Jeff Clarke did not need a chart to explain Dell Technologies’ problem. Asked on the earnings call which constraint was biting hardest, the vice chairman and chief operating officer answered with the same phrase he has used before. “The limiting factor is still the same, DRAM, DRAM, DRAM,” he said, “followed by NAND, NAND, NAND.”
The remark captured the paradox inside Dell’s latest results. Demand for the company’s artificial-intelligence servers has never been stronger, and its ability to satisfy that demand has never depended more on the memory industry’s output. Every record the company posted on Wednesday came with the same asterisk: it could have shipped more.
The numbers were large by any measure. Revenue for the quarter ended July 31 rose 58 percent from a year earlier to $46.97 billion, a record, and came in ahead of the roughly $44.8 billion that analysts had expected. Non-GAAP earnings of $7.04 a share were up 203 percent, far above the $4.90 consensus estimate. The stock rose about 11 percent in after-hours trading.
AI-optimized servers were the engine. Revenue in that business doubled from a year earlier to $16.4 billion in the quarter. Orders hit a record $60.9 billion, and the order backlog stood at $95 billion at the end of the quarter.
The gap between those two numbers, $16.4 billion of quarterly AI-server revenue against $60.9 billion of new orders, says something unusual about this market. Dell is adding to its backlog even while shipping AI servers at a record pace. Customers are not ordering one cluster at a time; they are committing to buildouts that run for years and accepting delivery schedules measured in quarters.
Management responded by raising its targets. Dell lifted the midpoint of its fiscal 2027 revenue guidance to $192 billion from $167 billion, and raised its full-year forecast for AI-optimized-server revenue to $74 billion from $60 billion. The reason, executives said, is that artificial intelligence has moved from experimentation to construction: customers are buying infrastructure not only to train large models but to run the inference systems and agentic workloads that depend on them.
Then the supply story. The constraint Mr. Clarke named is no mystery to anyone who watches the memory market. AI systems are built around accelerators that consume high-bandwidth memory in quantities the industry has never produced, and the same manufacturers that make that memory have been shifting capacity to serve the AI boom. The result has been tight supply and rising prices across DRAM, the memory that computers use for working data, and NAND, the flash memory used for storage.
For Dell, the shortage is a ceiling. The company buys memory in volumes that few others match, and its AI-server shipments are effectively capped by how many memory modules it can secure rather than by how many systems customers want. That is why orders and backlog keep setting records in the same quarter that revenue does: the queue grows faster than the line that builds the machines.
The wider industry feels the same squeeze. Every company building AI infrastructure, from cloud providers to server makers to the accelerator vendors themselves, is competing for the same memory, and the clearest winners of the boom so far include the memory manufacturers at whom Mr. Clarke’s complaint is aimed. For Dell, the squeeze carries a strategic twist: its AI servers are built around accelerators supplied by Nvidia, and the systems have made Dell one of the largest buyers of the memory that surrounds those chips.
There is a cost to the growth as well as a benefit. AI servers carry thinner margins than the personal computers and storage systems that Dell has sold for decades, so each shift in the sales mix toward AI changes the profit profile of a dollar of revenue. The earnings growth, with non-GAAP EPS up 203 percent, shows the AI business can be profitable at scale, but analysts who follow Dell watch the margin line as closely as the revenue line.
The raised guidance also implies a bet on the memory market’s recovery. If supply stays tight through the year, Dell will sell everything it can build, and the constraint Mr. Clarke named will keep showing up in the backlog. If memory makers bring new capacity online faster than expected, and they are expanding, the backlog should convert into revenue and the $95 billion pile at the end of the quarter will start to shrink.
For now, the company’s message to investors is simple: demand is not the problem, and the forecast is going up, not down. The problem, such as it is, sits in the supply chain, where a few memory makers hold the keys to Dell’s next several quarters. As Mr. Clarke put it, the limiting factor remains what it has been: DRAM, DRAM, DRAM, followed by NAND, NAND, NAND.
