For years, Nvidia was the obvious place to look if you wanted to own the AI boom.
Now there is another company sitting much closer to the bottleneck.
Micron.
And Nvidia’s own numbers are starting to explain why.
Nvidia just revealed that its supply commitments exploded from $119 billion to $279 billion in one quarter.
The company said the increase was primarily related to memory.
Think about that for a second.
Nvidia is the company everyone thinks of when they think about AI chips.
Yet Nvidia is committing hundreds of billions of dollars to suppliers because it needs the stuff that goes around those chips.
At the same time, Nvidia’s gross margin is starting to feel the squeeze.
It reported 75% gross margin in Q2 fiscal 2027. Management expects 74% in Q3 and says margins could fall to roughly 71%–72% before recovering toward 72%–73% in fiscal 2028.
The reason management gave investors was unusually blunt: memory prices are rising much faster than Nvidia expected.
Now look at Micron.
Micron’s fiscal Q3 revenue was $41.46 billion.
Net income was $28.24 billion.
Operating cash flow was $25.39 billion.
Its Cloud Memory business produced an 83% gross margin.
Its Core Data Center business produced an 87% gross margin.
And Micron guided its next quarter to roughly $50 billion of revenue with approximately 86% gross margin.
So while Nvidia is telling investors that memory is eating into its margins, Micron is selling that scarce memory at margins that would have looked ridiculous for a traditional memory company.
That is the part of the AI trade that deserves more attention.
Because memory isn’t some optional accessory attached to the GPU.
Nvidia’s next-generation Vera Rubin platform is being built around HBM4. Micron is already shipping HBM4 designed for Rubin, with its 36GB 12-high product in high-volume production.
And HBM is incredibly memory-hungry to manufacture.
Micron’s CEO said the company is deliberately keeping its HBM market share close to its overall DRAM share because HBM consumes a significant amount of wafers and puts pressure on the supply of conventional memory.
That’s a nasty little feedback loop.
AI companies want more compute.
More compute requires more memory.
The industry shifts manufacturing capacity toward HBM.
That leaves less capacity for conventional DRAM.
Then conventional memory gets tight too.
And suddenly the AI boom is competing with phones, laptops, servers and automobiles for the same underlying manufacturing resources.
We’re already seeing the consumer side get hit.
Reuters reported that memory can account for as much as 60% of the component cost of some $400 smartphones, while smaller manufacturers are struggling to secure supply. Counterpoint expects global smartphone shipments to fall 13.9% this year.
That’s not just a Micron story.
It’s an Nvidia story.
Because Nvidia’s extraordinary AI demand is helping create the very memory shortage that raises Nvidia’s own costs.
And Micron is sitting on the other side of that transaction.
Nvidia needs more memory.
Micron wants higher prices.
Nvidia wants enough supply to ship more AI systems.
Micron has to decide how aggressively to expand without destroying the pricing environment that just transformed its financial statements.
This is why the old description of Micron as a simple commodity memory company is becoming harder to use.
Micron has already moved into multi-year agreements with major customers.
It is investing billions into additional capacity.
It has HBM4 in volume production.
It is working on HBM4E for 2027.
And management says overall memory markets could remain tight beyond 2027 because capacity takes years to build and HBM consumes so much manufacturing capacity.
But there is a catch.
Micron cannot simply print money forever.
Memory is still a semiconductor business.
Eventually competitors add capacity.
Technology transitions change wafer economics.
Demand can slow.
And history says memory companies eventually discover that everyone building factories at the same time is a very expensive way to create another oversupply.
That is the risk Nvidia doesn’t have in quite the same way.
Nvidia can change architectures.
It can raise prices.
It can bundle more networking and software.
It can redesign systems.
It has CUDA and an enormous ecosystem surrounding the accelerator.
Micron has something different.
Scarcity.
Right now, scarcity is extraordinarily valuable.
And Nvidia’s own disclosures are basically telling us how valuable it is.
The company says supply remains a bottleneck through at least fiscal 2028.
It is locking up components years ahead.
It is carrying $32 billion of inventory ahead of the Vera Rubin launch.
And it is warning shareholders that memory pricing has become worse than expected.
That’s where the Micron versus Nvidia comparison gets interesting.
Nvidia owns the software ecosystem and the accelerator architecture.
Micron doesn’t.
Nvidia captures an enormous amount of value from every AI system.
Micron doesn’t capture anywhere near as much revenue per system.
But Nvidia is also the customer getting squeezed by one of its suppliers.
Micron is the supplier watching the world’s richest AI companies fight over its product.
That doesn’t make Micron the “next Nvidia.”
That’s too simplistic.
It means the AI supply chain is developing a second profit center.
The first wave made Nvidia extraordinarily profitable because compute was scarce.
The next wave could make memory extraordinarily profitable because the compute boom itself is consuming the memory required to build more compute.
And there’s the part I would watch next.
Micron is spending heavily to expand.
Nvidia is spending heavily to secure supply.
Both companies are effectively betting that AI demand will grow faster than semiconductor capacity.
If they’re right, Micron has an unusually favorable setup.
If they’re wrong, Micron has a much more familiar problem.
The memory industry builds too much capacity, prices collapse, margins disappear, and everyone suddenly remembers that memory has always been cyclical.
That’s the real comparison.
Nvidia is betting that AI demand keeps outrunning the system.
Micron is betting that one of the system’s scarcest components stays scarce long enough to turn a traditionally cyclical business into something much more profitable.
And right now, Nvidia’s own financial statements are giving Micron’s argument some pretty powerful evidence.
Got a news tip or correction? Let us know
If you got something out of this, please chip in to keep this site running, or subscribe to go ad-free.
0 views