The Cloud Native Computing Foundation (CNCF) is anticipating the “inevitable” emergence of an open source alternative to Nvidia’s dominant Cuda parallel computing platform used to program graphics processing units (GPUs) as the industry seeks to avoid supplier lock-in for artificial intelligence (AI) workloads.
Speaking at a press conference at KubeCon + CloudNativeCon China 2025 in Hong Kong last week, the CNCF’s chief technology officer, Chris Aniszczyk, addressed enterprise concerns and pointed to the historical precedent of open source providing standardised alternatives.
“People don’t generally like to depend on a single vendor or entity for anything,” Aniszczyk said in response to a question from Computer Weekly. “Open source is all about bringing together different organisations working on a standardised way to do things. I think that’s an inevitable thing.”
While acknowledging that a prevalent open source alternative to Cuda does not exist today, Aniszczyk was confident that the community will coalesce around one. “Like we came together to work on compute with Kubernetes, we’re going to do the same to ensure you can run AI workloads on different GPUs. It doesn’t happen overnight, and hopefully you’ll see CNCF projects that help with some aspects of that,” he said.
Aniszczyk’s comments came as he outlined the CNCF’s vision for the next decade. He noted that while the first 10 years of the CNCF were about standardising containerised workloads with projects like Kubernetes, the next 10 years will be a “continuation of that effort, especially focused on AI”.
He highlighted that AI workloads, which are primarily GPU-intensive, still require the same robust infrastructure that the cloud-native ecosystem provides. “Every workload needs to be secured, observed and be able to scale out,” he said. “All this technology that we’ve developed at CNCF is very applicable to AI workloads.”
Like we came together to work on compute with Kubernetes, we’re going to do the same to ensure you can run AI workloads on different GPUs
Chris Aniszczyk, Cloud Native Computing Foundation
During the press conference, Aniszczyk also pointed out the significant breakthroughs within the CNCF’s portfolio of projects. He singled out OpenTelemetry, the second-largest CNCF project after Kubernetes, for its success in standardising observability across metrics, logging, tracing, and now, profiling data.
“I consider that a breakthrough in the way that any time you get a huge portion of the industry globally to move to a certain system in a standardised way, it truly revolutionises how technology is adopted,” he said, noting that more than 50 major observability suppliers now adhere to the OpenTelemetry specification.
He also praised the “hard technical work” done on etcd, the distributed consensus mechanism at the core of Kubernetes. Collaborative efforts from Google, Alibaba, Huawei and others have significantly improved etcd’s scalability, allowing Kubernetes to manage far larger clusters.
The press conference also touched on the recent intent by the OpenInfra Foundation to join the Linux Foundation, the CNCF’s parent organisation. Jonathan Bryce, executive director of the OpenInfra Foundation, pointed out the long-standing technical collaboration between the OpenStack and Kubernetes communities. Both groups see the move as an opportunity to combine efforts in global outreach and community building, particularly in markets such as China, South America and Africa.
Addressing the geopolitical tensions between the US and China, Jim Zemlin, executive director of the Linux Foundation, reassured the community that open source collaboration remains unaffected.
“Open source as a freely available public good is exempt from export control rules and therefore is something that we still collaborate on and want to continue to collaborate on,” he said. He drew a parallel to the early days of CNCF, when fierce competitors like Amazon, Microsoft and Google collaborated on open source projects.
As the worlds of cloud-native and AI continue to merge, the CNCF is also addressing the complexities of licensing for AI models. Zemlin and Aniszczyk highlighted the new Open Model Definition and Weights (OpenMDW) licence from the Linux Foundation, designed to bring clarity and predictability to the components of open source AI. The framework aims to define the levels of openness for the various artifacts in an AI model, including data and weights, which differ from traditional source code.
