AI Economics

Training an AI model is a significant undertaking. Not only does it demand substantial financial investment, but it also requires highly specialized infrastructure and software capable of processing the data necessary for model creation. At the end of the day, this investment must generate a return—and a significant one at that. Google is known for its ad revenue, and Meta likely follows a similar strategy, focused on data and user engagement to drive ad sales. However, OpenAI operates somewhat differently.

While Sam Altman often projects an altruistic image as a tech founder—possibly because of his previous entrepreneurial success—OpenAI is still a business. It involves more than just Altman; indeed, had events unfolded differently, he might not even be the CEO today, following the recent attempted coup by a colleague. The essential point is that OpenAI, like any company, needs to generate revenue to survive. So, how is OpenAI planning to make money?

In addition to premium subscriptions, it has been selling access to its API to developers, who then build applications and charge their customers. This is a clever strategy, but not without risks. Recently, OpenAI released its latest model, ChatGPT-4o, for free. Well, sort of, it's restricted to a daily maximum of prompts. This decision likely wasn't made lightly but rather out of necessity. They probably realized that few people were paying for subscriptions and decided it was better to have a large user base, even if it didn't generate direct income. Now, OpenAI is hoping that developers will find the model valuable enough to keep paying for access to it.

However, a lot of hinges on this strategy. While Microsoft pays OpenAI a significant amount to use its model, reports suggest that Microsoft is developing its own AI behind the scenes. It may not yet match OpenAI's capabilities, but that might not matter in the long run. Many users, including myself, have been content with ChatGPT-3.5 because it meets their needs. Why pay for a better model if the existing one suffices? Moreover, if Google integrates its AI model with its ad business, what's to stop developers from choosing Google's Gemini over ChatGPT? Privacy could be a differentiator, but more on that later.

Meanwhile, Apple allows its customers to run AI models locally, and Microsoft has adopted a similar strategy with their latest AI PCs. This local processing approach poses a significant challenge for OpenAI, which relies on cloud-based access. This dependency on reliable internet connections can be problematic, as demonstrated by the poor reviews of AI Pin and Rabbit r1, where response time—or delay—was a critical issue. Slow performance drives users to seek alternatives, plain and simple. This suggests OpenAI is praying for both 5G to take off and wearables to become mainstream, not to mention AI itself.

Given the limited pathways for OpenAI to achieve profitability, it may need to explore additional revenue streams. Many people might not realize that every prompt in ChatGPT is saved for training purposes. These private conversations are stored on a server, ready to be used for model improvement. This raises significant privacy concerns. Imagine a scenario where this data gets stolen—this would be one of the most invasive breaches in history. Stronger systems have been compromised, so why not this one? By the way, the only way to disable this data processing is by turning off chat history, even though these functions are not inherently linked.

What becomes clear is that OpenAI definitely understands the importance of user data for training its models in order to keep an advantage over its competitors. But what's to stop them from using this same data for other purposes in the future, like advertising, especially if the company faces financial difficulties? They might argue that the data is anonymized and all that, but it would probably hinder people sharing personal information. Ultimately, OpenAI needs to find ways to sustain its business. Currently, their revenue largely depends on Microsoft and developers using their API. However, if either one shifts to a different, possibly cheaper, model, OpenAI will have to face a real crisis.

In conclusion, while OpenAI appears promising today, it must not underestimate the pragmatism of its market. Consumers and businesses are not particularly loyal to any brand at this stage; they simply want the most cost-effective, reliable AI model available. The market will undoubtedly shift towards whoever can provide this. To ensure fairness and prevent monopolistic practices, Apple and Google should offer choice screens for their AI engines, allowing users to switch easily between different models. This would help avoid lock-in situations and foster healthy competition.

As I've already discussed in my previous article about AI engines, achieving this balance requires an unprecedented level of trust and collaboration in the tech industry. But if we believe experts who claim that AI is as important as the invention of fire, this might be the most important partnership in history. Time will tell if these tech companies can unite and consider how future developments in this space may impact all parties involved.