Share with your CTO
Rafay Systems is positioning itself as the operating layer that turns raw GPU capacity into a sellable cloud service, and the company is betting that speed of deployment is the real moat. In a conversation covered by SiliconANGLE, CEO Haseeb Budhani argued that neoclouds, sovereign cloud providers, and telcos buying AI infrastructure at scale cannot afford to rebuild the control planes that AWS, Microsoft, and Google spent a decade constructing. Rafay’s platform spans bare metal, Kubernetes, virtual machines, serverless, and token-based consumption from a single surface, letting emerging AI clouds serve multiple customer segments without building each layer themselves.
What this means for your business
The gap Rafay is targeting is real, and it will land differently depending on which side of the infrastructure transaction you sit on. If your organization is evaluating regional or sovereign AI cloud providers as an alternative to the hyperscalers, the operational maturity of those providers is now a legitimate due-diligence question. A provider that cannot give you self-service provisioning, multitenancy, audit trails, and quota controls is not a cloud yet. It is custom infrastructure with a sales team, and that distinction carries real risk for enterprise workloads.
Budhani’s sharpest claim is that delivery capability, not software features, is the actual differentiator at this stage of the market. That tracks. The recurring failure mode in infrastructure platform plays is that the software ships, integration complexity multiplies, and the promised time-to-revenue stretches from weeks into quarters. Rafay’s counter is to embed its engineers before hardware even arrives and build toward a working service in days. Whether that delivery muscle scales as customer count grows is the question the sponsored format of this conversation, which tilts toward showcase rather than stress-test, leaves unanswered.
The deeper implication for enterprise CTOs is that the AI infrastructure vendor landscape is not settling into a few clean choices. Supply constraints alone mean a provider may plan around one set of networking and storage vendors and ship with a different mix entirely. That heterogeneity pushes the risk of integration upward, toward whoever owns the platform contract. If you are signing multi-year agreements with an emerging AI cloud, the right question to press on is not the GPU count. It is what happens to your SLAs when their hardware mix changes mid-contract and whether their operating software abstracts that change or exposes it to you.
Concept deep-dive: Control plane
A control plane is the management software that sits above physical hardware and lets customers request, configure, and monitor resources without touching the machines directly. Think of it as the brain that translates “give me a GPU cluster for this job” into the right sequence of network, compute, and storage instructions. Hyperscalers built these systems over years. Emerging AI clouds need one immediately, because without it they cannot offer self-service access, enforce security boundaries between customers, or track usage for billing.
Based on reporting from Rafay targets the AI infrastructure operating layer, originally published 2026-08-03 22:12:00.
