Cloud-native and AI: How open source shapes the future – News

AI and Kubernetes are coming together to reshape the entire stack. What will this mean for the future roles of cloud-native and open-source?

This is the central question that will be on the minds of key leaders from the open-source community when they gather in Atlanta for the annual KubeCon + CloudNativeCon NA event, kicking off on November 10. The past year has seen continued growth in everything from platform engineering techniques to new tools for cloud-native observability, yet AI has been topic number one and will continue to dominate the conversation.

Like ducks on a pond, many enterprises are seeking to appear calm on the surface, yet below water they are furiously paddling away. Even the most well-funded and experienced IT shops are beginning to feel like they are barely keeping up with the AI tsunami. This point was recently made clearer for Jonathan Bryce, executive director of the Cloud Native Computing Foundation and Linux Foundation. In an interview with News for this story, Bryce recalled a meeting with the leadership of one of the most advanced IT organizations in the world, during which they expressed concern about their ability to keep pace with evolving AI solutions.

“The whole time we were talking, they felt like they were behind,” Bryce told News. “The field is moving so quickly that every week there is a new thing. We’re at the point where people are trying a lot of different approaches simultaneously. We’re all actually learning very quickly.”

This feature is part of News Media’s ongoing exploration into the evolution of cloud-native computing, open-source innovation and the future of Kubernetes. Be sure to watch theCUBE’s analyst-led coverage of KubeCon + CloudNativeCon NA from November 11-13. (* Disclosure below.)

Cloud-native protocols for agents

One of the fastest-evolving areas of AI today involves agents, autonomous software to perform key enterprise tasks. The cloud-native world has been an active participant in shaping the future direction of this autonomous technology as interoperable, secure and specialized multi-agent systems take shape.

Two recent moves in particular highlighted the growing role of the open-source community in agentic AI. The first was a decision by Google LLC’s cloud unit to contribute the Agent2Agent Protocol to the Linux Foundation. The protocol, known as A2A, provides features for moving data between agents, thus taking the burden off developers who previously had to write code that enabled cross-agent communication.

The other significant event was an announcement by the Linux Foundation in August that it would take over Solo.io’s open-source Agentgateway project — an AI-native proxy that optimizes connectivity and observability for agentic environments. Another proposed standard for agentic workflows – Agntcy – was accepted by the foundation from a consortium that included Google and Red Hat Inc.

“Agents are capturing people’s attention,” Bryce said. “Where we are right now is the very early stage of coming up with the right frameworks and protocols. When I look at the agent space, I see that these frameworks and projects are open-source.”

Building permissions through OpenFGA

Along with protocols for agentic AI, the cloud-native world has been developing new tools encompassing security and model management. One of the tools on the horizon is Open Fine-Grained Authorization, or OpenFGA. The open-source authorization engine, which sprang from Google’s Zanzibar system for role-based access control, moved into CNCF’s incubation maturity level last month.

OpenFGA is designed to define who can do what within systems, providing authorization checks in milliseconds. The open-source tool is expected to appeal to developers seeking flexibility as organizations build increasingly more complex AI-based workflows.

“We kind of put these permissions into boxes,” Bryce said. “OpenFGA can layer different permissions together. It’s a more complex model, but it is the right kind of model when you think about autonomous AI systems.”

Reasoning models on the rise

Language models themselves are also becoming more of a focus within the cloud-native community as a growing number of enterprises move to adopt open-source AI solutions. A study released by McKinsey & Co. found that more than half of over 700 technology leaders surveyed globally were using open-source AI models, with that number increasing to over 70% within the technology industry itself.

A key moment in the development and acceptance of open-source AI models took place nearly a year ago when Chinese startup DeepSeek Ltd. released the first of several low-cost LLMs.

“That was really the catalyst for the open-source wave that we’ve seen in the AI space over the last year,” Bryce noted. “DeepSeek really kickstarted that.”

What has taken place since has been not only a proliferation of new open-source models, but the development of reasoning capabilities within them. Reasoning models differ from standard LLMs in their ability to “fact-check” responses on-the-fly.

DeepSeek’s R1 model, released last November, was touted as comparing favorably to similar offerings from OpenAI Group PBC. Chinese tech giant Alibaba Holding Group Ltd. claimed that its new family of open-sourced Qwen-3 AI reasoning models launched in April outperformed others on the market.

This activity in new models with reasoning capabilities represents a key opportunity for growth within the open-source world, according to Robert Shaw, director of engineering at Red Hat.

“Reasoning capabilities that have been added to some of the proprietary models are starting to show up in the open-source models,” Shaw noted during a recent interview with theCUBE. “We’re seeing really, really large open-source models be released that have that frontier capability. It’s just the continued improvement in the overall quality of those open-source models, which is what actually unlocks the business use cases and consumer use cases that are powering all this frenzy. I think that’s one really, really great trend.”

Open source for inference platforms

Another trend underway within the open-source community is a move toward open-weight models. “Weights” are the parameters learned during training that are made publicly available, allowing users to download and run them locally.

Because these models are not fully open source, they represent a middle ground between open and closed-source offerings. In March, OpenAI announced that it would launch its first open-weight model since 2019. Three months ago, Amazon Web Services Inc. made OpenAI’s open-weight models available on Amazon Bedrock and Amazon SageMaker JumpStart.

Growing interest in open-source models and early signs of a shift toward making open-weight more readily available are being driven by the reality that AI platforms are not going to run on a single model or inference server. New platforms are beginning to emerge, and this will likely be a blend of proprietary and open architectures.

“Companies are really start think through what their inference platform looks like,” Shaw said. “I’ve talked to a lot of customers, many of them are running on proprietary APIs. I’m starting to see lots of companies plan for the future, as these open-source models get better, as the costs start to rise. What does my platform look like? What is the foundation of the next 20 years of my private LLM cloud?”

AI infrastructure takes shape

Questions such as the ones posed by Shaw are giving rise to the growth of a new set of companies designed to provide answers in the form of infrastructure for the evolving world of AI. One such company is Spacelift Inc., an infrastructure-as-code orchestration platform that enables codeless provisioning for cloud workloads. Enhancements released by the company last month provided an agentic infrastructure deployment model for allocating cloud resources using natural language commands.

The latest open-source release, Spacelift Intent, enables more precise control over cloud infrastructure through the use of agentic provisioning.

“Spacelift will control the capabilities,” said Marcin Wyszynski, co-founder and Chief R&D Officer at Spacelift, in a recent interview with theCUBE. “We’ll have policies that will prevent your LLM from doing something that you would not want to do but, other than that, let your coding agent continue with the magic. And if you want to show your project to someone else and you want to deploy it on AWS, then Intent kicks in and allows you to do this.”

Another key player redefining AI infrastructure orchestration is Vast Data Inc., which recently introduced serverless compute functions and triggers designed to operate directly on incoming data. The approach lets enterprises run inference and indexing workloads in real time across multi-tenant environments.

“The combination of those two things — functions and triggers — allow us to invoke compute in real time as data comes into the system,” said Alon Horev, co-founder and chief technology officer of Vast Data Inc., during an interview with theCUBE. “That allows us to index information but also to trigger applications to get their job done as soon as possible, as fast as possible.”

Another cloud-native infrastructure firm — Backblaze Inc. — provides cloud storage and services built to accommodate developers. The company recently announced a set of enterprise web console and role-based access controls for cloud-native security and simplified management.

Backblaze’s role in facilitating cloud-native development has allowed it to see firsthand how firms are building the plumbing to channel AI implementations. “We’re seeing the entire workflow of what it looks like to build and use AI in the processing pipelines,” explained Gleb Budman, CEO of Backblaze, in conversation with theCUBE. “First, companies need to collect all the data that they want to use, then they want to label it, they want to build the models, they want to then do inferencing off of those models, and then they want to log and track all of the information. So, we’re seeing Backblaze be used in all of those different parts of the pipeline.”

What is clear as the open-source community prepares to gather for KubeCon this month is that the rapidly moving pace of AI is having its own effect on the cloud-native world. Companies are busily building new tools to support an infrastructure that has yet to be fully formed for a technology that is still in its infancy.

Yet there is also a clear tailwind carrying the open-source ecosystem to a new level, where it will have a significant impact on the future direction of enterprise AI.

“We’re talking in completely new paradigms here, and the new paradigms are best built in the open,” Wyszynski told theCUBE. “If people don’t understand how it works, if they can’t see the code, if they can’t play around with it, if they can’t extend what’s happening, then they will not trust it, and they will not adopt it. It’s the open standards that do win.”

(* Disclosure: TheCUBE is a paid media partner for the KubeCon + CloudNativeCon NA event. Neither Red Hat and Google Cloud, the headline and premier sponsors of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or News.)

Image: Microsoft Designer/ News