Summary of Key Points
Recently, there has been a subtle shift in the focus of competition within the AI industry: while in the past, companies were vying for large models (such as GPT), computing power (like NVIDIA GPUs), and data centers, the giants are now aggressively competing for third-party platforms that do not develop models, manufacture chips, or build data centers. For example, the open-source AI community Hugging Face has been rumored to be sold for $13 billion, and the model routing platform OpenRouter was acquired by Stripe for $8 billion. The value of these platforms lies not in their technology itself, but in their role as “hubs” within the AI ecosystem—connecting developers, models, and users, making them crucial entry points to the AI arena.
1. Can Platforms Worth Billions Be Valuable Without Developing Models?
Hugging Face and OpenRouter share a common characteristic: they do not produce models themselves, but they enable everyone to use them.
- Hugging Face acts as a “big marketplace” for AI models: Developers can upload, download, and test various open-source models here (with models like DeepSeek-R1 and Kimi K3 being released first on this platform), and they can also develop new models based on existing ones. It’s like the AI version of GitHub, with 3 million models, 1 million datasets, and 13 million users—bringing together AI developers from around the world.
- OpenRouter functions as a “central hub” for model invocation: Previously, developers had to connect to each company’s interface separately to use multiple models (e.g., OpenAI, Claude, and open-source models). OpenRouter provides a unified interface that allows access to over 400 models and automatically selects the most suitable one for a given task. It handles trillions of model requests daily, attracting users ranging from individual developers to enterprises.
Their core value lies in their ability to “connect things” rather than in creating products, thereby addressing issues of information asymmetry and complexity in the AI supply chain.
2. What Do the Giants Want by Acquiring These Platforms?
The real goal of the giants is to gain control of the “voice” within the AI ecosystem.
- This is similar to Microsoft’s acquisition of GitHub in 2018: it didn’t buy GitHub for its profitability (it was not yet profitable at the time), but because it controlled the global developer community’s code ecosystem. Now, Microsoft’s AI programming tool Copilot is integrated with GitHub, making it an essential tool for developers.
- In the AI context, controlling Hugging Face means controlling the distribution of open-source models, and controlling OpenRouter means controlling the entry points for enterprises to use models. For instance, Stripe’s acquisition of OpenRouter aims to apply its expertise in payment services to the AI domain. Previously, businesses didn’t need to handle payment processes themselves; now, they can use OpenRouter to simplify this process. For giants, controlling these entry points is more important than developing models themselves. Even if you have a powerful model, it’s useless if no one uses it; having control over the entry points ensures that your model gets noticed and used.
3. The Concerns Around Billion-Dollar Acquisitions: Does Loss of Neutrality Reduce Platform Value?
The fundamental value of these platforms is their neutrality—models from Meta, Google, or individual developers can all be displayed equally on them. However, this advantage may be lost after acquisition by a giant.
- For example, if Hugging Face is acquired by Microsoft, might it start recommending Microsoft’s models? If OpenRouter is acquired by Stripe, might it favor models used by Stripe? If neutrality is compromised, developers might switch to other platforms, significantly reducing the platform’s value.
This is why Microsoft promised to maintain Hugging Face’s independent operation when it acquired it, recognizing that losing neutrality would be detrimental to its reputation. The biggest challenge for these platforms in the future will be balancing the interests of the giants with the trust of developers.
4. The Logic of AI Competition Has Changed: From “Who Is the Strongest” to “Who Can Connect the Most People”
In the past, AI competition focused on “hard power”—who had the largest model size, more computing power, or higher accuracy. Now, it’s about “ecosystem competition”: whoever can connect the most developers, users, and models will win.
- Even the strongest models are of little use if there’s no development or integration into real-world applications (e.g., healthcare, education).
- Users are less likely to use AI if there’s no convenient access to it.
- Giants have realized that the future of AI does not lie in one company dominating the market, but in a collaborative ecosystem. The company that can create the best environment for developers and users will be the most profitable.
This is similar to the role of platforms like淘宝 in the internet era:淘宝 doesn’t produce goods; it connects sellers and buyers, becoming the most profitable platform. Hugging Face and OpenRouter in the AI world serve the same purpose.
In Conclusion
The AI industry is transitioning from a “technology race” to an “ecosystem race,” with third-party platforms that connect everyone becoming new “gold mines.” However, their ability to maintain neutrality is key to their continued value.
(End of translation)