In a hyperscale cloud market crowded with abstractions and layered virtualization, Voltage Park is staking its claim on the raw simplicity of bare metal infrastructure.
With over a billion dollars invested exclusively in AI hardware, six dedicated AI datacenters, and full-stack designs that reject conventional virtualization in favor of direct hardware access, the specifics reflect not mere preference but strategic choice
This is a neocloud built on the premise that performance demands exacting control. Rather than hiding resource allocation behind cloud abstractions, Voltage Park provides direct access to GPU nodes deployed on Nvidia HGX H100 clusters, connected through InfiniBand fabrics capable of speeds up to 3.2 terabits per second.
Ethernet networks complement this architecture, ensuring comprehensive high-speed connectivity. Workload management is handled by Kubernetes and Slurm schedulers, providing predictable orchestration of their AI workloads without hidden constraints.
Bare metal, as Voltage Park’s Principal Network Engineer, Drew Pletcher, describes it, is more than marketing. It solves real-world bottlenecks and operational bumps in the road AI customers face with the largest hyperscale cloud providers.
Pletcher points to cases of customers who struggled with unpredictability and performance inconsistencies in virtualized hyperscale cloud environments. He cites one customer example involving a rapid expansion from an initial 64 GPUs to over 2,000 GPUs within six to eight months.
That kind of scaling is not so much about raw capacity as it is predictability and insight into performance, all of which are enabled by their unabstracted infrastructure.
Yet, scaling bare metal GPU infrastructure to thousands of nodes brings storage into critical focus. And here is where Pletcher is clear about differentiation.
Pletcher points to benchmarks where VAST exceeded anticipated performance levels. They specifically cite rigorous compliance capabilities required by enterprise customers, including SOC 2, HIPAA, FINRA, and SEC requirements, which VAST reliably meets.
Voltage Park explains their technical evaluation openly, noting that their core requirements were flexibility, performance, redundancy, and multi-tenant security, all provided by VAST’s EBox hardware architecture.
Pletcher explains that workloads on their infrastructure frequently run uninterrupted for weeks, sometimes months. In this case, demand for sustained checkpointing is where VAST shines. He also points to a range of customer data handling preferences, from direct streaming via S3 interfaces to hierarchical data staging.
Across all these scenarios, VAST reliably accommodates intense, continuous data flows without compromise.
But what’s the best metric for Pletcher and team? VAST, is “definitely is not keeping anybody awake at night.”
That sleep metric translates technical reliability into a human, measurable impact.
Infrastructure is fundamentally about predictability, a fact Pletcher openly acknowledges. Predictability, in turn, is directly tied to peace of mind. The ability for engineers and data scientists to rest assured that their workloads will proceed uninterrupted is arguably the most meaningful form of infrastructure performance.
Voltage Park’s embrace of bare-metal systems, coupled with explicit integration of best-in-class storage from VAST, reinforces an infrastructure imperative that prioritizes straightforward excellence over layered obfuscation.
By investing over one billion dollars into dedicated hardware, including Nvidia HGX H100 GPU nodes connected by InfiniBand and Ethernet, managed transparently through Kubernetes and Slurm, and complemented by VAST’s superior storage, Voltage Park provides a template of how infrastructure should be done.