Can someone explain the underlying economics between Micron and NVIDIA, and how the Street is coming up with earnings estimates for Micron that are ~1/2 NVDA's? I got the following numbers from ChatGPT, are these numbers wrong or is some assumption off?

Based on these numbers, NVDA should make at least ~7x whatever earnings MU makes? In that case, either the consensus earnings estimates for MU are super high, or NVDA is super low? These are both widely followed companies so I doubt analysts would be this off?

(The "score" it references below can be understood as overall market share as a % of AI spend).

But in peak pricing mode

If HBM prices double or memory tightness pushes memory content value much higher, then:

|Scenario|HBM content per accelerator|MU share|MU weighted capture of accelerator ASP| |:-|:-|:-|:-| ||||| |Normal / base|~$2.9k|20–25%|~1.5–2.0%| |Tight market|~$5k–$6k|20–25%|~3–4%| |Extreme memory squeeze|~$8k–$10k|20–25%|~5–7%|

That is the answer: peak pricing can take MU’s effective value capture from ~2% to maybe ~5–7% of accelerator ASP. That is a big jump, but still nowhere close to NVDA’s platform capture.

At rack/system level

For full systems, memory can be a much larger cost bucket. Recent reporting on future Rubin-class racks suggested memory could reach around 25% of total system build cost in an extreme configuration.

But even then:

So even under a very favorable system-level view, MU’s weighted deployment value capture is probably mid-single digits, not half of NVDA.

Bottom line

Peak-pricing MU score should not be 1–4. It is more like:

|Company|Normalized structural capture|Peak-pricing capture| |:-|:-|:-| |||| |MU|~1–4|~5–8| |NVDA|~35–60|~35–60+|

The relative earnings assumption is from these numbers, this suggest people are estimating a future earnings for MU that is about 1/2 NVDA:

Current valuation snapshot

|Metric|MU|NVDA| |:-|:-|:-| |Market cap|~$1.02T|~$5.24T| |Share price|$895.88|$214.86| |Trailing P/E|~42.3x|~32.7x| |Forward P/E|~9–10x|~24–25x|