Why great earnings can sink a stock — reading AI results
Stocks move on surprise versus expectations, not on the absolute numbers
The three lines
- The same evening: AMD -9% on record revenue, SpaceX +9% on a $541M loss
- Markets grade against expectations, and the pre-report rally is part of the bar
- Five checks: core segment, consensus gap, guidance, pre-pricing, margin path
Key questions
- Why does a record quarter drop a stock
- Because the record was already in the price — and then some. AMD entered its report up 7% on the day; beating data-center consensus by ~$200M couldn't clear what the rally had priced. 'Record' and 'disappointing' can be the same number.
- What should I actually check in an AI earnings report
- Five boxes: ① core-segment growth (data center, for chipmakers) ② the gap versus consensus ③ next-quarter guidance ④ how much the stock already rose before the print ⑤ margin direction. These five explain the next morning far better than the headline.
- Why do loss-making companies jump on earnings
- Growth names are priced on the speed of the path to profit. SpaceX lost $541M but nearly halved its year-ago loss and doubled the expected EBITDA — so it jumped. Different baseline, same principle.
On the evening of August 4, two companies reported. The one with record revenue fell nearly 9% after hours. The one with a half-billion-dollar loss rose 9%. That pair — AMD and SpaceX — is the cleanest demonstration of how earnings season actually works, and this reference answers the search it provokes: why did the stock drop on good earnings?
1. The principle: markets grade the gap, not the score
By report day, expectations are already in the price. What moves the stock is the difference between expected and delivered — and the pre-report rally is part of expectations. AMD's $6.7B data-center print beat the $6.5B consensus, but the stock had climbed 7% that day pricing more. SpaceX's $541M loss looks bad in absolute terms and excellent against a $1B expectation.
Two sides of one rule: good number < high bar → down. Bad number > low bar → up.
2. The five-box test
| Box | Question | This week's case |
|---|---|---|
| Core segment | How fast is the AI segment growing? | AMD DC +107% (58% of firm) |
| Consensus gap | Beat by how much? | +$200M — a slight beat |
| Guidance | Next quarter's outlook? | judged against raised bars |
| Pre-pricing | How much had the stock run? | +7% that session |
| Trajectory | If loss-making, shrinking how fast? | SpaceX loss nearly halved |
The headline ("record revenue") answers none of the five. Filling this table, not quoting the press release, is how to read an earnings story.
3. What is still open
This document gets re-linked every earnings season. This week's specifics — AMD's record-and-drop, SpaceX's debut — run in their own pieces; the AI release race and the frontier-model glossary complete the cluster.