RXT: AMD Partnership; Single-Operator Accountability for Governed Enterprise AI

研报英文原文证据摘录

RXT: AMD Partnership; Single-Operator Accountability for Governed Enterprise AI

Rackspace Technology Inc

Financial benefits continued. Management indicated that by operating up stack to enterprise

AI, RXT expects to capture margin throughput higher than commodity bare metal GPU services,

though specific pricing guidance was not provided. We believe over time RXT can increase Private

Cloud adj gross margins (36% in 1Q26) to 40-55% over the mid-term.

Workforce realignment. Concurrent with this infrastructure commitment, RXT announced a

workforce realignment reducing headcount by up to 15% globally, with one-time restructuring

charges of $14-19m and anticipated annualized run-rate savings of $75-85m. The reduction is

mainly driven by strategic decisions to de-emphasize certain legacy service delivery functions,

primarily within Public Cloud. A significant portion of the anticipated savings will be reinvested

into forward-deployed engineering, AI solutions delivery, and enterprise AI infrastructure build

out. Management characterized this as deliberate capital reallocation from offerings towards

governed enterprise AI platforms.

Efficient token utilization. The combination of AMD Instinct GPUs and EPYC CPUs enables

RXT to optimize token efficiency by partitioning workloads across heterogeneous compute.

AMD Instinct GPUs (MI355X and MI350P) handle tensor operations for inference while

EPYC CPUs execute CPU-intensive tasks such as tokenization, embedding operations, and

context management. This architecture reduces GPU idle cycles and memory bottlenecks,

improving tokens-per-second throughput per watt. For the Enterprise Inference Engine tier

specifically, EPYC CPUs manage context operations while MI-series GPUs execute forward passes,