@DollarCostAvg Those numbers are ridiculous: +196% revenue and +682% EPS? HBM sold out + long-term contracts. $MU is executing perfectly. $1500 is not crazy at all, could even go higher.
It's kinda crazy how HFT algorithms can tank a perfectly healthy market in minutes for no reason. Case in point, memory stocks on Friday (aka $MU $SNDK $WDC seagate and the general $SMH ). Here's what I think happened (NFA of course):
A @SemiAnalysis_ report said NVIDIA's Rubin NVL72 racks might ship with less memory per rack, roughly 55TB down to 28TB. The bots saw "memory" plus "NVIDIA" plus "halved" and flushed the whole sector in minutes.
But that 55TB number is system memory, not GPU memory. Two completely different things living in the same rack.
A Rubin rack has two memory tiers. HBM4 sits stacked right next to the GPU and feeds it data at very high bandwidth. SOCAMM and LPDDR sit on the CPU side (Vera) and act as system main memory for scheduling, host work, and spillover.
The cut everyone panicked about is the second tier. NVIDIA looks to be moving from 192GB SOCAMM modules to 96GB ones to protect the delivery schedule. That halves capacity per rack on the system side. HBM4 config and GPU shipments are untouched.
If you've ever spec'd a box, this is the difference between dropping your DDR5 from 1TB to 512GB versus touching the HBM on the accelerator. One changes how much you can hold in host memory. The other changes how fast you feed the compute. Not interchangeable, and only one of them is the genuinely scarce piece.
She slept on the office floor on a Wednesday night.
Alarm set for 4am. Jacket folded under her head.
Her team hit the deadline.
4 months later they hit her with a layoff.