Nvidia Vera CPUs to power Modal’s AI agents, challenging Intel and AMD - Tech Wire Asia

Nvidia's Vera CPU isn't a product launch—it's a sovereignty play. By anchoring Modal's agent orchestration layer to its own ISA, Nvidia is severing the x86 dependency that has underpinned data-center economics for two decades. The technical cascade is immediate: agent scheduling, tool invocation, and multi-step reasoning—tasks historically CPU-bound—now execute within NVLink's unified domain. x86's generality moat evaporates in agent workloads. Server OEM BOMs will see CPU share climb from roughly 18% toward 30%, reshaping memory and interconnect demand. Supply-chain exposure is acute. Vera's fab dependency on advanced-node capacity in Taiwan, China makes it a direct target of any export-control escalation. More insidiously, once enterprises bake agent runtimes into Nvidia's ISA, migration costs dwarf the x86-to-ARM transition of the 2010s—this is ecosystem lock-in, not product switching. Intel's likely counter: lean harder on Granite Rapids inference optimization. AMD: double down on EPYC's open-architecture pitch with hyperscalers. Both face the same structural trap—CPU margins can't fund an AI-ecosystem subsidy war. Within 18 months, agent-dedicated CPUs emerge as a discrete category. By month 24, data centers bifurcate into Nvidia's vertical stack versus the x86 open stack—a replay of the 2000s GPU-versus-CPU schism, with materially higher stakes.