作者：Think AI
The storage chip industry has followed a single rule for decades: shortage drives price hikes, capacity expansion leads to oversupply, then prices are cut to clear inventory.
Last night, Micron’s Q3 smashed that iron rule to pieces.
Revenue hit $41.456 billion, more than tripling year-on-year. Gross margin surged to 84.9%, an all-time high in the company’s 48-year history.
Single-quarter net profit reached $28.24 billion (roughly 192 billion RMB), compared to just $1.89 billion in the same period last year.
That translates to earning 24,700 RMB per second and 2 billion RMB per day. That is absolutely insane.
It even surpassed NVIDIA and Meta.
What does this mean? A company that sells memory sticks now has a higher gross margin than one selling GPUs.
“Long-term contracts with locked prices – the cyclical stock has turned into infrastructure.”
Micron locked contracts all the way out to five years. It signed 16 agreements, nailing down 20% of DRAM shipments and 30% of NAND shipments.
These long-term agreement customers include NVIDIA, Anthropic, Microsoft, and Google Cloud. The contracts specify minimum prices and minimum shipment volumes.
Based on these contracts, future guaranteed revenue backlog has already piled up to $100 billion.
This kind of play never existed in the memory business before. Multi-year long-term contracts with locked pricing are a privilege normally reserved for chip design houses and cloud providers.
Now it’s a memory stick seller’s turn.
“HBM sold out for the full year, supply gap visible through 2027”
Micron’s 2026 HBM capacity is 100% locked. It can only meet half to two-thirds of customer demand.
New production lines won’t release large-scale output until the second half of 2027 at the earliest.
The reason is simple. The memory used in an AI server is 6 to 8 times that of an ordinary server.
The world’s four major cloud providers have total computing capex this year of $725 billion. AI server shipments have doubled.
High-end wafers are all being redirected to HBM, which squeezes general-purpose memory supply and drives up its prices.
This is AI’s siphon effect on hardware. When one link explodes, the entire chain gets dragged along.
For ordinary consumer memory, gross margins top out at 30 to 40 percentage points. HBM’s gross margin is steady above 85%, and the selling price per unit is 10 times that of regular memory.
Micron relentlessly tilted capacity toward HBM. Overall gross margin was pulled straight into the 85% range.
That kind of profit margin previously belonged only to chip design companies. A wafer manufacturer, riding AI demand, has muscled its way right into the profit layer of a design house.
“What should we watch going forward?”
In the second half of 2027, new capacity from Samsung and SK Hynix will come crashing in. Their combined HBM capacity is 2 to 3 times Micron’s, and at lower cost.
Once the growth rate of AI computing capex slows, an 85% gross margin cannot hang on forever.
There is another layer of risk. Micron’s products basically cannot enter the Chinese mainland market.
The world’s second-largest market for computing demand can only watch from the sidelines. Over the next two years, more than $50 billion in capex will have to be digested entirely by overseas customers.
Honestly, Micron’s performance today is extremely good. Precisely because it is sitting in the window where AI demand is exploding and new capacity has not yet caught up.
That window only lasts one to two years at most.
The storage industry is indeed transforming from a cyclical commodity into infrastructure. The 84.9% gross margin is the best evidence of that.
But permanently elevating a memory company’s valuation to that of an AI chip stock requires more than just a few quarters of good results.
It requires AI computing demand to keep exploding, without any pause.
