Netflix Now Runs Single Containers on Dedicated VMs

Netflix's container platform now isolates single tasks on their own virtual machines, a major architectural shift for enhanced workload security.

Netflix has made a significant architectural change to its in-house container management platform, Titus. According to a report from InfoQ, the company introduced a new “single-tenant” mode that fundamentally alters how it isolates containerized workloads. In this model, each individual container task is run on its own dedicated, right-sized Amazon EC2 virtual machine. This is a major departure from the conventional multi-tenant approach, where numerous containers from different applications share the resources of a single, larger host machine. For Netflix's application developers, the change is designed to be seamless; the company’s platform team continues to manage and patch the underlying hosts, preserving the existing API and deployment workflows developers are used to.

The new mechanism provides a much stronger security boundary between applications. In a typical container setup, multiple containers share the same operating system kernel on a host machine. While technologies like namespaces and cgroups provide a degree of isolation, a sophisticated attacker who compromises one container could potentially exploit a kernel vulnerability to “escape” the container and gain access to the host and other containers running on it. By dedicating an entire virtual machine to a single container task, Netflix is now using the hypervisor—the technology that creates and manages VMs—as the primary isolation layer. This hardware-level separation is widely considered a more robust and secure boundary than the software-based isolation used within a single operating system, effectively eliminating the risk of a container escape affecting neighboring workloads.

This decision by Netflix reflects a broader industry trend toward prioritizing security, even at the cost of some resource efficiency. While running one container per VM is less dense and potentially more expensive than packing many containers onto a single host, it simplifies the security model dramatically. This approach is not entirely new; technologies like AWS Fargate and projects like Kata Containers have explored similar concepts of providing stronger isolation for container workloads. However, Netflix’s implementation using standard, right-sized EC2 instances is a pragmatic endorsement of using well-understood, battle-tested virtualization technology to solve modern container security challenges. It represents a calculated trade-off, favoring ironclad security and operational simplicity over maximizing server utilization.

For CTOs, security professionals, and infrastructure teams, Netflix’s move serves as a powerful validation of using VM-level isolation for high-stakes containerized applications. This pattern is particularly relevant for services that handle sensitive data, run untrusted code, or require strict compliance with security regulations. It suggests that the future of container orchestration is not a one-size-fits-all model but a flexible approach that allows teams to select the appropriate level of isolation based on a workload's risk profile. The key takeaway is that for the most critical applications, the simplicity and strength of hypervisor-based isolation may outweigh the efficiency gains of traditional multi-tenant container hosting. The industry will be watching to see if other major technology companies adopt similar patterns and how cloud providers evolve their services to make this high-security model more cost-effective and accessible.

For engineering and security teams, this move validates a "VM-per-container" model as a viable strategy for high-security workloads. It offers a strong alternative to complex in-kernel isolation mechanisms, trading some density for a much harder security boundary that is simpler to reason about and manage.

This architecture signals a trend where companies prioritize security over maximum resource density, especially for critical applications. While potentially increasing cloud costs, it drastically reduces the risk of container-escape vulnerabilities affecting multiple applications, lowering the potential cost of a security breach.