Dell Leverages CPUs for AI Inference Growth

Seeking Alpha reports that Dell Technologies was rerated to a "BUY" on Apr 25, 2026, based on accelerating AI infrastructure demand and an anticipated growth cycle driven by inference workloads. The article attributes 40% FY26 growth to Dell's ISG business and cites an $43 billion AI backlog for that segment. Seeking Alpha describes Dell's CSG business as flat, noting margin pressure from rising DRAM/NAND memory prices, and observes a forward price-to-sales below 1.

Seeking Alpha frames the shift toward inference as favoring higher CPU capacity in servers, a pattern seen when workloads emphasize many smaller, lower-latency queries rather than large-batch training. Companies deploying inference at scale commonly evaluate throughput-per-dollar, latency, and utilization; these factors can make CPU-centric nodes or specialized inference ASICs more attractive than GPU-heavy designs for certain workloads.

Industry-pattern observations: memory price volatility tends to compress margins in client and storage-heavy segments even as enterprise infrastructure benefits from backlog visibility. Comparable vendor cycles show that visible, multi-billion-dollar backlogs can sustain revenue growth through hardware refresh cycles, but supply-chain constraints and component-price swings materially affect gross margin dynamics across vendors.

Key indicators to follow include reported ISG revenue growth and backlog conversion rates in upcoming earnings, trends in server CPU shipments and average core counts, enterprise adoption of CPU-optimized inference frameworks and libraries, and DRAM/NAND price trajectories. Observers should also track announcements from major CPU and accelerator vendors for product timing that alters the CPU-versus-GPU inference economics.

For infrastructure and ML ops teams, the broader pattern reported here implies revisiting procurement models and TCO calculations for inference deployments, including benchmarking CPU-based nodes for target latency and cost profiles rather than assuming GPU-first architectures.