Dell Just Proved The Memory Bill Gets Passed Down — And Named Who Pays It

Dell's Infrastructure Solutions Group posted a 15.0% operating margin in its fiscal second quarter, against 10.5% the prior quarter and 8.8% a year ago. Operating income of $4.78 billion rose 225% on revenue up 89% to $31.78 billion. Dell attributed the expansion to pricing discipline, scale leverage and high-margin Dell IP storage attachment. The same quarter, TrendForce had conventional server DRAM contract prices rising 13 to 18 percent sequentially.

The setup going in was clean. Dell's own CFO had guided to a sequential improvement in ISG margin. A rise from 10.5% meant Dell was passing higher memory costs through. A flat or lower print meant some of the bill was landing on Dell. It came in at 15.0% — not a hold, a 450 basis point sequential expansion and 620 year over year, in the quarter memory costs rose fastest. That settles a question the whole memory trade rests on. The shortage is real, and the cost of it is not being eaten by the box makers. The chain now has a number and a named source at every link. HBM spot pricing is running four to five times long-term contract prices — a 36GB HBM3E stack trades near $2,100 on spot against a contract price roughly a fifth of that, per MegaGrid Supply. That premium pulls fab capacity toward HBM and away from conventional DRAM, and HBM4 makes it worse rather than better: it yields worse than HBM3E, so each unit shipped consumes more DRAM. Samsung has committed roughly 70% of its capacity through 2031 to long-term agreements, so what is left for anyone outside a contract is a residual market. Server DRAM contract prices then rise 13 to 18 percent. Dell passes that through at an expanding margin, while telling investors it expects continued supply constraints in DRAM and NAND. Palo Alto Networks, on the same day, independently guided that rising memory and storage costs will keep pressuring its hardware business. Then the last link, which Dell also supplied. Jeff Clarke said on the call that the AI infrastructure business is broadening across neoclouds, sovereign customers and enterprises. Dell booked a record $60.9 billion of AI orders in the quarter — more in three months than in all of fiscal 2026 — and exited with a $95 billion backlog. Neoclouds are the weakest credit in that list. Goldman Sachs offers clients a basket of 18 US high-yield AI issuers, CoreWeave and Applied Digital among them, tradeable as total return swaps, averaging a 319 basis point spread against 267 for broad high yield. Financing identical GPU collateral costs roughly 300 basis points more for a non-investment-grade offtaker than an investment-grade one. So the memory shortage is real, its cost is being transmitted intact through the manufacturer, and it terminates on the balance sheets that pay the widest spread in the chain. Dell books it as backlog. Someone else books it as debt.