Vultr joins Agentic AI Foundation as Gold member to push open agent stacks

Why Vultr is joining

According to Vultr, agentic AI is moving from experimentation to production. Agents have to reason, call tools, access data, interact with other agents and operate across distributed environments. The company argues that this takes more than models: it needs infrastructure, orchestration, frameworks and open protocols that work together.

Vultr says the membership extends its commitment to an open, composable AI stack into the agentic layer. Its position is that developers should be able to choose the infrastructure, frameworks, models, tools and protocols that fit their applications, rather than being tied to a single technology stack. The post quotes Manik Surtani of the Agentic AI Foundation, who describes a mission "to bring this vision of open, composable stacks so we can have an open ecosystem, open innovation, we can have freedom, choice, and flexibility."

The layers of an open AI stack

Vultr describes production AI as several layers, each handled by different open technologies and Vultr products.

• Agents and applications: open protocols such as MCP and A2A connect agents with tools, data, services and other agents.
• Model development and inference: PyTorch, vLLM and Ray support training, fine-tuning and inference. Vultr Cloud GPU offers AMD Instinct and NVIDIA GPUs, plus high-performance CPU infrastructure.
• Orchestration and scaling: Kubernetes and CNCF technologies cover orchestration, observability, networking and scaling. Vultr Kubernetes Engine (VKE) is CNCF-certified managed Kubernetes with storage, networking, a container registry, CPUs and GPUs.
• Infrastructure and isolation: Cloud Compute, Bare Metal, VPC networking and storage.
• Heterogeneous compute: AMD EPYC CPUs, AMD Instinct GPUs and NVIDIA GPUs, so developers can match compute to training, inference, orchestration, data processing, application logic and agent execution.

Chris Aniszczyk, CTO of CNCF, spoke with Vultr at AgentCon and MCPCon about how CNCF technologies already support AI infrastructure across inference, service mesh and storage. He said, "Open source gives them optionality." Vultr reads that as a practical benefit: developers can pick the right technology per layer and swap components as requirements change.

Connecting agents to infrastructure

Vultr points out that a single agent task can involve many steps: querying a model, retrieving enterprise data, calling an API or MCP tool, talking to another agent and taking an action. Each step may use a different technology, service or infrastructure resource, which makes interoperability more important.

The post lists AAIF projects that address parts of this layer: MCP, goose, AGENTS.md, agentgateway, A2A and Agent Router. Naseem Al-Naji, co-founder of AgentCat, framed the issue during AgentX as thinking about the entire "agent's experience", since agents increasingly work across multiple tools and systems.

Operations matter as well. Oz Osorio of Stacklok discussed designing agent infrastructure for Kubernetes from the start and building identity and delegated permissions into agents and sub-agents. For enterprises, Vultr frames the questions as: which agent is acting, what can it access, what actions can it perform, and how is it monitored and governed. The company also stresses that GPUs are not the whole picture, because CPUs handle application logic, orchestration, data processing, networking and agent execution.

What it means for developers and enterprises

Vultr compares the goal to earlier open foundations: Linux for computing and Kubernetes for cloud-native orchestration. It says agentic AI needs the same openness at the application layer. The aim, according to the company, is not for every organization to use the same stack, but to let developers choose a model, run it on suitable compute, orchestrate it with Kubernetes, connect it to tools and data and build agents on top, without a single provider controlling every layer.

Vultr says that for developers this means choice across the stack, and for enterprises the flexibility and control to take agentic AI into production on their own terms. The post does not give details on specific commitments, contributions or new products tied to the Gold membership.