Eric S. Raymond’s The Cathedral and the Bazaar introduced two dominant models of software development. The Cathedral is centralized, controlled by a select group of elites who dictate its structure. The Bazaar is decentralized and open, allowing rapid innovation through widespread participation.
But AI has disrupted this dichotomy. AI is not just software—it requires vast amounts of data and computing power, operates within complex ethical and political frameworks, and is deeply tied to corporate and national interests. AI race became a race that one nation can not afford to lose. Thus AI is no longer just about Cathedrals or Bazaars; it has evolved into something new: the Megachurch. Where the line between cathedral and bazaar gets blurred.
It is entirely possible to argue that Android was the first true “Megachurch” model in tech. However, Android didn’t start that way. Initially, it was closer to a Bazaar—an open-source operating system that manufacturers could freely modify. Over time, Google transformed it into a Megachurch by locking key services behind Google Mobile Services (GMS). While Android remains technically open-source, Google now exerts near-total control over its ecosystem.
The AI industry, on the other hand, is skipping the open-source phase entirely—many AI companies are launching directly as Megachurches, ensuring they maintain dominance from the start. This can be contributed to many things including the nature of AI models where intense resource and investment is required.
The Cathedral: OpenAI, Google, and Groq—AI’s High Priests
The Cathedral model in AI is embodied by OpenAI’s ChatGPT, Google’s Gemini, and Groq’s AI compute. These companies operate in a closed, hierarchical structure, where AI development is centralized and tightly regulated. OpenAI and Google treat their models as sacred knowledge—only accessible through carefully curated APIs, with users having no direct control. They position themselves as ethical gatekeepers, ensuring AI is used “responsibly,” but in practice, this consolidates power in the hands of a few, monopolizing AI’s future under the guise of protection.
Groq, while different in function, still belongs to the Cathedral model. Instead of developing AI models, it provides high-speed inference hardware, making AI access faster and more efficient. But like OpenAI and Google, Groq remains a centralized, corporate-controlled entity, dictating how its infrastructure is deployed. Groq might not control the AI models themselves, but it controls who gets to run them at scale.
The Bazaar: Mistral AI and the True Open-Weight Movement
In contrast to the Cathedral, the Bazaar in AI is represented by companies and organizations releasing fully open-weight models. Mistral AI is currently the best example—its models, like Mistral 7B and Mixtral, are not only powerful but also fully open and permissively licensed, allowing anyone to modify and fine-tune them. Unlike Meta’s LLaMA, which still requires licensing approval, Mistral’s models embrace the true Bazaar philosophy of open, decentralized innovation.
Other projects, such as Falcon (from the UAE’s Technology Innovation Institute), OpenAssistant, OLMo, and Stability AI’s StableLM, further support this Bazaar model. These initiatives encourage grassroots AI development, enabling smaller companies, research groups, and independent developers to participate without relying on the infrastructure of large tech giants.
However, the Bazaar remains an underdog in the AI arms race. While open-weight models offer freedom, they lack the massive capital and computational resources of the Cathedrals and Megachurches. Whether this model can remain competitive without eventually being absorbed by corporate influence remains to be seen.
The Megachurch: Meta and DeepSeek—AI’s False Prophets
The Megachurch model is distinct from both Cathedral and Bazaar. It blends the centralized power of the Cathedral with the mass appeal of the Bazaar, all while ensuring corporate dominance. Meta’s LLaMA and DeepSeek exemplify this approach. These companies present themselves as open and accessible, yet they retain strategic control over their ecosystems.
DeepSeek, in particular, exemplifies the Megachurch model—but not without controversy. It is widely speculated that DeepSeek’s AI models were developed with substantial assistance—or outright appropriation—of OpenAI’s ChatGPT technology. Reports suggest DeepSeek may have used distillation techniques—extracting knowledge from OpenAI models—to train its own AI, prompting an ongoing investigation by OpenAI and its major investor, Microsoft.
Adding to the speculation, DeepSeek’s rapid progress in training large-scale language models has raised questions, given China’s restricted access to cutting-edge AI chips. Despite U.S. export bans on advanced semiconductors, there is growing suspicion that DeepSeek circumvented restrictions by acquiring Nvidia’s H100 GPUs through shadow companies in Singapore, a region responsible for nearly a quarter of Nvidia’s global GPU sales.
Furthermore, DeepSeek’s claim that it developed its AI model for $5.6 million is misleading. This figure only accounts for the final training phase and does not include the substantial costs associated with early-stage research, data acquisition, and preliminary experiments. Large-scale AI models require billions of dollars in cumulative investment before reaching the final training phase. The true cost of DeepSeek’s AI development is almost certainly far higher than what has been publicly disclosed.
But in the Wild Wild West—or perhaps the Wild Wild East—of the AI race, there is no time to dwell on it. The AI war is already in full swing, and DeepSeek has made its move. Whether through innovation, opportunism, or strategic rule-bending, it has established itself as a formidable competitor. Complaining about fairness will not change the reality on the ground—the ball is now on our side.
This strategy aligns perfectly with the Megachurch model—creating the illusion of openness while selectively controlling access. DeepSeek and Meta distribute AI as if it were a universal right, yet in reality, they dictate how, when, and to whom this power is granted. The battle for AI dominance is not about who follows the rules, but who writes them.
The Future of AI: Who Will Control the Faith?
I cautiously predict that even OpenAI will eventually transition toward the Megachurch model. The Cathedral is too rigid, and the Bazaar too unpredictable. The Megachurch is the perfect middle ground—it offers just enough openness to attract mass adoption while ensuring centralized control remains intact.
Historically, religious institutions evolved from strict hierarchies to mass-market televangelism. AI is following the same trajectory—pretending to empower the people while keeping real power in corporate hands.
This article was developed with the help of several AI models, such as OpenAI o1, Google Gemini, among others.
The original plot and direction of the article were provided by the human writer, Jieon Choi.
Comments