Interview with the founder of the Linux Foundation: AI foundation models are destined to be fully open-source, and the battlefields lie only in the application end.

The debate between open source and closed source has always been a core battleground in the early stages of new technology development, and the AI era is no exception.

On one side of this war are the "high - wall proponents" led by OpenAI and Anthropic. They hold the most powerful models and advocate guiding the development of AI "responsibly" through a centralized, cautious, and closed - system, using this to build the "moat" for their business empires. On the other side are the "open - source proponents" represented by Meta, Mistral, and the emerging Chinese company DeepSeek. They firmly believe that the only sustainable path for technological progress is the open - source model that is decentralized, transparent, and collaboratively developed by the global community. They also think this is the fundamental guarantee to prevent technology from being monopolized by a few giants.

This debate has become extremely intense in the past year due to a key variable - for the first time, open - source models have truly reached the "throne" of closed - source models in terms of performance.

Among them, the rise of DeepSeek is particularly crucial. It has proven to the world that a Chinese company with relatively limited resources, through excellent technological innovation, is fully capable of creating products comparable to GPT - 4 level models and has brought the API cost to an unprecedented low.

How should we understand what is happening? Is this irresistible open - source wave just a short - lived "challenger's stance" or an irreversible "historical law"?

No one is more suitable to answer these questions than Jim Zemlin, the executive director of the Linux Foundation. As a witness, shaper, and leader of the global open - source movement in the past three decades, Zemlin's career is itself an epic of "open source defeating closed source". He led Linux to compete with the Microsoft Empire in the "final battle" of operating systems. He also promoted the containerization standard worldwide through Kubernetes in the cloud - computing era.

For him, today's "open - vs - closed" debate in the AI field is not something new, but a classic replay of history on a different technological foundation.

In June 2025, Jim Zemlin had an in - depth conversation with Tencent Technology. In the conversation, Jim Zemlin, from his unique historical perspective and profound insights into the industry, systematically answered all the core questions about open source in the AI era. From the uniqueness and new challenges of AI open source, to the underlying reasons for the success of "rising stars" like DeepSeek; from the hidden business logic behind open source, to its essential advantage of attracting top - notch talents.

This is not only a discussion about technological routes but also a review of innovation laws and business philosophy. This conversation itself is a comprehensive sorting out of all the core questions about AI open source.

The Uniqueness and Challenges of Open Source in the AI Era

Tencent Technology: You once described yourself as a builder of technological bridges. So in the AI era, what kind of bridge does the Linux Foundation want to build?

Jim Zemlin: We hope to provide people with channels to access the "building blocks" for creating AI applications and technologies. You can think of us more as building a community where people can innovate together.

If we can achieve this, the significance of this bridge lies in accelerating the ongoing basic AI work and transforming it into practical applications that can touch our daily lives.

Tencent Technology: So, the main goal of the Linux Foundation is to promote the transformation of AI from the basic model level to the application level.

Jim Zemlin: Yes, we are an "innovation accelerator". The more freely accessible the basic building blocks for creating AI applications, AI agents, and tools that can empower drug discovery and enable autonomous driving are, the faster innovation results will be produced. We have also witnessed this in the cloud - computing field, where technologies like Kubernetes and OpenStack were used to build the cloud - computing industry.

(With them) The speed of building cloud services is much faster than if each cloud service provider built all the technologies from scratch. As a result, we can use cloud services at a lower cost and have more innovative tools. Large companies like Tencent can also grow faster because they can freely use many basic technologies that their customers may not care about.

Tencent Technology: In recent years, especially after the emergence of ChatGPT, you and the Linux Foundation have launched many initiatives. Are they aimed at making AI more free and shared? In your opinion, what tasks are essential to ensure the "democratization of artificial intelligence"?

Jim Zemlin: Obviously, it is crucial to enable people to access technology so that more people can create more solutions.

Let me say it again. All value is created at the application layer. And whether you are a small startup or a large company, if everyone can access the basic building blocks for building AI solutions to create those applications, it is better for consumers because they can get cool things. It is also better for society as we can have more scientific research. This is exactly what we hope to see in the AI field.

So in some aspects, this is not different from the past, but some things are different. For example, a different legal framework may need to be established around data sharing. These are exactly the things we want to promote.

Tencent Technology: Then do you think there are any differences in the rules of open source in the artificial - intelligence era compared with the cloud - computing service era, the Microsoft era, or the PC interface era?

Jim Zemlin: This is a good question.

Traditionally, open source mainly focused on the source code itself. That is, the software used to create personal computer operating systems or a platform that can schedule cloud services. In those cases, people jointly developed Linux or Kubernetes to build solutions for running the Internet and creating cloud services. So at that time, the main output (artifact) was the software code itself.

In the field of artificial intelligence, you need more components to create a large - language model and AI applications. You need very good data, you need model weights, and you need pre - training and post - training. You need more than just code.

Actually, many models also have a superset that includes various different components, and we also hope to promote the sharing of these components.

Tencent Technology: You mentioned that the levels of open source in the AI model are very different, including the data layer, the weight layer, and the code layer. Yes, there is also the tool layer. I think in the work done by the Linux Foundation, such as Open MFW (Model Open Framework), you have defined open source at different levels. Can you specifically explain this new open - source framework in the AI era?

Jim Zemlin: Open MFW is a license that covers all work related to models. At that time, a clear legal framework was needed to cover all the outputs (artifacts) of models, and that's what Open MFW does.

People need to consciously decide how to share their intellectual property or not. One thing we often discuss at LF is that we hope people can share what they want to share under the terms they want, and also retain what they want to retain under these terms.

What Open MDW does is to provide a clear framework on how you can share the technology and how consumers can use the technology under the Open MDW license terms.

This is a very permissive license. It clearly indicates whether the model is open source, whether it is provided "as is", and whether there are no additional requirements. You can use it as you like.

If you go to the Linux Foundation website and look at our definition of the Model Open Framework, it will clearly show you the different levels of open source, such as the data layer, the training layer, etc., including all the different components covered by the model.

Thanks to DeepSeek, the Open - Source Movement is Stronger in the AI Era

Tencent Technology: Compared with the operating - system or cloud - computing era, do you think the open - source movement has strengthened or weakened in recent years?

Jim Zemlin: I think it has become stronger.

If you look back at the major technological trends - mobile computing, cloud computing, and now artificial intelligence - open source has actually been at the forefront of all these trends. And it has successfully accelerated industrial applications in each era.

I can give you some examples. In the field of mobile computing, Android - which was originally essentially an open - source platform of Google, but later a huge community was built around it.

Now Android is used in various mobile phones of many different device manufacturers. It is used in automotive systems, and its code is used in many places. There is no doubt that in the mobile field, open source has created most of the market share and accelerated innovation in this field.

In the cloud - computing field, OpenStack, Kubernetes, these technologies have triggered new cloud services and have indeed reduced the cost and threshold of using cloud - computing resources.

So we are seeing the same situation in the AI field. I think a huge breakthrough is that DeepSeek has proven that a small Chinese company can open - source a cutting - edge model with very, very good performance, and it is completely free. This has completely changed the competitive landscape of the industry.

Now you can see this happening almost every week or even every day, with new models emerging continuously. I think people are realizing that the AI infrastructure at the model level will be free and will perform quite well.

Tencent Technology: How do you explain the reason for DeepSeek's success?

Jim Zemlin: They came up with new technologies and used knowledge distillation as a method to improve performance.

These are real innovations. And they made those innovations public for everyone to see. I think many companies are adopting these technologies.

So, sometimes when you are a small startup, you have to be innovative because your resources are limited.

This is no different in China or the United States. This is exactly the story of Silicon Valley - small companies with limited resources think about the world in a new way and come up with more innovative methods to solve similar problems.

Tencent Technology: In the era of operating - system platforms, Linux was a small company, while Microsoft was a giant. But later, large companies gradually joined the open - source camp. In the mobile - phone era, Google, a giant, took the lead in joining the open - source movement and achieved success. The same was true in the cloud era, where Amazon directly joined the open - source process. Open source seems to be becoming an increasingly accepted consensus among large companies.

But in the AI era, although Meta has joined the open - source camp, overall, the form where giants take the lead in joining open source and become the main force has disappeared. Why is this?

Jim Zemlin: Open source may not always be the first to enter the game, but it usually has the last laugh.

One of the difficulties in building cutting - edge models is that it requires a large amount of data, a large amount of computing resources, and GPUs. These things are very expensive. It also takes a lot of courage to open source.

And smaller organizations do not have the ability to obtain these resources. This makes large companies think they can maintain their advantages without open - sourcing.

That's why DeepSeek is so surprising. They built a model with extremely high performance with limited resources. This shocked everyone, like 'Oh my god, a small company can actually do this'.

They also showed that other organizations can effectively open - source their models, build communities around these models, co - develop these models with other organizations, or let users adopt these models.

The results show that this approach helps their own companies and promotes competition. That's why open source is so powerful.

Tencent Technology: You published a blog in February this year titled "AI models have no moat" (We have no moat), which is actually an extended analysis of Google's (memo about the threat of open source). In it, you think that in fact, there is no moat at all. Is it DeepSeek that gave you this impression?

Jim Zemlin: Businessmen are very smart. They came up with a term called "moat" to show their competitive advantages.

I should point out that Sam Altman of OpenAI has said that he will build and release an open - source model. So let's wait and see what happens next.

This shows that at the lower levels of the basic model technology stack, it doesn't make sense to work alone. Since open source has become such a powerful movement, you have to look for places to create value in other fields.

One reason I respect Anthropic and OpenAI is that they have a large number of consumer users. In the case of Anthropic, they have a large number of enterprise users because they have built an excellent user experience. So that's where they create a lot of value.

ChatGPT now has a "memory" of how users interact with the model. This is very valuable. It's like a moat because if you switch to a new model, what will happen to all my memories?

I think you will see more similar things, so I think AI companies are now moving towards higher - level innovation. These are different from the underlying code and the model itself. They are more like the uniqueness of a specific product or service.

Tencent Technology: So do you think that no company will be able to maintain a continuous competitive advantage in the basic model, which is the underlying technology stack?

Jim Zemlin: I think over time, the main competition is unlikely to remain at this level.

As the lower levels of the technology stack become freely available in open source, value is shifting to higher levels.

In the Future, Basic Models Will Be Completely Open - Source, and Value Will Shift to Higher Levels

Tencent Technology: In the basic model layer, you said that there are several different levels of things that should be open - sourced, not just can be open - sourced. Such as data and code. But in the application layer, you said that there is also a possibility of open source for applications, right?

Jim Zemlin: I think infrastructure and software for building AI agents (agent - tec) will be open source.

When I talk about the application layer, open source has traditionally not been good at quickly creating a specialized user experience or meeting very specific customer needs, which companies can provide to their customers. So at that level, we see companies finding their unique value because they know their customers better than anyone else.

For example, do people care if TikTok runs on Kubernetes? No, they care about making videos. Do people care if there is open - source technology in self - driving cars? No, they just want to go somewhere. So I think that's where value is created.

But they can meet those customer needs because a lot of the underlying infrastructure software is very low - cost due to open source. And the closer the software is to the customer, the less open - source it tends to be.

Tencent Technology: Okay, so now, actually, I think ultimately, the basic model will become a completely open - source arena. Will it be like this?

Jim Zemlin: Yes, what is truly closed - source is the application layer.

For example, in the distant future, there will be an Agent library. These Agents are extremely professional in a specific skill and are trained in a unique and excellent way. Maybe these Agents will be closed - source, but they are built using open - source technology. That's the key.

There is a Hidden Business Logic in Open Source

Tencent Technology: I've observed that the top three companies in the AI field - Google, OpenAI, and Anthropic - their most advanced models are basically closed - source.

Companies that open - source their models, such as Meta, are in a sense catch - up players. They are a bit behind. So, open - sourcing actually doesn't cause them any business losses because even if they keep their models closed - source, they can't actually profit from them due to the inferior quality of the models.

I think this may be a logic of participation and development behind the open - source movement. Do you think this is a common phenomenon?

Jim Zemlin: Let's take Kubernetes as an example.

I remember around 2014, I received calls from some people at Google when they were just starting Google Cloud. At that time, Amazon was the far - leading market leader in the cloud - computing field. And what Google wanted to do was to change the way of building cloud applications.

Amazon created EC2, a virtual - machine technology that people used to build cloud infrastructure. So, at that time, most clouds were running on virtual machines. Google didn't use virtual machines; they used containers.

They hoped to create a more level playing field for cloud computing by getting people to adopt the technology they used - containers. In exchange, they open - sourced this incredible technology - Kubernetes, which they had been developing internally at Google for a long time.

In this case, Google was not an existing giant in the cloud - computing field but a challenger. They were not the market leader; they were behind.

By open - sourcing a technology that they understood better than anyone else,