🧠 Food for thought

Implications, context, and why it matters.

Uber already did the work for Arm adoption

• Uber’s latest shift builds on a seven-year move off its own data centers 1.
• During the 2023 move to Oracle Cloud Infrastructure (OCI), Oracle’s cloud computing platform, and Google Cloud Platform, Google’s cloud computing platform, Uber added support for Arm-based Ampere processors, server chips built on Arm’s energy-efficient architecture, to cut costs and improve efficiency 2.
• That effort rebuilt the build pipeline for more than 5,000 software services, which lowers the effort needed to adopt another Arm-based chip such as AWS Graviton 2.

Uber’s Trainium test aims to cut AI costs

• Uber is testing Amazon’s Trainium3 AI chip as part of its push to reduce spending on large-scale AI workloads.
• AWS pitches its custom Trainium chips as delivering up to 50% lower training costs than graphics processing units (GPUs) for some workloads, including models with more than 100 billion parameters 3.
• Uber keeps Google Cloud and Oracle Cloud Infrastructure (OCI) in rotation while it tests Trainium3 1.
• With multiple providers, Uber can compare performance and seek better pricing, which pressures NVIDIA’s pricing leverage; AWS and some third-party analysis argue Trainium can run cheaper in some cost comparisons 4.