Haider Aziz of VAST Data discusses how modular infrastructure could address power and capacity constraints, support local data control and bring AI inference closer to users.
Power, cooling and data centre capacity are becoming critical constraints on AI deployment, prompting infrastructure providers to rethink where and how organisations run inference. VAST Data and UAE-based Hyperfusion have announced plans to integrate the VAST AI Operating System into Hyperfusion’s Sentinel modular AI infrastructure as its preferred data platform, with the aim of bringing sovereign, production AI inference online in weeks rather than waiting years for conventional data centre capacity.
Sentinel packages a complete AI data centre into a transportable 45-foot ISO High Cube unit, with a reference configuration of 96 GPUs. The proposed integration combines Hyperfusion’s modular deployment approach with VAST’s data platform to support inference closer to enterprise data, users and available power.
Technical validation and the integrated product architecture remain under development. Speaking to TahawulTech.com, Haider Aziz, General Manager – Middle East, Turkey and Africa at VAST Data, explains the infrastructure requirements, sovereignty considerations and operational work needed to turn the joint vision into deployable capacity.
Interview excerpts
What infrastructure bottlenecks does the VAST–Hyperfusion partnership address, and what must customers have ready to bring AI inference online in weeks?
The biggest constraint is no longer access to GPUs; it is access to power, cooling and suitable data centre capacity. Hyperfusion’s modular approach is designed to deploy around available power rather than waiting years for large greenfield sites, while VAST provides the data layer needed to make that compute useful quickly.
Customers still need the underlying power allocation, site readiness and network or operational requirements defined, but the goal is to remove much of the traditional build-out delay and get sites online faster.
How will VAST AI OS support data sovereignty and confidential computing across Sentinel deployments, including fully air-gapped environments?
VAST provides the control layer that lets an organisation enable sovereign AI, keeping data, models and inference governed within the environment where they need to run. That also extends to confidential computing and isolated deployments, including air-gapped environments where inference can continue locally without relying on external cloud connectivity. The remaining work is around validating the operational model for securely updating models and data in those disconnected environments.
How does a shared global namespace help customers scale across Sentinel and HyperPod deployments while maintaining local control of their data?
A shared namespace, like VAST’s DataSpace, is intended to let customers treat distributed infrastructure as one logical data environment rather than a collection of separate silos. That means models and datasets can be managed consistently across Sentinel and larger HyperPod deployments, while individual sites can still retain local control over where data is stored and processed. The benefit our customers see and value most is simpler scaling without forcing them to centralise everything physically. It’s a much more pragmatic view on modern infrastructure.
Which UAE and Middle East use cases stand to benefit most from placing inference closer to users, enterprise data and available power?
We see a lot of very different use cases, but typically the common thread is those organisations where data sensitivity, latency or infrastructure availability make centralised cloud inference less practical: government, sovereign AI, energy, telecommunications, critical infrastructure and enterprises with sensitive operational data. In those environments, placing inference closer to the data can improve control and responsiveness while making better use of distributed power capacity.
What technical validation remains before the joint offering becomes available, and how would a common control plane simplify operations across multiple sites?
The companies still need to complete technical validation of VAST within the Sentinel and HyperPod architectures, including sizing, integration, performance and operating requirements. A future common control plane is intended to simplify multi-site operations by giving customers one way to monitor and manage distributed deployments rather than treating each Sentinel as a standalone system. The technical collaboration is active, but the integrated product architecture is still being finalised.
Looking ahead, what will define the next phase of AI infrastructure, and why does the location of inference matter?
This next phase of AI infrastructure is not just about building bigger AI factories. It is about putting inference where the data, users and available power actually are. Globally, that matters because organisations want the benefits of advanced AI without giving up control over sensitive data or waiting years for conventional infrastructure to be built.
Source: Tahawul Tech