billion in Intel. The circularity is the point, and the number that should worry you is the one that hasn't been marked yet.
Apollo's Torsten Slok says the boom's returns are funded by investors, not customers. He's describing a shift from equity to credit — and credit prices default, not disappointment.
Revenue grew 93 percent and the company raised its full-year guidance by the most it ever has. None of that is the question. The question, as always, is what you paid for it.
Microsoft and Amazon were rewarded for the same capex that cost Meta close to a tenth of its value in a night. The number that separated them wasn't how much they spent. It was whose revenue it became.
A transformer-only chip startup is reportedly raising two rounds at once, at valuations twice apart, before its first chip has earned a dollar. The gap tells you what neither figure will.
It's lining up an IPO at a $965 billion private mark, ahead of OpenAI. The company is real. The number is a multiple of a run-rate — and a run-rate is last month wearing a calendar.
SK hynix is a great company that just sold $26.5 billion of new shares into the greatest memory shortage on record. The record is a fact about the moment, not the company — and the proceeds are the tell.
Together AI raised $800 million and tripled its valuation in sixteen months renting out Nvidia's chips. Whether the company is good was never the question. The question, as always, is the price — and who's already long.
A $269 billion single-session drop wasn't a verdict on Google's chips. It was the market repricing the assumption underneath the whole AI trade: that the people who make the compute worth anything come with the building.
The cash-rich chipmaker just sold its biggest-ever slug of bonds. The more interesting debt is the borrowing it isn't on the hook for — the loans its lease commitments quietly make safe. We have seen a vendor underwrite its own demand before.
On June 22, every S&P 500 fund on earth must buy Marvell, a chip designer whose stock has roughly tripled this year. Forced buying is not the same as conviction — and the gap between them is where the AI trade keeps its risk.
Morgan Stanley expects AI debt issuance to nearly double to $570 billion this year. An equity mania costs the believers. A debt mania sends the bill to everyone else.
SpaceX, OpenAI and Anthropic are walking through the same door at once. The door is the tell. The trillions are the question.
The chips are real and the demand is real. That was never the question. The question is what you paid, and whether the money going around the circle is the same money counted twice.
