The Memory Supercycle: Why SanDisk Just Took the AI Infrastructure Crown

TL;DR: I ran five AI infrastructure names — Micron, AMD, SanDisk, ASML, Seagate — through six brutal financial rounds. SanDisk took the win with 13 points, combining 162.9% revenue growth, a 33.8% levered FCF margin, and a 5.9% debt-to-equity ratio. Micron came in second on raw profitability, ASML third on stability.

How SanDisk Built a 13-Point Lead

What stands out to me isn't that SanDisk won — it's how it won. The company didn't crush a single category. It quietly placed near the top of nearly every one: top in levered FCF margin (33.8%), top in debt-to-equity safety (5.9%), second in net margin (34.2%), second in revenue growth forecast (162.9%).

That kind of distributed strength is rare. Most memory and storage names are lopsided — Seagate has elite cash ROIC (50.3%) but a frankly scary 381.6% debt-to-equity. AMD has scale but lands last on net margin (13.4%). SanDisk is the only one in this group I'd describe as having no obvious weakness right now.

Why Memory Stopped Being a Commodity

For most of the last decade I treated memory as a cyclical commodity trade. Capital-intensive, price-taker, thin margins at the trough. This cycle is rewriting that assumption.

HBM is the bottleneck for every AI accelerator shipping today, and capacity through 2026 is already booked. That's not a marketing line — it's a structural shift. When NVIDIA, AMD, and the hyperscalers lock in multi-year supply contracts, the negotiation looks nothing like the spot DRAM market of 2018.

This is why I think the 41.5% net margin Micron is currently printing isn't an anomaly. It's what happens when a commodity supplier suddenly has multi-year visibility on price and volume.

Micron: The Best Value-for-Profit in the Group

Micron actually leads on three of the six rounds — net margin (41.5%), forward revenue growth (194.1%), and profit-adjusted PE (0.27). Where it stumbled was on cash ROIC (14%) and levered FCF margin (17.7%).

Here's how I read that. Micron is in a heavy capex phase — HBM3E ramp and next-gen DRAM nodes consume cash faster than they generate it on a quarterly basis. The short-term efficiency metrics get squeezed, but the forward revenue and margin tell you what the back end of this investment looks like.

The profit-adjusted PE comparison is the part I keep coming back to. Micron at 0.27. AMD at 3.89. That's a roughly 14x spread per unit of profit. Even if you discount the memory growth thesis aggressively, that gap is hard to defend on fundamentals alone.

What 2026 Sold-Out Actually Means

'Sold out' in a normal semiconductor cycle is a warning. It's the peak. It's the moment when capacity overshoots and the rollover begins.

I don't think it means that this time. AI infrastructure buildout is a 3-to-5 year capex cycle, not a 12-month inventory pulse. If 2026 is already gone, the conversations about 2027 and 2028 are already happening — and the companies that can deliver supply now are the ones writing those contracts.

Risks I'm Tracking

Three things I'd watch.

First, the buying side. HBM pricing power exists because supply is short. Once 2027 capacity additions ship, the balance shifts back toward the hyperscalers.

Second, SanDisk's debt-to-equity at 5.9% will not stay there. Capex cycles always lever up. The right question isn't whether it rises — it's whether it stays under 50%.

Third, Seagate's 381.6% debt-to-equity is a problem in any rate environment that isn't friendly. AI storage demand helps the top line, but it doesn't fix the balance sheet.

FAQ

Q: Why weren't NVIDIA and TSMC included?

A: The point of this comparison was to find the bottleneck names, not the obvious headline AI plays. GPU and foundry get plenty of analyst coverage already.

Q: How did Seagate score so low with a 50.3% cash ROIC win?

A: That single round gave them three points, but the 381.6% debt-to-equity in the final round kept them off the podium. High ROIC built on heavy leverage scores differently than high ROIC built on a clean balance sheet.

Q: Is ASML a 'loser' here?

A: No. It came third with 7 points and posted the second-best cash ROIC (40.1%) and third-best debt-to-equity (13%). It lost the growth round because lithography revenue cycles slower than memory — that's a feature of the business, not a bug.