Yi Shi, founder of Flashintel, pioneers AI agent software. A computer science expert & e/acc proponent shaping transformative tech.
With AI, innovation isn’t just measured in raw computational power—it’s about creating systems that can reshape markets. The DeepSeek R1 paper has captured attention for its novel approach to training large language models (LLMs), demonstrating advanced reasoning through a streamlined reinforcement learning (RL) process.
Yet, as LLMs become increasingly commoditized, deeper questions arise: How will companies like OpenAI, Anthropic and their peers sustain value? Can closed models build a winning ecosystem reminiscent of the Windows era?
A Deeper Look At DeepSeek R1
DeepSeek R1 represents a breakthrough in model training with a focus on efficiency and emergent reasoning, though information about the model is still coming out and some experts have disputed some of the claims about the technology.
Rather than relying heavily on supervised fine-tuning, DeepSeek R1 introduces a variant called R1-Zero that is trained predominantly using RL. By rewarding the model not just for correct answers but also for a transparent, step-by-step reasoning process, the approach encourages what many call a “chain-of-thought” methodology.
In essence, the model is prompted to “think out loud,” which helps in breaking down complex problems into manageable steps.
After this initial phase, DeepSeek R1 undergoes a concise round of fine-tuning to polish its responses and improve clarity. One of the most exciting aspects of the project is its ability to distill these advanced reasoning capabilities into smaller, more accessible models. This would mean that the sophisticated problem-solving skills developed in the giant model can be effectively transferred to lighter versions, potentially making cutting-edge AI more democratic and available to a wider range of developers.
The Commoditization Of LLMs
The rapid evolution of models like DeepSeek R1 is occurring alongside a broader trend: LLMs are quickly becoming a commodity, as Microsoft and others have pointed out. Today’s market is witnessing an explosion of high-performing language models—both proprietary and open source.
This democratization means that the raw technology is less of a competitive advantage than it once was.
When multiple players can offer comparable performance, the unique value of any single model diminishes. Instead, success now hinges on how these models are applied. The focus is shifting from having a powerful engine to creating value-added services and user experiences that leverage that engine.
Standardized APIs and common training methodologies further lower the barrier to entry, making it easier for companies to switch between providers or integrate multiple models into their workflows.
Accruing Value In A Crowded Market
In this new landscape, traditional model providers like OpenAI and Anthropic are rethinking their strategies. With foundational models quickly becoming interchangeable, the battle for market share is shifting to the application layer. Here’s how these companies are positioning themselves for long-term success:
• Proprietary Data And Fine-Tuning: While the base models are available to everyone, companies that can leverage unique, proprietary data to fine-tune their models stand to gain a competitive edge. Customization tailored to specific industries or tasks can offer superior performance that generic models simply can’t match.
• Integrated Platforms And Services: Rather than simply providing API access, these companies are building full-fledged platforms. OpenAI’s ChatGPT and Anthropic’s Claude, for example, are not just models—they’re part of broader ecosystems that offer analytics, safety features and robust customer support. These integrated solutions are far more attractive to enterprises looking for end-to-end solutions.
• User Experience And Ecosystem Lock-In: The goal isn’t just to power applications but to become the de facto standard for them. By building intuitive, customer-facing applications, AI companies can capture greater value. When a user interacts with a polished, all-in-one product, the underlying model becomes secondary. This is a strategy reminiscent of how Microsoft leveraged Windows to become an essential part of the computing ecosystem.
Drawing Parallels With Windows And Office
The most compelling parallel is with Microsoft’s strategy in the personal computer era. Microsoft didn’t just offer an operating system; it bundled Windows with the Office suite, creating an ecosystem that locked in users and developers alike. This synergy between a platform and its applications generated powerful network effects and established a high barrier to entry for competitors.
For AI, the lesson is clear: Owning the customer interface is critical. A closed-source model that remains solely a backend service risks being undercut by open-source alternatives. However, if that model is paired with a strong application—something that seamlessly integrates into daily workflows—it can command a premium.
The integrated ecosystem approach means that even if the underlying AI is commoditized, the user experience can create a moat around the product.
The Ecosystem Question
The path to ecosystem dominance is not without its challenges. The nature of AI makes switching costs relatively low compared to the era of desktop software. Developers can often swap out one LLM for another with minimal disruption, thanks to standardized APIs and shared performance benchmarks. Furthermore, the collaborative spirit in the AI research community accelerates innovation, meaning that any proprietary advantage is likely to be short-lived.
The critical question remains: Can closed-source AI providers build an ecosystem robust enough to rival the enduring dominance of Windows and Office?
If the future of AI is to mirror past successes, companies must go beyond merely offering a high-performing model. They must create compelling, user-friendly applications that become indispensable in everyday workflows, capturing and sustaining value even as the technology itself becomes widely available.
Conclusion
DeepSeek R1 is a testament to the rapid strides in AI research, offering a glimpse into a future where advanced reasoning can emerge from innovative training techniques.
Yet, as the market for LLMs becomes increasingly crowded and commoditized, the true battleground shifts to how these models are deployed and integrated into real-world applications. The race is on for closed-source providers to build ecosystems that lock in users and create lasting value.
Only time will tell whether these efforts can replicate the legendary success of the Windows-Office synergy, or if the openness of today’s AI landscape will continue to democratize access to cutting-edge technology.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
