A cloud built for Python data scientists, not infrastructure engineers (thenewstack.io)

Coiled — founded by Dask creator Matthew Rocklin — is arguing that industrial-scale Python for data science should target raw VMs, not Kubernetes or Docker, and they’ve built tooling to prove it. Using a Python-first API (you can define clusters with simple decorators like vm_type, region, keepalive and toggle ARM with a comment), Coiled spins up hundreds to thousands of EC2 machines from a notebook, copies the user’s existing environment instead of forcing Docker image cycles, and wraps VM lifecycle, logging controls and cost guardrails (auto-shutdown of idle clusters, warnings for “chatty” logs). Rocklin demonstrated two 1,000-core clusters live and cited per-cluster costs that were tiny when configured properly; he also contrasts this with serverless solutions that carry 4–5x cost premiums and can’t access large machines or GPUs. For the AI/ML community this matters because it lowers the infrastructure friction that often blocks experimentation: data scientists can quickly try different instance types (ARM vs Intel vs AMD), GPUs, and regions from familiar Python workflows, speeding hardware and model exploration without becoming infrastructure engineers. The technical implications include easier reproducibility of local environments (avoiding Docker push cycles), richer hardware access for large-scale training/benchmarking, and the need for robust defaults and monitoring to prevent surprise bills. In short, Coiled reframes the cloud as a playful, experiment-friendly platform for production-scale Python compute.