Micron's $150 Billion Boogeyman Has a Wafer Problem (And It's Not Micron's)
Everybody's screaming that $150 billion in fab spending will drown the memory market by 2027, as if memory makers have never once been dumb in a group. But HBM eats wafers like a teenage boy eats pizza, and that changes the math the doomsayers are doing on their cocktail napkins.
- Roarable rates Micron (MU) a STRONG BUY at $108.50 with a $142.00 target, an RQD Score of 83/100, and a forward P/E of just 13.8x. That's a bargain price on a business the market still treats like a commodity.
- The bear case counts dollars of capex. The bull case counts wafers. HBM3e consumes roughly 3x the silicon of regular DRAM per bit, so a big chunk of that $150B gets swallowed just keeping up with HBM, not flooding the shelves with cheap memory.
- Quality is showing up in the numbers: ROIC 19.4%, gross margin 46.2%, Profitability grade A, and 91% of EPS revisions heading up in 90 days. Analysts are quietly raising their hands and saying 'oops, we were too low.'
$MU Ratings Summary & 5-Factor Quant Grades (Now vs. 3M vs. 6M)
| Factor Grade | Now (Unlocked Free) | 3M Ago | 6M Ago | Primary Forensic Metric |
|---|---|---|---|---|
| Valuation | A | A | A- | Fwd P/E 13.8x vs. Sector Median |
| Growth | A+ | A | B+ | YoY Revenue & FCF Per Share Trajectory |
| Profitability | A | A | B+ | ROIC 19.4% β’ Gross Margin 46.2% |
| Momentum | A- | B+ | B | 3M / 6M Relative Strength vs. S&P 500 |
| EPS Revisions | A | B+ | B | 90-Day Upward vs. Downward Consensus Revisions |
The $150 Billion Headline Is Scary. The Denominator Is Where the Money Is.
Here is how the doom trade works. Somebody adds up every announced fab, every cleanroom, every shiny new tool order, lands on $150 billion, and then declares that supply is about to bury demand. It makes a lovely chart. It also treats a wafer like a wafer, which is like treating a pickup truck and a Prius as the same thing because they both have four wheels.
HBM is not a regular wafer. It takes about three times the silicon of standard DRAM to produce the same number of bits, and then you stack it, bond it, test it, and lose some along the way to yield. So when the industry pours money into capacity, a big slice of it vanishes into a black hole of wafer consumption. You build a bigger bathtub, then the new drain is three times wider. That is the part the headline skips, and it is the part that pays dividends to the people who read past the headline.
Why Memory Isn't Acting Like Memory This Time
Old-school memory was a miserable business for a simple reason: everybody made the same thing, nobody could differentiate, and when prices rose, every CEO ran out to build a fab at the same time. Boom, glut, bust, repeat. Investors learned to sell the good news because they'd been burned so often they could smell smoke in their sleep.
HBM scrambles that script. It's qualified customer by customer, sold against tight allocations, and tied to the most cash-hungry compute build-out in a generation. Look at what Micron is showing: a 46.2% gross margin, 19.4% ROIC, and a Profitability grade of A. That isn't a commodity wearing a nice suit. That's a business getting paid for being hard to make. And when 91% of EPS revisions over 90 days point north, the sell-side isn't being generous. They're being dragged to the numbers.
So Where's the Catch? There's Always a Catch.
The catch is timing and discipline. If the big three go full lemming in 2026 and 2027, the 'reckoning' crowd gets to take a victory lap, and nobody should pretend that can't happen. Memory has humiliated smarter people than me. Our Valuation grade is a B+, not an A, because you're not buying this at a fire-sale price anymore, and a stock with an A- Momentum grade can fall hard if the cycle blinks.
But look at what you're being asked to pay. At 13.8x forward earnings and $108.50, the market is pricing Micron as if the cycle peaks tomorrow, while Growth (A-) and EPS Revisions (A) say the opposite. Our $142.00 target doesn't require a miracle. It requires memory to be less stupid than usual and HBM to stay scarce. Position it like an adult: size it so a capex hiccup is a bad week, not a bad decade. Then let the wafer math do the heavy lifting.
π¬ Community Debate (4 Trader Comments)
Finally somebody talking wafer starts instead of dollars of capex. The 3x silicon per bit on HBM is the whole ballgame. Every bear deck I've seen ignores trade ratio. Add the stacking yield losses and effective bit supply grows way slower than the announced spend implies. Still holding my 400.
I've watched this movie in 2018 and again in 2022. 'This time memory is different' is the most expensive sentence in semis. Everyone's HBM roadmap is ramping at once, and the moment one player gets desperate for share, pricing cracks. 13.8x is cheap because it's peak-ish earnings. Not buying the 'discipline' story.
@CycleScarred I respect the scar tissue, I really do. But 2018 and 2022 were commodity DRAM and NAND gluts, where every wafer was interchangeable. HBM is allocated, qualified, and chews up three times the silicon. Different animal. And if 13.8x is 'peak earnings,' explain why 91% of revisions in 90 days are going up. Peak doesn't usually get raised that much. That said, I'm not asking you to love it. I'm asking you to size it like it can hurt you, and watch the capex guides like a hawk. If they start going wild, I'll be first to tell you.
I run MU alongside VRT for the same reason: AI buildouts need memory and need cooling, and both are bottlenecks. A 19.4% ROIC with that gross margin tells me they're getting paid for the scarcity, not just riding a wave. Only thing I'm watching is free cash flow conversion once the capex bill hits.