AI Compute Demand Broadens Beyond GPUs as BofA Lifts Server CPU TAM Above $210 Billion

The AI infrastructure trade widened materially over the past session as Bank of America lifted its total addressable market estimate for server CPUs to above $210 billion, arguing that the deployment of autonomous AI agents requires far more general-purpose compute, memory and I/O capacity than the accelerator-only narrative implies. The call landed alongside a strong quarterly report from Lenovo, which dragged Dell and HP higher in sympathy and offered the clearest evidence yet that enterprise hardware refresh demand is no longer confined to hyperscale GPU clusters. The read-through for Nvidia is nuanced but constructive. Sell-side commentary circulating on the wires contends that the concentration bear case — the argument that a handful of hyperscalers underwrite the bulk of data-center revenue — has lost roughly half its footing as sovereign, enterprise and neocloud buyers scale orders. That broadening is precisely what infrastructure-linked equity strategies are positioning for, with thematic vehicles tying AI demand to physical build-out in power, cooling and construction. The counterweight is financing. A widely circulated analysis of AI compute asset-backed securitisation frames the emergence of structured credit against GPU collateral as the mechanism by which the cycle could ultimately unwind, echoing cautionary remarks attributed to Nvidia's leadership. For now, however, order books and TAM revisions are moving in the same direction, and the marginal buyer is paying up for hardware exposure rather than trimming it.