AINEWS
COO Jeff Clarke says leading edge chip supply is the binding constraint, even as Dell booked $130 billion in AI server orders over the past year.
Dell's $95 billion AI backlog, more than twice its quarterly revenue, puts a number on the gap between AI orders and the hardware to fill them.
Chief Operating Officer Jeff Clarke put the bottleneck on the record during Dell's Q2 FY27 earnings call. Revenue for the quarter ending July 31 hit $47 billion, up 58% year over year. Of that, the Infrastructure Solutions Group, Dell's server and storage unit, contributed a record $31.8 billion, up 89%. Clarke also disclosed that Dell booked $61 billion in AI server orders in just the three months ending July 31, bringing the trailing twelve-month total past $130 billion. The constraint, he said, runs through the most advanced chip manufacturing processes: "just about every product going through a leading node" is supply-limited (Reseller News coverage, SEC filing index).
Dell split the quarter's growth across three customer classes: enterprise IT, the newer "neoclouds" built primarily around AI workloads, and sovereign cloud buyers in regions that want their data and compute inside national borders. The neocloud class has grown quickly, channeling private capital raised specifically to rent GPU capacity to AI labs and enterprises that cannot get it directly from the hyperscalers. CPU-based server revenue, which supports the agentic and inference workloads that much of the new buildout serves, climbed 122% year over year. The Dell AI Factory, the company's packaged platform for AI training and inference, has signed 6,500 customers to date. Three thousand three hundred of those were added in just the last three quarters, against 3,200 in the first two years after the offering launched in May 2024 (Reseller News coverage).
"Enterprise agentic AI is expected to be the single largest workload by 2028," Clarke said, treating it as an on-the-record projection rather than analyst consensus. He also called inference "pure demand in our industry" and said "agentic demand is reshaping the data center." Those statements sit alongside a separate forecast: 3,600 quadrillion tokens in use by 2030, an 87-fold jump from today, and training demand rising to 850 zettaflops, a fivefold increase. Both numbers are Clarke's forward-looking statements, not third-party measurements (Reseller News coverage, SEC filing index).
Hewlett Packard Enterprise reported a record Q3 of its own, with AI infrastructure and networking demand pulling growth forward, before supply constraints "tapped the brakes" on the back half of the quarter. Read together, the two prints describe the same sector mechanism: order books are full, factories are not, and the lag is now a line item in earnings calls rather than a logistics footnote (HPE Q3 coverage, Reseller News).
The shared chokepoint is upstream of Dell: foundry capacity at the leading edge, advanced packaging, and the high-bandwidth memory that sits next to those processors. Clarke's "just about every product going through a leading node" line is the mechanism in one sentence. The test for the next four quarters is how much of the $131.7 billion in demand Dell has converted into orders over the last year actually clears into shipped, installed, and billed systems before the 2028 workload mix he described arrives.