HPE's $7.6 billion AI backlog waits on memory supply

HPE sells the servers, networking gear, and storage that businesses and governments use to run AI in their own data centers. Its latest quarter showed demand at record highs: revenue grew 34% year over year to $12.2 billion, adjusted earnings jumped 152% to $1.11 per share, and total AI orders of $3.1 billion pushed the backlog to $7.6 billion. Bookings for traditional servers were also up 75% as customers bought AI-enabled hardware.

The catch is supply. Memory and storage drives are short, and CEO Antonio Neri warned the tightness has been an issue all year and will likely last longer, with consequences for cost and pricing. The networking segment shows the gap clearly: orders rose 36% year over year while normalized revenue rose just 10%. Server revenue grew 35%, but that was driven primarily by higher average selling prices from higher input costs, not by shipping more units.

The company is leaning on its networking business, now roughly 30% of total operating profit, to offset some of the margin pressure. Networking carries higher margins and is expected to land in the mid-to-high 20% range for fiscal 2027, compared with around 13% for servers. It is also winning large deals, including a multi-gigawatt deployment for Oracle's AI data centers, and may supply switches for AMD's Helios rack-scale systems. For investors, the outcome hinges on whether HPE's locked-up memory capacity lets components align with its order book in the current quarter, and on whether supply constraints prove worse or last longer than expected.