Lam Research: Etch and Deposition Intensity Is the AI Memory Trade

Chips don't get faster by getting smaller anymore. They get faster by getting taller — and that shift hands the industry's hardest manufacturing problems to one company.

Lam Research builds the machines that carve and coat silicon wafers into working chips. Etch tools cut material away with plasma, like a sandblaster with atomic aim. Deposition tools lay material down, sometimes one atomic layer at a time — and Lam sells both to TSMC, Samsung, SK hynix, Micron and Intel.

For decades the roadmap meant one thing: shrink the transistor. That road is running out of pavement. So the industry turned the map ninety degrees and started building upward.

Look at what the AI boom actually demands. HBM memory stacks dies like floors in a tower. 3D NAND has blown past 300 layers and is heading toward 1,000. Gate-all-around nanosheets rebuild the transistor, backside power delivery moves the wiring under the silicon, and advanced packaging fuses chiplets into one piece.

Every one of those moves is etch- and deposition-intensive. Lam's own CTO says the 3D era roughly doubles etch-and-deposition intensity per wafer. More layers means more cuts and more coatings — and more tools.

The June quarter showed that mechanism at work. Revenue hit $6.72 billion, up 30% year over year, and non-GAAP gross margin reached 52.0% — the highest in about 20 years. Memory was 46% of systems revenue, split evenly between NAND and DRAM, with NAND dollars more than doubling quarter over quarter.

Zoom out and the pattern holds. Full fiscal 2026 revenue reached $23.23 billion, up 26% from $18.44 billion a year earlier. Systems revenue in the June quarter was $4.25 billion, with the rest coming from keeping the installed base running.

The forward view is louder. For the September quarter, management guided to $8.10 billion, more than 20% growth sequentially. At Citi's TMT conference this month, CFO Doug Bettinger raised Lam's calendar-2026 wafer-fab-equipment outlook to the low $150 billion range — his second hike this year.

His explanation is the interesting part. The bottleneck isn't demand, he said. It's cleanroom space, and the industry is fundamentally under-supplying demand. When your constraint is floor space rather than orders, you're looking at a different kind of cycle.

That's why Lam's serviceable market has climbed past 36% of total WFE, up from the low 30s. Gate-all-around transistors add roughly $1 billion of served market for every 100,000 wafer-starts a month. Backside power delivery adds about another billion on the same basis.

The tooling follows the roadmap. Akara, Lam's conductor etch platform for the 3D era, targets gate-all-around, 6F2 DRAM and 3D NAND. Lam Cryo 3.0 etches at cryogenic temperatures roughly 2.5 times faster, aimed at NAND past 400 layers, while Flex and Vantex handle dielectric etch.

Competition is real. Applied Materials spans a broader portfolio, holding roughly half of deposition and about a third of etch. Tokyo Electron is the genuine dielectric-etch challenger, with KLA in process control and Screen in cleaning. Lam leads etch and sits a strong number two in deposition.

Then there's the honest part — the risks. China was 26% of June-quarter revenue and keeps shrinking under U.S. export controls, down from 35% a year earlier. Customer revenue is concentrated in Taiwan and Korea, and memory capex has always been boom-bust, sold out or not.

The stock has had a huge run, too. After nearly tripling over the past year it trades at a premium multiple, and it slipped in a recent broad chip-equipment selloff. Chipmaking may be going vertical, but this is still a cyclical business wearing a structural story.

Micron says its entire 2026 HBM supply is sold out; SK hynix says DRAM, NAND and HBM are booked through 2026. Sold-out capacity is what turns a roadmap into purchase orders.

Lam is putting money behind its own thesis. The dividend rose 27% to $0.33 a quarter, and a more-than-$3 billion, five-year R&D lab expansion is underway, targeting over 50% more experiment capacity.

There's a quieter business underneath, too. Customer support runs about $2.5 billion a quarter, and trailing free cash flow sits near $4.9 billion. The installed base keeps paying long after the tool ships.

So here's the trade in one line. If AI memory keeps stacking higher, Lam gets paid for every layer. The market still files Lam under cyclical equipment, but content per wafer is quietly turning it into something more durable.

Disclosure: The Signal holds no position in LRCX. Positions may change. This is not financial advice.