The emergence of neocloud providers is shifting the enterprise infrastructure focus from raw GPU acquisition to the complex orchestration of multi-tenant AI services. To address the operational friction inherent in this transition, Aviz Networks and hosted·ai have announced a partnership to integrate networking automation with GPU commerce capabilities. This collaboration aims to bridge the gap between disconnected compute, networking, and monitoring workflows that currently complicate service delivery for operators. By combining Aviz’s AI factory portfolio with the hosted·ai platform, the two companies are positioning a unified operating model designed to help neoclouds deploy, monitor, and monetize AI infrastructure more efficiently through automated networking and token-level observability.
Integrating Aviz AI Factory with hosted·ai Commerce
The partnership centers on merging Aviz Networks’ software layer for the NVIDIA AI Factory into the hosted·ai platform, targeting environments spanning NVIDIA Spectrum X Ethernet, InfiniBand, and NVIDIA BlueField, as well as multi-vendor networking setups. Aviz is contributing four specific technical components to the integration: ONES AI, an NVIDIA-certified automation software for bootstrapping and orchestrating multi-tenancy; Aviz Service Gateway, which manages tenant connectivity to the internet and other data centers via BGP/EVPN and VXLAN; AI Parser, which provides observability into token-level inference usage and latency; and AI NOC, an agentic AI operations platform providing customizable SRE skills.
In return, hosted·ai provides the commercial and compute orchestration layer necessary to turn hardware into a service. The hosted·ai platform includes a GPU Hypervisor capable of sharing a single GPU among up to ten isolated customers running unmodified CUDA workloads. It also manages diverse consumption models—ranging from long-term leases to per-token billing—and automates the associated billing and payouts. This integration effectively extends the hosted·ai platform from pure compute orchestration into the networking domain, creating a foundation for bare-metal AI infrastructure, Kubernetes environments, and large-scale AI factories.
Addressing Neocloud Multi-Tenancy and Observability
As neocloud operators attempt to maximize the utility of every GPU, the complexity of sharing hardware across multiple customers increases the pressure on the underlying network fabric. The collaboration seeks to mitigate the manual work and troubleshooting delays caused by fragmented workflows. By implementing Aviz’s networking software, operators can manage tenant isolation and service performance through a single interface. This is particularly critical for models that sell capacity by the token, where understanding the relationship between network throughput and inference latency is vital for accurate billing and service level agreements.
The inclusion of multi-vendor support through Aviz suggests a strategic move to prevent vendor lock-in, allowing neoclouds to maintain a consistent operating experience even when utilizing different hardware architectures across various locations. For operators, the goal is to accelerate the path from initial GPU cluster deployment to full-scale monetization. By automating the provisioning of networking and compute, the partnership aims to provide a more complete infrastructure stack that supports the rapid scaling required by enterprises and developers running inference workloads on the growing network of specialized AI clouds.
Key Takeaways
- The partnership integrates Aviz’s AI factory portfolio, including ONES AI and AI Parser, with the hosted·ai GPU commerce platform.
- hosted·ai’s GPU Hypervisor enables the sharing of a single GPU among up to ten isolated customers running unmodified CUDA workloads.
- The integration supports diverse networking environments, including NVIDIA Spectrum X Ethernet, InfiniBand, and NVIDIA BlueField.
TechInsyte's Take
In our view, this partnership highlights a critical realization in the AI infrastructure market: owning GPUs is no longer the primary hurdle; managing the complex, multi-tenant networking required to monetize them is. As neoclouds move away from simple hardware rentals toward sophisticated "GPU-as-a-Service" models, the ability to provide granular, token-level observability becomes a competitive necessity rather than a luxury. By linking Aviz’s networking automation with hosted·ai’s commercial orchestration, these companies are attempting to solve the "fragmentation problem" that plagues many emerging AI providers. If successful, this unified approach could lower the barrier to entry for smaller operators looking to compete with hyperscalers by offering highly specialized, automated, and observable AI compute environments. This signals a broader industry trend toward software-defined, highly automated AI factories where the network is as programmable as the compute.
Questions & Answers
How does this partnership impact the way neoclouds bill their customers?
The integration enables more sophisticated consumption models by linking network performance with commercial metrics. Specifically, the use of Aviz’s AI Parser allows operators to gain token-level observability into inference usage, latency, and throughput, which supports hosted·ai’s ability to meter and bill capacity by the token.
What technical capabilities are provided to ensure tenant isolation in these AI environments?
Isolation is managed through a multi-layered approach. The hosted·ai GPU Hypervisor allows up to ten customers to share a single GPU while remaining isolated, while the Aviz Service Gateway utilizes BGP/EVPN routing, VRF/VXLAN termination, and stateful L3/L4 policy to manage secure, software-defined service edges for isolated tenants.
Does this integration limit operators to specific hardware vendors?
No. While the integration supports NVIDIA-specific technologies like Spectrum X Ethernet, InfiniBand, and BlueField, Aviz brings a multi-vendor networking approach to the partnership. This is intended to give neocloud operators the ability to maintain a consistent operating experience across different architectures and locations.
What is the primary operational problem this collaboration intends to solve?
The partnership targets the inefficiency of "disconnected compute, networking, and monitoring workflows." By integrating these layers, the companies aim to reduce the manual work, complicated troubleshooting, and slow service delivery that currently occur when operators manage AI infrastructure using fragmented tools.
Source: Aviz Networks
