A memory card printed with a large speed figure can still stop a recording within seconds. The number on the label and the number video recording depends on are usually not the same measurement.
The headline number is a peak read speed
The largest figure on most packaging describes reading under laboratory conditions, which matters when transferring files and is irrelevant while recording.
Writing is consistently slower than reading, and the write figure is either printed smaller or omitted entirely.
Even the quoted write speed is typically a burst, achieved while the card's fast buffer has room, and not a rate that can be held for minutes.
Video needs a guaranteed floor, not a maximum
A camera writes a continuous stream and cannot pause. If the card drops below the required rate for even a moment, the recording stops.
This is why video speed classes exist. They state a minimum sustained write rate the card will not fall beneath, which is the only figure that predicts whether a mode will record.
A card with a very high peak and no guaranteed floor can be less suitable for video than a slower card carrying an appropriate class rating.
Sustained performance degrades as a card fills
Flash memory writes fastest into empty blocks. As a card fills, the controller must reorganise existing data before writing, and the sustained rate falls.
A card that records a demanding mode reliably when empty can fail the same mode when nearly full, which makes the failure look intermittent and mysterious.
Heat compounds this, since controllers reduce speed to protect themselves, and a card inside a warm camera during a long take is working at its worst.
The camera is the other half of the equation
A card can only perform as well as the slot it sits in, and a fast card in an older interface delivers the interface's speed rather than its own.
Cameras with two slots do not always run both at the same speed, and recording simultaneously to a slower second card can throttle the whole operation.
Manufacturer compatibility lists exist because of these interactions, and testing a specific card in a specific body remains the only reliable check.
Counterfeits fail exactly here
Fake cards report a large capacity and a fast rating to the host while containing slower memory, and they pass casual tests that only write a small file.
The failure appears under sustained load, which means during a long recording rather than during the check that was supposed to catch it.
Verifying a new card by filling it completely and reading it back is slow, and it is the only test that exercises the conditions where a card actually lets you down.