A feature on how generative AI is redrawing data centre design — from storage-led optimisation to power-hungry GPU compute and high-bandwidth interconnect — and the energy, permitting, and supply-chain constraints that come with it. John Bradshaw is quoted on why retrieval-augmented generation carries most enterprise workloads: cheaper, faster to deploy, no scarce ML engineering talent required, and proprietary data never touches a training pipeline.
#infrastructure
4 items
John Bradshaw argues that major cloud outages reveal design failures, not just technology failures, and explains how abstraction across multiple providers creates true resilience rather than dependency on a single cloud's availability guarantees.
John's August 2024 interview with TFiR on the problems edge computing actually creates: distributed data gravity and workload orchestration. He argues the binding constraint is operator and developer experience rather than raw capacity, and sets out how Akamai approaches it.
An examination of the trade-offs between serverless and serverful computing approaches in modern cloud environments, exploring the infrastructure management decisions organisations must consider when building their cloud strategy.