"I think there was a level of disbelief that there would be a significant pickup because it felt like oversupply had been there forever," he said.
The Wall Street analyst community had been able to see spending on AI chips continuing for training, Bryson noted - but beyond that, questions revolved around how AI models would continue to progress and whether the spending on AI was sustainable. Additionally, Bryson said he thinks chip manufacturers would have thought more about adding capacity sooner if they had anticipated an uptick in AI inference fueling demand.
Micron wasn't alone in underestimating the level of demand for AI-linked products. Bryson also pointed to Intel, which found itself without enough supply of central processing units as they became more essential to inference workloads.
And although Micron was a recipient of funding from the Biden administration's Chips and Science Act, Bryson said the company didn't seem to be pushing hard on building its facilities in New York and Idaho to bring capacity online.
In an attempt to manage the boom-and-bust cycle this time, Micron has introduced a new type of long-term agreement that it has said will help smooth out demand trends from its big customers.
