The memory shortage made Micron rich. Now watch who pays — the bill is starting to show up in the filings

The memory cycle has a winner and the winner is loud. Micron's FY2026 Q2 revenue hit $23.86 billion at a 74.4% GAAP gross margin, up from 36.8% a year earlier — we walked through that filing here. DRAM contract prices rose an estimated 58–63% quarter-over-quarter in Q2 2026 and NAND 70–75%, the steepest moves in roughly fifteen years, with the three memory makers reportedly shifting the bulk of combined output toward HBM for AI.

That's the side of the trade everyone is positioned for. The more interesting research question — the one that fits how we work — is the other side of every contract: someone is paying those prices. When a commodity input doubles, the cost lands somewhere downstream, and it lands first on the companies that turn memory into systems: server OEMs, AI system builders, and PC makers. The useful question isn't "is memory hot" (it is). It's where the cost is being absorbed, who can pass it through, and whether any of it is yet visible in a filing rather than a headline.

(Financial research, not investment advice. Built from Aether search over SEC filings; evidence labels at the end. Where a figure comes from press/industry data rather than a filing, we say so.)

The squeeze, stated plainly

Memory is a large, volatile line in the cost of any server. When DRAM and NAND prices move 50–75% in a quarter, three things happen to the companies one tier down:

Cost of sales inflates faster than they can reprice existing backlog.
Gross margin compresses unless they pass the cost through fully and immediately.
Working capital and inventory risk rise — buy ahead to secure supply and you carry price/obsolescence risk; buy as-needed and you may not be able to fill orders.

None of that is a forecast. It's the mechanical consequence of a commodity spike hitting a bill of materials. What we wanted to know is how much of it the filings already describe — because risk-factor language written years ago, when read against today's prices, is often the most honest forward indicator a company gives you.

What Super Micro's own filings say about exactly this scenario

Super Micro ($SMCI) is the cleanest case: a pure AI-server builder whose cost of sales is memory, CPUs, GPUs and storage. Its risk factors describe the current squeeze almost prophetically. From the 10-K (SEC source):

"Prices of certain materials and core components utilized in the manufacture of our server and storage solutions, such as serverboards, chassis, CPUs, memory, hard drives and SSDs, represent a significant portion of our cost of sales… we do not have long-term supply contracts for all critical materials and core components, but instead often purchase these materials and components on a purchase order basis. Prices of these core components and materials are volatile… if shortages, supply or demand imbalances or delays arise, the prices of these materials and key components may increase or the materials and key components may not be available at all."

And the part that matters most when you read it against a 60% DRAM move:

"From time to time, we have accepted customer orders with various types of component pricing protection. Such arrangements have increased our exposure to component pricing fluctuations and have adversely affected our financial results in certain quarters."

That is a company telling you, in its own SEC filing, that when component prices spike and it has locked customer pricing, the margin damage shows up in results. It also flags that in shortages "some of our larger competitors may have greater abilities to obtain materials and key components due to their larger purchasing power" — i.e. allocation favors scale, which is itself a competitive variable in a shortage.

Evidence label: SEC-filed for the risk language above. The application to this specific 2026 price spike is an inferred relationship — the filing describes the mechanism, not the current quarter's realized hit. The honest move is to read the next 10-Q's cost-of-sales and gross-margin lines against these words.

Supplier concentration: the shortage rewards whoever is already big

The same SMCI filings quantify how concentrated the input side is. In one quarter, two suppliers accounted for 55.1% and 10.3% of total purchases (SMCI 10-Q, SEC source) — the 55% driven by GPU purchases to build customer solutions. The filing also notes that "certain materials… are available from a limited number of suppliers" and "shortages could occur in these materials due to an interruption of supply or increased demand in the industry."

Read that against a memory market where three vendors control essentially all DRAM and HBM supply and are reallocating wafers toward the highest-margin product. A system builder buying on purchase orders, from a handful of suppliers, into a shortage where capacity is being steered elsewhere, is structurally the price taker in the relationship. That's not a knock on the company — it's the position the supply chain puts it in.

This is exactly the kind of thing our supply-chain map is built to make visible: who depends on whom, and which direction the pricing power runs. (SMCI's chain →)

Evidence label: SEC-filed for the supplier-concentration percentages (as of the cited filing date). The read-through to current pricing power is inferred.

The cost is already being acknowledged out loud

Two more data points sit at different evidence tiers, and both point the same way.

In its own SEC-filed earnings commentary, Alpha & Omega Semiconductor ($AOSL) — a power-MOSFET maker that sells into PCs and data center — told investors that for calendar 2026, "visibility into the PC market remains limited, driven primarily by uncertainty around memory shortages," and that "memory availability may impact end PC demand" while "data center investment continues to provide an important offset" (AOSL earnings exhibit, SEC source). That's a third party, in a filing, naming the memory shortage as a swing factor in downstream demand — and flagging the split between a pressured PC channel and a still-funded data-center channel.

AOSL's own earnings call three months later (Q3 FY2026, 2026-05-06) made the split sharper: PC weakness was "likely exacerbated by earlier pull-ins… and potential demand impacts from rising memory pricing," even as Advanced Computing (AI, servers, graphics cards) "more than doubled sequentially" (AOSL transcript, SEC source). Same memory shortage, opposite signs on the two channels — in one company's filing.

And the AI-server side is now visible in a filing, not just the press. Dell's Q1 FY2027 release (2026-05-28) reported AI-optimized server revenue of $16.1 billion, up 757% year-over-year, $24.4 billion of AI orders booked, and raised its FY2027 AI-server revenue target to $60 billion — with CFO David Kennedy noting execution was "exceptionally strong across the business — from supply chain to sales to pricing" (DELL exhibit, SEC source). That's the easier-pass-through channel in numbers: explosive AI-server revenue while the company explicitly names pricing and supply chain as the levers it's pulling — exactly where a memory spike gets absorbed or passed on.

Separately, and at a lower evidence tier: press coverage of Dell and HPE earnings has Dell's COO calling the memory shortage "unprecedented" and saying the company is "repricing… every day," with server makers planning price increases on the order of 15% and HP guiding a roughly 30-cent FY2026 EPS impact from commodity costs. We label that press-reported rather than filing-cited — but it's consistent with what the filings' risk language predicts, and it tells you where to look in the next set of 10-Qs.

Evidence label: SEC-filed for the AOSL commentary; press-reported for the Dell/HPE/HP figures (verify against each company's next 10-Q before treating as confirmed).

The two-speed channel: who passes it through, who eats it

Put the pieces together and the downstream picture is not uniform — it's a spread:

Hyperscaler-facing AI servers are the easier pass-through. Customers want capacity now, supply is the binding constraint, and contracts can reprice. Cost inflation here is more revenue than margin destruction — though "component pricing protection" clauses, where they exist, are exactly where a spike still bites (SMCI says so itself).
PC and commodity-server channels are the harder pass-through. Buyers are price-sensitive, BOM targets are fixed, and a 50–75% memory move against a fixed-price target is straight margin compression or lost volume — the AOSL "may impact end PC demand" risk.

So the same shortage that prints record margins for the memory maker can show up downstream as higher revenue with squeezed gross margin for the system builder, and as demand destruction for the consumer-PC tier. One catalyst, three different income-statement signatures. That's the part generic "memory is hot" coverage misses — and the part you can actually check in filings.

How to research this yourself (and what to check)

This is a watch-the-next-filing thesis, not a price call. Concretely:

Read cost-of-sales and gross-margin lines in the next 10-Qs from server OEMs and builders against the memory price moves. SMCI's own filings tell you the mechanism; the numbers will tell you the magnitude.
Watch for "component pricing protection" or fixed-price backlog language — that's where a spike converts directly into a margin miss.
Separate the channels. AI-server revenue can rise while blended margin falls; don't read the top line alone.
Trace the dependency on the supply-chain map: who buys memory, from whom, with how much concentration.

Search the evidence directly with Aether: try component pricing protection memory cost of sales or supplier concentration GPU purchases to pull the risk language and concentration disclosures yourself, then run the financials on the names — SMCI, DELL, HPE, MU — on EvidInvest, free.

Bottom line

The memory super-cycle is real and the supplier side is already in the income statement — Micron proved that. The under-covered half is the demand side of the same contracts: server OEMs and system builders whose largest cost line just spiked, whose own filings warn that locked customer pricing "adversely affected our financial results in certain quarters," and whose supplier concentration leaves them as price takers into a shortage. None of that is a sell signal. It's a map of where to look — and the next round of 10-Qs is where the bill becomes a number.

Evidence labels: SEC-filed — SMCI component-pricing-protection and core-component risk language (10-K 2023-08-28), SMCI supplier concentration 55.1%/10.3% (10-Q 2023-11-03), AOSL memory-shortage / PC-demand commentary (earnings exhibit 2026-02-05) and the AI-vs-PC split (Q3 FY2026 transcript 2026-05-06), Dell AI-optimized server revenue $16.1B/+757% and raised $60B FY2027 guide (Q1 FY2027 exhibit 2026-05-28), Micron FY2026 Q2 results (8-K exhibit 2026-03-18). Inferred — application of SMCI's general risk language to the specific 2026 price spike; downstream margin/demand read-through. Press-reported — Q2-2026 DRAM/NAND contract-price moves (industry trackers), Dell/HPE/HP "unprecedented"/repricing/EPS-impact commentary. Verify each against the relevant company's next 10-Q before treating as confirmed. Not investment advice.