HPE Q3 FY26 — The Warehouse Did the Quarter, and the Suppliers Financed It.

One print, two readings, six and a half hours apart

HPE opened at $47.60 against a prior close of $51.83, down 8.16%. Within the first half hour it traded at $45.70, down 11.83%. It then climbed all session and closed at $54.44, up 5.04%, with the final 30-minute bar running from $52.46 to the close on 9,681,362 shares — comfortably the largest volume bar of the day. Low to high, the range was 17.71% of the prior close, against a pre-print implied move of 9.94%.

No new information arrived in between. The release and the call were both out the previous evening. So the tape read one document twice and reached opposite conclusions, and the useful question is which read the filing supports. I think the morning was reading the balance sheet and the afternoon was reading the margin, and that the two are the same fact seen from different ends.

Juniper did not do this

The obvious explanation for a 1,090bps gross-margin expansion at a server company is that it stopped being a server company. Networking revenue was $2,893M against $1,654M, up 74.9%, and it is the high-margin half of the portfolio. That explanation fails on one line in the segment table.

Networking's segment operating margin was 22.02% this quarter, 21.60% last quarter, and 22.07% a year ago. It did not move. What moved was Cloud & AI, from 6.99% to 17.02%, on revenue of $9,042M against $7,212M. In dollars: total segment earnings from operations rose $1,259M year on year, and Cloud & AI supplied $1,035M of it — 82.2%. Networking supplied $272M, or 21.6%, with Corporate Investments giving back $(48)M. The chart puts the two segments side by side, because the sentence keeps getting read as though the mix argument survived it.

The question that was asked and not answered

Erik Woodring of Morgan Stanley put it to the CFO directly: "have you benefited from low-cost inventory now that's starting to uptick in your bill of materials?" That is the whole thesis of this piece, asked on the call, in one sentence, by someone who could see the same balance sheet.

Marie Myers answered: "I would think about Q3 more as a confluence of everything coming together at once." She then named four contributors — revenue scale, "disciplined pricing," the Catalyst cost programme, and "a mix of deals in the quarter that also impacted and really frankly benefited us" — and quantified none of them. The one thing she did not address is the one that was asked. A non-answer is not evidence of anything on its own; a non-answer to a question about the cost basis of inventory, in a quarter when inventory rose $2,789M on the balance sheet, is worth writing down.

Two of the four named contributors also carry their own warnings. "Disciplined pricing in traditional servers" is a price margin, and Antonio Neri described the environment it was earned in as "this constrained component environment" — that is a shortage rent, and shortage rents transfer to whoever adds supply next. "A mix of deals in the quarter" is mix is the softest margin to capitalize stated by the company: no price increase to point to, no cost removed to verify, just the weighted average of what happened to ship. Myers said as much about its durability in the next breath — "the mix of deals even inside of traditional server will also moderate as we go into Q4 and into '27."

The warehouse is the position

Inventory ended the quarter at $11,823M. A year earlier it was $7,163M; at the October fiscal year end, $6,352M; one quarter ago, $9,034M. That is +65.06% year on year, +86.13% from the year end and +30.87% sequentially. Expressed against the cost of goods it will be sold at, inventory days went 101.95 to 118.62 to 148.72.

This is the model this newsletter wrote on Longsys and Biwin in August, arriving at a $46B American server OEM: the warehouse is the position. For a downstream buyer in a component spike, gross margin is not a manufacturing margin — it is a holding-period margin, the spread between what the input cost when it was bought and what the finished good clears at now. Two consequences follow, and both are visible here. Inventory is not a warning label attached to the earnings; it is the asset that produced them, still on the books. And the P&L records the win at revaluation while the cash statement records it a quarter or two later, when the goods ship — so cash conversion, not margin, is the number that settles the print.

The model comes with a test, and the test is a differencing exercise the disclosures make possible. Cumulative cash-flow statements give the inventory outflow at six months, $(2,956)M, and at nine months, $(5,847)M. Subtract: the standalone third quarter consumed $2,891M — within $65M of what the first two quarters consumed together. Price-long, cash-late says a stockpile that is genuinely converting shows a narrowing quarterly outflow even as the balance keeps building, and a one-way bet shows a widening one. This one is widening, and close to doubling.

Who is financing the warehouse

Myers gave the working-capital summary in one sentence: "Our cash conversion cycle improved by 1 day from Q2, driven primarily by a decrease in days receivable due to more favorable billings linearity within the quarter, along with stronger collections. This was offset by an increase in days of inventory due to higher purchases in anticipation of future shipments." Every word of that is accurate. It is also the composite, and the composite is where a 30-day move hides.

Compute the three components against each quarter's own cost of sales and day count. Days of inventory ran 101.95 a year ago, 118.62 last quarter and 148.72 now, roughly 30 days added in a single quarter. Days sales outstanding ran 56.96, 52.39 and 46.59, about 6 days better, which is the collections Myers credited. Days payable outstanding ran 123.28, 148.52 and 172.75 — roughly 24 days longer. Net those and the cycle lands at 22.55 days against 22.49, which is how a company adds a month of inventory and reports a working-capital cycle that did not move.

Accounts payable is $13,734M against $8,662M a year ago and $7,731M at the fiscal year end — up 58.55% and 77.65%, and $2,423M sequentially. Against inventory of $11,823M, HPE now owes its suppliers more than the entire stockpile is worth. That is not a criticism; extending terms into a shortage is what leverage over a supply chain is for. It is the answer to who ultimately pays for the capex, asked at the working-capital layer rather than the fixed-asset one, and the answer is the component vendors. What it means for the reader is narrower: the cash cost of this position has not been paid yet, and the payables that deferred it come due on a schedule nobody disclosed.

The company guides the margin away itself

The strongest evidence for reading Q3's margin as temporary is not in my arithmetic. It is in HPE's fiscal 2027 framework. Myers put Cloud & AI's operating margin at "approximately 13%" — the segment that just printed 17.02%, guided down roughly four points, back to about where it sat in Q2. She was equally plain on gross margin: "Going forward, we expect our gross margin to moderate toward more historical levels driven by the growth in AI systems and the normalization in traditional servers, offset by the growing mix of networking."

Read that against the raise it arrived with. FY26 revenue growth goes to 34% to 37%, FY26 non-GAAP EPS to $3.75 to $3.85, FY27 revenue growth to 13% to 17% and FY27 EPS to $4.40 to $4.60. The raise is real and it is a volume raise: more revenue at a lower rate, with the FY27 non-GAAP operating margin framed at 14% to 15% against the 16.2% just reported. And the free-cash-flow raise to at least $5.0B came with its own attribution — Myers said "one of the biggest drivers of that improvement in the rate is really the fact that we've got a lot less restructuring as we go into '27," with the Juniper synergy tail to "bleed off by the sort of end of Q4 of next year." That is cash released by ceasing to spend it, not cash released by the working capital unwinding.

One more line deserves marking, because it is the difference between a stockpile and a position. A commitment is an allocation claim before it is a cost, and Neri made two of them: orders are "supported by our historically highest level of purchase commitments," and HPE has "doubled the number of commitments in terms of inventory," on supplier agreements "now multiyear, in some cases." Commitments of that size lock allocation, which is why they are also the reason the inventory line keeps growing after the quarter that gets applauded for it.

How this reads across the complex

The most valuable sentence on the call for anyone covering memory came from Neri, and it was not about HPE at all: "you have traditional DRAM moving to HBM. And that HBM demand is super high because it's driven by the GPU and the better memory that comes with it... Now in the traditional server, we use DRAM. We don't use HBMs."

• The mechanism, stated by the buyer. Wafer capacity converted to HBM is wafer capacity not making DDR5, and the DDR5 buyer is the traditional server. HPE's cost problem is the direct arithmetic consequence of the accelerator build-out — it is paying the scarcity rent created by demand it does not itself serve. That is one mechanism running through two pages published here today: Broadcom guided its consolidated gross margin down to approximately 73% from 78% a year ago on the memory content inside its XPUs, and HPE's margin went up because it bought DRAM before that content bid it away. Same wafer, opposite ends.
• Dell, one day earlier. Dell reported inventories of $21.29B, up 104.0% from its own fiscal year end, in the same window. Two server OEMs roughly doubling inventory in the same three months is not a company-specific decision; it is the industry taking a position on component prices.
• Networking is the clean part. Routing at $788M, up 270.0%, Data Center Networking at $382M, up 112.2%, orders up 36% against revenue up 10% normalized, and a segment margin that has not moved in a year. Neri also noted that "the working capital demand in networking is significantly lower because it's a faster turn to revenue once you get the inventory on hand." The part of HPE that is growing fastest is the part that does not need the warehouse.
• What is not disclosed. Backlog is called "record," "record-breaking," "at its highest ever" and "sizable," and the AI Systems backlog is given as up 14% sequentially. No dollar figure appears in the release or on the call. Dell put $95B on the same page one day earlier. An unquantified backlog is not evidence against the demand; it is an absence of the one number that would settle whether the inventory is spoken for.

The strongest case against this read

The best argument on the other side is that a build against contracted demand is working capital, not speculation, and HPE has more support for that than most. Orders grew 42% on a normalized basis against revenue up 33.7%; Networking orders grew 36%, about 3.5x its normalized revenue growth; AI Systems backlog rose 14% sequentially to a new high on AI orders of $1.4B, up over 30% sequentially, with AI Systems revenue of almost $1.6B and Q4 revenue explicitly guided higher on backlog conversion. Neri said the constraint is supply, not demand — "Growth in our backlog shows strong customer demand is running ahead of available supply" — and Myers tied the inventory directly to it: "you can see that in our inventory number, inventory number went up. So we're positioning ourselves to ship some larger AI transactions." On that reading, buying components you have already sold, using suppliers' terms to do it, while free cash flow still set a third-quarter record at $958M and net leverage reached 1.8x more than a year ahead of the 2x target, is not a position at all. It is a company converting a shortage into share.

I think that is the more likely case and it does not change the conclusion, because it does not survive its own test. If the inventory were pre-sold and converting, the quarterly cash outflow funding it would be narrowing while the balance kept climbing, and it is doing the opposite: $2,891M in one quarter against $2,956M in the previous two combined. The backlog that would resolve the ambiguity is the one number management declined to size. And the segment guide concedes the rest: a business shipping into contracted demand at scale should hold its margin, and HPE guides Cloud & AI from 17.02% to approximately 13%. Both readings can be true at once — the demand real, the margin temporary — which is exactly what a holding-period margin looks like from the inside.

Valuation, the trade, and what settles it

At $54.44, HPE is worth approximately $72.73B on 1,336M basic shares and approximately $78.88B on 1,449M diluted. Against the FY26 non-GAAP EPS midpoint of $3.80 that is roughly 14.33x, and against the FY27 midpoint of $4.50 roughly 12.10x. Neither is expensive, and neither is the argument. The argument is what the $3.80 is made of: GAAP net earnings of $1,540M this quarter include a $444M gain on the sale of an equity interest — 28.83% of the total — and the operating profit underneath was earned at a margin the company has already guided down.

On structure — analyst framing, not advice — the reaction day repriced the optionality as violently as the equity. Thirty-day implied volatility fell to 0.546 from 0.729, with an IV rank of 26.37 and a 10.7% implied move still attached to a name that just travelled 17.71% low-to-high in a session. For a holder, the setup argues for owning the Q4 cash-flow statement rather than the Q4 headline, because the headline is guided and the cash-flow statement is not.

Bottom line. Everything HPE reported is true and most of it is very good. Revenue of $12,213M, up 33.7%. Operating earnings of $1,393M against $247M. Free cash flow of $958M, a third-quarter record. Leverage at 1.8x, a year early. What the print does not establish is that the margin producing it belongs to the business rather than to the timing of a purchase order, and the one disclosure that would settle it — the dollar value of the backlog the inventory is supposedly against — was the one not given.

So the test is the cash-flow statement, not the income statement. HPE reported its Q4 and full-year fiscal 2025 results on December 4, 2025; fiscal 2026's fourth quarter ends October 31, 2026, and the full-year cash-flow statement will give the Q4 inventory movement by subtraction from the nine-month $(5,847)M. If that standalone outflow narrows below roughly $1.5B with revenue inside the $13.9B to $14.8B guide, the warehouse was shipping and this read was wrong. If it holds at or above $2,891M, with accounts payable still funding four-fifths of it, then the position is still open — and a holding-period margin gives the quarters back as readily as it takes them.