BofA: AI Memory Boom Unstoppable — Micron Earnings to Explode 34%
Bank of America Says the AI Memory Boom Isn’t Over: Why Micron Could Be a Hidden Gem
Quick Summary: Bank of America thinks Wall Street is way too pessimistic about memory chip maker Micron. Thanks to AI’s insatiable hunger for a special type of memory called HBM, BofA projects Micron’s earnings could grow 34% every year until 2030—yet the stock trades like the party ends tomorrow.
The Big Idea: AI Changed the Rules of the Game
For decades, the memory chip business has been a roller coaster:
- Boom: High demand → high prices → huge profits
- Bust: Factories catch up → oversupply → prices crash → profits vanish
Investors expect this cycle. So they treat Micron (NASDAQ: MU) like a cyclical stock—buy at the bottom, sell at the top.
Important Point: This time might actually be different. AI doesn’t just want more memory—it wants a specific, hard-to-make kind called High Bandwidth Memory (HBM). And making HBM eats up 3x the factory space per bit compared to regular memory. That means supply can’t flood the market overnight like it used to.
Micron’s Recent Report Card: Straight A’s
| Metric | Fiscal Q3 2026 | Year Ago | Change |
|---|---|---|---|
| Revenue | $41.46 billion | ~$22 billion | ~88% ↑ |
| Gross Margin | 85% | 38% | Massive jump |
| DRAM Revenue | $31.3 billion | ~$7 billion | 343% ↑ |
| NAND Revenue | $9.9 billion | ~$2 billion | 361% ↑ |
| Operating Cash Flow | $25.39 billion | — | Record high |
ELI5: Gross margin = (Revenue – Cost to make chips) ÷ Revenue. 85% means for every $1 of sales, Micron keeps 85¢ after manufacturing costs. That’s insane for a commodity business.
BofA’s "SanDisk-Like" Dream Scenario vs. Consensus
BofA built a bull case modeled after SanDisk’s historic run (before it was acquired). Here’s how it stacks up against Wall Street’s consensus:
| Fiscal 2030 Projection | Consensus (Average Forecast) | BofA’s Bull Case |
|---|---|---|
| Sales | $280.5 billion | $377.3 billion |
| Gross Margin | 78.0% | 80.0% |
| EPS (Earnings Per Share) | $136.24 | $236.16 |
| Free Cash Flow | $190.8 billion | $188.6 billion |
What this means: BofA sees 30.7% annual sales growth and 34.1% annual EPS growth through 2030.
The Disconnect: Despite this, Micron trades at ~6x forward earnings (based on 2027 consensus EPS of $151.37). That’s "priced for a bust," not a multi-year AI supercycle.
Why HBM Is the Secret Sauce
High Bandwidth Memory (HBM) is not your grandpa’s RAM. It’s:
- Stacked vertically (like a layer cake) using TSVs (Through-Silicon Vias)—tiny elevator shafts for data
- Glued right next to the GPU (the brain of AI) for lightning-fast data transfer
- Insanely complex to make → 3x wafer capacity per bit vs. regular DRAM
Micron’s HBM roadmap:
- HBM3E (8-high & 12-high) → Shipping now
- HBM4 → Volume production started
- HBM4E → Expected 2027
Supply Constraint = Pricing Power: Hyperscalers (Microsoft, Google, Amazon, Meta) need more HBM now. But you can’t just "turn on" a new HBM line—it takes years. This keeps the market tight.
Other Tailwinds Pushing the Boat
| Trend | Why It Matters for Micron |
|---|---|
| Enterprise SSDs | AI data centers generate massive data → need fast, high-capacity storage (NAND) |
| AI Inference | Running AI models (not just training) needs new memory architectures |
| High Bandwidth Flash (HBF) | New category from SK hynix & SanDisk: sits between HBM and SSDs, up to 512GB & 3 TB/s |
Note: HBF doesn’t hurt Micron—it shows how fast memory is evolving. Today’s winners won’t own every new category.
The "Yeah, But…" Section: Why BofA Might Be Too Optimistic
| Risk | Reality Check |
|---|---|
| 80% Margins Forever? | Historically, memory margins average 30–40%. 85% today = unusually tight market. |
| Competition Is Waking Up | SK hynix (HBM leader), Samsung (catching up), Chinese suppliers (ramping) |
| Capacity Will Come Online | If everyone builds fabs → oversupply → price war → margins normalize |
| $236 EPS Requires Perfection | That’s 73% higher than consensus. Needs structural change, not just a long cycle. |
The Middle Ground: You don’t need BofA’s max case to like Micron. Even at consensus numbers, 6x earnings for a company printing $25B/quarter in cash flow with 85% margins is… cheap.
Key Takeaway: Directionally Right, Destination Debatable
| BofA Gets Right | BofA Might Stretch |
|---|---|
| AI fundamentally changes memory economics | 80% margins sustained to 2030 |
| HBM + Enterprise SSD + HBF = multi-year demand | $236 EPS (vs. $136 consensus) |
| Long lead times = slower supply response | Micron becomes "structurally different" |
| Market pricing in a bust too early | Ignores competitive dynamics |
Bottom Line: Treat BofA’s $236 as the bull case, not the base case. At 6x earnings, the market may be underestimating how long this boom lasts—even if it doesn’t last forever.
Summary: Should You Care?
- AI needs special memory (HBM) → Hard to make → Supply constrained → Good for Micron
- Micron is crushing it right now → $41B revenue, 85% margins, $25B quarterly cash flow
- Wall Street expects a crash → Stock priced at 6x earnings (very cheap)
- BofA says "not so fast" → Projects 34% annual EPS growth to $236 by 2030
- Risks are real → Competition, cyclicality, margin normalization
- Even the "boring" case looks interesting → Cheap valuation + strong fundamentals = asymmetric bet
FAQ: Your Questions, Answered Simply
1. What is HBM and why is it such a big deal?
HBM (High Bandwidth Memory) is memory chips stacked vertically and placed right next to the GPU. It uses tiny vertical connections (TSVs) to move data super fast. AI models are huge— they need this speed. Regular RAM can’t keep up. Making HBM takes 3x the factory space per bit, so supply is naturally limited.
2. Why does Micron trade at only 6x earnings if things are so good?
Because investors have PTSD from past memory cycles. Every time margins hit 50%+, they assume a crash is coming. They’re pricing Micron like the bust starts tomorrow, not in 3–5 years.
3. What’s the difference between DRAM and NAND?
- DRAM = Fast, temporary workspace (like your desk). Loses data when power off. Micron’s DRAM revenue +343%.
- NAND = Slower, permanent storage (like a filing cabinet). Keeps data without power. Micron’s NAND revenue +361%.
4. Can’t Samsung and SK hynix just build more factories and crush margins?
They’re trying. But:
- Building a leading-edge memory fab takes 3–5 years and $20B+
- HBM is much harder than regular DRAM
- By the time new capacity comes online, AI demand may have grown even more
5. Is this a "buy" recommendation?
No—this is education, not advice. But the setup is intriguing: a cyclical stock priced for a bust, in a structural upcycle driven by AI, with record profits and a clean balance sheet. Do your own homework (DYOR).
Want to Go Deeper? The team behind this analysis (24/7 Wall St.) partners with Doomberg—an independent, subscriber-supported newsletter written by heavy industry & private equity veterans. They cover energy, macro, and geopolitics with zero ads and zero institutional agendas. 24/7 Wall St. readers get 17% off their first year →
Article based on Bank of America Global Research analysis and Micron fiscal Q3 2026 results. All figures in USD. Not investment advice.