Wall Street has taken note as AI moves from massive model training to inference, reviving demand for legacy hardware. Central‑processing units that once powered generic servers are seeing renewed orders, lifting stocks of chipmakers such as AMD and ARM. Analysts now gauge how much of the expanding AI spend will flow back to this “old‑IT” stack.

Shift to inference is not the only catalyst. AI agents—software that automates routine tasks—require CPUs in roughly a one‑to‑one ratio with GPUs, according to Intel’s CFO. That balance helped drive a 400% rally in Intel shares after a government stake and sparked a 60% surge in Seagate. Companies that can position legacy components as core to agent‑driven workloads may preserve pricing power.

Software vendors are feeling the pressure too. Salesforce rolled out a “headless” version of its cloud platform, aiming to let AI agents tap its vast customer data without a human‑focused UI. The gamble hinges on whether its data moat can outweigh the risk of reduced user‑based revenue. The market is already testing that hypothesis, forcing a reassessment of the value chain that underpins modern AI deployments.