Artificial intelligence news.
Straight from the original announcements.

Modal joins gVisor maintainers as Google donates project to CNCF

November 11, 2026By the AI News Bites Editorial TeamSource: Modal
Modal joins gVisor maintainers as Google donates project to CNCF
Image: Modal

Modal is joining gVisor’s maintainers as Google donates the container runtime, including its name and trademarks, to the Cloud Native Computing Foundation. The move places Modal alongside Ant Group and Tines, while OpenAI, Tencent and NVIDIA will continue contributing to the project.

Key points
  • Modal will join gVisor’s maintainers under a new maintainer led governance model.
  • Google is donating gVisor, its name and trademarks, to the CNCF.
  • Modal says it launches upwards of 40 million gVisor containers a day.
  • Modal has contributed to gVisor since 2023.
  • The company has upstreamed GPU, RDMA, snapshot and performance improvements.

A maintainer led future for gVisor

The announcement was made during KubeCon in Salt Lake City, according to Modal. With the project moving to a maintainer led governance model, Modal says it has committed to increasing its engineering work on gVisor.

Modal has used gVisor as its major container runtime since the company began. The infrastructure provider says it launches upwards of 40 million gVisor containers a day, covering GPU inference, multi node training, AI coding agents and sandboxes.

Modal has contributed to gVisor since 2023 and says it has significantly increased the amount of code it sends upstream. The company also built a team around gVisor this year to support its customers’ requirements.

Why Modal uses gVisor for AI workloads

Ordinary Linux containers share the host kernel. Modal explains that a kernel bug or vulnerability can therefore allow code to escape a container and compromise the machine. gVisor adds a separate application kernel, called the Sentry, between the container and the host.

The Sentry implements the Linux system call interface and asks the host for a small, filtered set of operations. gVisor also uses namespaces, cgroups, pivot_root and a tight seccomp filter as additional layers of protection. According to Modal, an attacker would need to break the Sentry and then defeat those hardening measures.

The runtime does not require hardware virtualization, so Modal says it can run on cloud virtual machines where nested virtualization is unavailable or slow, as well as bare metal GPU servers. Modal says gVisor provides a real security boundary between tenants while supporting shared infrastructure.

The company says gVisor can boot a sandbox in about 25 ms with 12 MB of memory overhead. Its checkpoint and restore capabilities can capture a sandbox’s processes, files and sockets, as well as GPU memory through cuda-checkpoint, and restore it on another machine.

GPU, networking and performance work

Modal’s upstream contributions include improvements to nvproxy, the gVisor component that allows sandboxed processes to use NVIDIA GPUs through a proxy that validates the driver’s ioctl interface. The work covers newer NVIDIA drivers and GPUs, including Blackwell and the RTX PRO 6000, as well as multi GPU workloads, CUDA MPS, Vulkan, headless OpenGL, video encoding and decoding, monitoring, profiling and debugging.

Modal also added rdmaproxy for RDMA support in gVisor. The feature supports NVIDIA ConnectX with InfiniBand and RoCE, as well as AWS EFA, allowing multi node training jobs to exchange gradients inside the sandbox. Modal says NCCL all reduce benchmarks reached about 99% of native bus bandwidth.

Other work has focused on snapshots, storage and memory performance. Modal says about 670k gVisor containers are restored from a snapshot each day. In one PyTorch DataLoader benchmark, recycling huge pages increased throughput from 2.9 to 22 GB/s while keeping memory use flat. Optimizations to gVisor tmpfs made SQLite 2.7x faster, according to the company.

For users, the governance change could bring broader participation around a runtime designed for shared AI infrastructure. Modal expects the CNCF move to help expand the number of providers offering general purpose gVisor powered compute, while its own maintainer role gives the company a formal position in the project’s development.

Related news

Syntys Starts Construction of 20MW Data Center in Qatar
Chips and infrastructureNovember 12, 2026, 13:11

Syntys Starts Construction of 20MW Data Center in Qatar

Syntys has broken ground on a 20-megawatt data center development in Qatar, adding capacity for hyperscale, cloud, AI and advanced computing deployments. The facility is…

NVIDIA Adds Runtime AI Inventory to Its GPU Platform Tooling
For developersNovember 11, 2026, 18:00

NVIDIA Adds Runtime AI Inventory to Its GPU Platform Tooling

NVIDIA AI Cluster Runtime now includes a qualified AI bill of materials component for tracking supported AI workloads running on Kubernetes. The integration connects…

NVIDIA Gives OpenShell and AI Infrastructure Tools to Open Source
For developersNovember 10, 2026, 17:00

NVIDIA Gives OpenShell and AI Infrastructure Tools to Open Source

NVIDIA is contributing its OpenShell runtime for securing autonomous agents to the Cloud Native Computing Foundation, alongside work on Agent Substrate and new AI…