Analysis
Dell's fiscal Q2 earnings landed September 2 with a record $95 billion AI order backlog, and Pulse covered the number then. Nearly a week later, Fortune is arguing the market read the wrong headline. Its Tuesday analysis says the real story isn't the backlog size -- it's where that demand is coming from: enterprises pulling AI workloads out of hyperscaler clouds and back onto infrastructure they own.
The underlying numbers, restated:
- Revenue — $47B this quarter, up 58% year over year, vs. adjusted EPS of $7.04 against expectations near $4.90
- Guidance — raised $25B to $192B for the full year, implying roughly 70% annual growth
- Traditional servers — revenue up 122%; storage revenue up 26%
- Margin — infrastructure operating margin expanded 620 bps to 15%, which Morgan Stanley called unprecedented for a hardware business at this scale
- Customers — enterprise AI customers passed 6,500, with 3,300 added in the last three quarters alone
“The underlying numbers, restated: - Revenue — $47B this quarter, up 58% year over year, vs.”
Why on-premise is the sharper story
The consensus narrative all year has been that AI compute consolidates into a handful of hyperscaler and neocloud data centers -- the Microsoft, Google, Amazon, Crusoe and Nscale story Pulse has tracked repeatedly. Dell's numbers argue a second track exists alongside it: banks, healthcare systems and manufacturers that will not put certain workloads on someone else's cloud, for reasons ranging from data residency to regulatory exposure to plain cost control at scale. That is a fundamentally different buyer than a hyperscaler renting GPU capacity by the hour, and it is a buyer Dell, HPE and Supermicro are positioned to sell to in a way cloud-native infrastructure players are not.
The traditional-server growth number is the tell. A 122% jump in non-AI servers alongside AI backlog growth means enterprises are not just buying GPU boxes -- they are refreshing the surrounding infrastructure that makes on-premise AI viable at all: networking, storage, power distribution. That is a multi-year capital cycle, not a single earnings beat.
The counterweight
A backlog is a promise, not a shipment. Dell's $95 billion figure counts orders that have not yet been fulfilled or paid for, and order backlogs in capital equipment businesses routinely get pushed, renegotiated or partially cancelled when a customer's own AI budget tightens. The 6,500-customer figure is impressive breadth, but Fortune's piece does not disclose customer concentration -- how much of that $60.9 billion in quarterly orders came from a small number of very large buyers building their own data centers rather than a genuinely broad enterprise base. And Dell's stock move on the print was muted relative to the size of the beat, which suggests investors are already pricing some of this skepticism in.
For anyone selling AI infrastructure software or services, the practical read is that the on-premise buyer is underserved relative to the cloud-native buyer -- most AI tooling still assumes a hyperscaler environment. That is a real market gap, and Dell's backlog is the clearest evidence yet of its size.
