HPE and Vultr accelerate deployment of AI inference-focused datacenter infrastructure.

Cloud infrastructure provider Vultr has selected Hewlett Packard Enterprise (HPE) and Nvidia technology for a new wave of AI infrastructure deployments as enterprise demand switches from AI experimentation toward production workloads. The announcement, made at HPE Discover 2026 on Wednesday, positions Vultr among a growing group of cloud providers expanding dedicated AI infrastructure to serve enterprise training and inference workloads. Under the agreement, Vultr plans to deploy Nvidia GB300 NVL72 systems supplied through the Nvidia AI Computing by HPE portfolio. The deployments will incorporate Nvidia Spectrum-X Ethernet networking and HPE liquid cooling technologies as part of new AI-focused data center environments. Antonio Neri, president and CEO of HPE, said: "Vultr represents a new generation of AI cloud providers, and the company's selection of HPE validates the importance of AI data center architectures designed to support the next wave of global AI growth."

The deployment reflects a significant shift in how enterprises approach AI infrastructure. According to Ron Westfall, vice president and practice lead for networking and infrastructure at HyperFrame Research, "We have reached the inflection point where inference is transforming from a secondary operational phase into a primary, long-term driver of AI infrastructure investment." He noted that "the industry is shifting its spending focus toward infrastructure optimized for production throughput, cost efficiency and rack-to-rack networking rather than just raw training power." Vultr CEO J.J. Kardwell confirmed this market movement, stating that customer demand has shifted sharply over the past year from AI experimentation toward production deployments. Three years ago, most GPU demand came from startups and AI developers building and training new models, but today, a growing share comes from organizations running production AI services, including inference workloads tied to customer-facing applications and business operations.

Kardwell emphasized the competitive pressure enterprises face in deploying AI capabilities. He noted that "when you reach a point in the market where you're seeing operating margins for public companies be favorably impacted, we're now seeing changes to the way large public firms think about staffing levels because of the economic impact of these AI capabilities." He also highlighted how traditional enterprise procurement cycles are struggling to keep pace with AI infrastructure evolution. Organizations accustomed to six- to eighteen-month planning cycles often find capacity unavailable by the time purchasing decisions are finalized, creating demand for cloud-based AI infrastructure that can be deployed more quickly. According to Kardwell, more enterprises are adopting hybrid infrastructure strategies rather than choosing between cloud and on-premises deployments, with large organizations continuing to build their own AI infrastructure while also turning to cloud providers to gain faster access to the latest GPU platforms and additional capacity.

Networking has emerged as a critical component in large-scale AI deployments. Kardwell stated that performance bottlenecks emerge once workloads extend beyond individual servers and racks, making high-bandwidth fabrics essential to AI infrastructure design. He emphasized that "the bottleneck is the second you leave that rack," pointing to the growing importance of east-west traffic within large AI clusters. Westfall concurred, observing that "when AI clusters expand beyond individual racks, networking evolves from a standard infrastructure component into a performance bottleneck and a key competitive advantage." Future deployments will feature Nvidia Spectrum-X networking with 400 GbE and 800 GbE connectivity designed to support large AI clusters and rack-scale GPU architectures. Additionally, data sovereignty requirements are becoming increasingly important as enterprises integrate AI systems into core business processes. Vultr currently operates cloud infrastructure across 33 data center locations in 17 countries, a footprint the company says helps address customer requirements around data residency and regional deployment.