Microsoft Azure customers can now run workloads on Cray supercomputers
Mary Jo Foley / ZDNet :
Context & Ripple Effects
This is another step in Microsoft's campaign to pull specialized hardware under the Azure umbrella. Earlier in 2016 it partnered with HPE, Dell, and Lenovo on Azure Stack, bringing Azure services into customers' own datacenters; now the direction reverses, with Cray's supercomputers exposed as an Azure workload target.
The endpoint of that trajectory shows up three years later, when Microsoft builds a 285,000-processor supercomputer inside Azure exclusively for OpenAI — evidence that folding exotic compute into the cloud became core strategy, not a side offering.
First-order effects
- HPC customers who could never justify buying a Cray can now rent supercomputer capacity by the workload, while Cray gains a distribution channel through Azure instead of relying on direct system sales.
Second-order effects
- Cloud rivals face pressure to match on-demand access to top-tier HPC hardware, and Cray's economics shift from selling boxes to selling utilization through hyperscaler channels.
Third-order effects
- If the pattern holds, supercomputing consolidates around cloud platforms: specialty system vendors become capacity suppliers inside hyperscalers, culminating in arrangements like Azure's exclusive OpenAI machine rather than standalone national-lab-style procurements.
The trend: High-performance computing is migrating from owned machines to rented hyperscaler capacity, turning specialist vendors like Cray into suppliers inside the cloud platforms.