DeepSeek-R1 is AI's Sputnik moment
In recent years, artificial intelligence has evolved from a futuristic vision to an omnipresent reality. Whether in medicine, transportation, entertainment, or economics, AI systems are fundamentally changing the way we live and work. This rapid development has also sparked a fundamental debate: Should AI models and their underlying technologies be freely accessible (open source) or remain under proprietary control (closed source)?
This question is much more than a technical detail. It touches on fundamental aspects of our digital future: Who controls the power of AI? Who benefits from its advantages? And how can we ensure that AI systems are used safely, ethically and for the benefit of all?
On the one hand, tech giants such as OpenAI, Anthropic and Google are developing their most advanced AI models behind closed doors and offering them as commercial products. On the other hand, a vibrant open-source community is growing, making impressive progress and developing freely available alternatives, such as the Llama model from Meta, the Mistral model from the French startup of the same name, or the DeepSeek models from the Chinese startup.
In this article, we examine the strengths and weaknesses of both approaches and venture a prediction for the future. The central question is: open source vs. closed source in the world of artificial intelligence – who will win the race?
The Closed-Source Perspective: Control, Security and Commercial Interests
Today, several tech companies are developing leading closed models. But open source is quickly closing the gap. Last year, Llama 2 was only comparable to an older generation of models behind the frontier. This year, Llama 3 is competitive with the most advanced models and leading in some areas. Starting next year, we expect future Llama models to become the most advanced in the industry. But even before that, Llama is already leading on openness, modifiability, and cost efficiency.
Proponents of closed AI systems argue primarily on three key points: security, quality, and economic sustainability.
Security concerns: Companies such as OpenAI and Anthropic emphasize the potential risks that could arise from the uncontrolled proliferation of advanced AI models. They argue that powerful AI systems could be misused to spread disinformation, carry out cyberattacks, or develop other harmful applications. By controlling access to their models, they can implement security measures and monitor who is using their technology and for what.
At the end of 2023, a large number of scientists led by Elon Musk had come together and issued an urgent appeal stating: “Powerful AI systems should not be developed until we are sure that their impact will be positive and their risks manageable.”
Quality and resources: Developing state-of-the-art AI models requires enormous computing power, data and expertise. Companies like Google DeepMind and Microsoft have the financial resources to invest in advanced research and recruit top talent. The closed nature of their development enables them to protect their competitive advantages and continuously invest in improvements. Nevertheless, it should be noted that, according to the benchmarks, small open-source models can now keep up well with closed-source models.
Economic sustainability: The development and operation of advanced AI systems involves enormous costs. Training a large language model alone can cost several hundred million dollars. By commercializing their technologies, companies can cover these costs and invest in future research. Demis Hassabis, CEO of Google DeepMind, therefore emphasized the need for sustainable business models.
The Open-Source Perspective: Transparency, Democratization and Collective Innovation
It’s as open as you can be,” Mensch said. “We share the weights, we share the inference, we share a lot of findings around how we built it. There’s obviously some trade secrets that we keep, because how we bring our core value to work with customers.
Proponents of open-source AI, on the other hand, emphasize the advantages of transparency, accessibility and distributed innovation.
Transparency and trust: Open-source models allow insight into their code and development, which builds trust and enables independent reviews. Researchers can identify and fix potential problems before they become serious security risks. However, open access to the code alone is not enough: only by publishing the so-called weights of an AI model is true traceability possible. These weights contain all the trained parameters that determine how a model arrives at its decisions. Without this data, crucial aspects of the AI – in particular, its behavior and potential biases – remain opaque and not fully comprehensible. Only the disclosure of these weights allows independent researchers and the public to analyze AI models in depth, discover vulnerabilities, identify and correct biases, and ensure trustworthy, safe use.
Democratization and accessibility: Open-source AI makes advanced technology accessible to a wider audience, including small businesses, educational institutions, and developers in regions with limited resources. This fosters innovation and applications tailored to specific local needs. Instead of relying on a few tech giants, diverse communities can develop AI tools that align with their own values and priorities.
Collective innovation: The open-source movement has proven that distributed collaboration can lead to remarkable innovations. Linux, Wikipedia and numerous other open-source projects demonstrate the power of collective intelligence. In the world of AI, the open-source community has made impressive strides, as evidenced by the rapid advancement of models like Llama 3 and Falcon.
The Current State: An Increasingly Dynamic Competition
Competition between open- and closed-source AI has intensified in recent years, with notable developments on both sides.
Closed-source successes: Models such as OpenAI's GPT-4.5, Anthropic's Claude Sonnet 3.7 and Google DeepMind's Gemini 2.0 have demonstrated impressive capabilities in areas such as language comprehension, reasoning and creative writing. These companies have attracted significant investment – OpenAI, for example, received a $10 billion investment from Microsoft – and are increasingly developing business models built on their AI services. Recently, OpenAI attracted $500 million in funding that they are investing in the Stargate project.
Open-source progress and market disruption: At the same time, the open-source community has made significant progress. Meta's Llama-3 model, released under a commercially friendly license, has demonstrated comparable performance to proprietary models in benchmarks. Mistral AI from France has developed models with significantly fewer resources that can compete with much larger models in certain tasks.
One particularly impressive example of the disruptive power of open-source AI was the publication of DeepSeek r1 by the Chinese AI company DeepSeek. This open model demonstrated surprisingly advanced capabilities at significantly lower hardware requirements than comparable commercial models. The announcement triggered a shockwave across financial markets and led to significant price drops for chip manufacturers such as NVIDIA, AMD and Intel. Investors feared that the efficiency gains of models like DeepSeek r1 could reduce the future demand for expensive high-performance hardware.
News around the world: DeepSeek r1 shocked the stock market
This “DeepSeek shock” highlighted a central argument of the open-source movement: through collaborative development and the open exchange of knowledge, efficiency improvements can be achieved faster, potentially challenging the market dominance of established players. It also showed that open-source models can be disruptive not only technologically but also economically.
Initiatives such as Hugging Face have created platforms that facilitate access to and collaboration on open-source models, further accelerating this development.
Hybrid approaches: Interestingly, the boundaries between open and closed source are increasingly blurring. Some companies are taking a hybrid approach, making certain models or components open while keeping their most advanced technologies private. Meta has open-sourced its Llama models but kept the training data and certain methods to itself. Google has open-sourced TensorFlow as an open-source framework but kept its most advanced AI systems like Gemini under wraps.
Key Factors for the Future
Several decisive factors will shape future competition between open- and closed-source AI:
Computing resources: Training advanced AI models requires enormous computing power. Currently, large tech companies have a significant advantage here. However, new developments such as more efficient training methods and specialized hardware could reduce this advantage.
Regulatory developments: Governments worldwide are developing frameworks for AI. These could either favor open-source approaches by demanding transparency and control, or favor closed-source models by imposing strict security requirements that are difficult for smaller players to meet. The EU AI Act, for example, places special demands on “high-risk AI systems”, which could have different effects on open and closed source developers.
Trust and acceptance: Ultimately, the acceptance of AI systems by users, companies and institutions will be crucial. Here, open-source models could have advantages due to their transparency, while closed-source providers may be able to score points with stronger security guarantees and support services.
Specialization vs. general AI: The AI landscape is increasingly developing in the direction of specialized models for certain application areas, alongside the large general models. This specialization could open up new opportunities for open-source projects that can focus on specific niches.
Initial Findings: A Multifaceted Competition With An Unclear Outcome
The analysis so far shows that the competition between open- and closed-source AI is more multifaceted than it might appear at first glance. It is not a simple either/or, but a complex ecosystem with different strengths, weaknesses and areas of application.
In certain areas, closed-source models currently appear to have the upper hand – especially for the most powerful general AI systems, which require significant resources for training and security measures. In other areas, such as specialized applications and local implementations, open source could increasingly offer advantages.
It appears that both approaches will coexist and influence each other, similar to what we have seen in other technology areas. Linux (open source) and Windows (proprietary) have coexisted in the operating system market for decades, with different strengths in different application areas.
Conclusion
The competition between open-source and closed-source AI is more than a technological contest – it is a battle for the future of one of the most transformative technologies of our time. After extensive analysis, it seems unlikely that one approach will completely displace the other. Rather, developments point to a complex ecosystem in which both models coexist and enrich each other.
Closed-source systems are likely to continue to push the boundaries of what is technically possible, with significant resources for research and development and strict security protocols. They will be used particularly in highly sensitive areas such as healthcare, financial services and critical infrastructure, where reliability and liability are crucial.
Open-source AI, on the other hand, is expected to excel in the breadth and diversity of its applications. It will democratize innovation, enable localized and specialized solutions, and serve as an important corrective and control mechanism for closed systems. The transparency of open-source models will be crucial to building trust and enabling societal oversight.
Society as a whole could ultimately be the real “winner” in this competition – if we manage to use the strengths of both approaches while offsetting their respective weaknesses. The challenge is to create a regulatory framework that both promotes innovation and ensures security, and to develop ethical principles that apply to both commercial and collaborative AI development.
In this sense, the answer to the initial question “Open source vs. closed source: Who will win?” is perhaps: Both – but only if they coexist in a balanced ecosystem characterized by responsible governance and societal engagement. The real challenge is not to choose a winner, but to ensure that AI technology as a whole is developed and used for the benefit of humanity – regardless of whether it comes from open or closed sources.
Kai-Fu Lee (Co-Chair of the World Economic Forum's Artificial Intelligence Council) however sees the triumph of open source as inevitable, questioning the long-term viability of closed AI models. "Sam Altman is probably not sleeping well", he said recently. He may be right - but we will see.
—
Ready for more content from Kim Isenberg? Subscribe to FF Daily for free!
