ChatGPT wants to be your new Windows

BarcelonaOpenAI announced last Monday at its DevDay conference for developers that ChatGPT will become a universal platform for interacting with third-party services, from Spotify to Zillow. The official message: a revolution that will turn the popular chatbot into "something more like an operating system," in the words of Nick Turley, head of ChatGPT. The reality: a desperate flight forward by a company that lost $5 billion in 2024 and spends $2.25 for every dollar it earns.

Call me a dirty thinker, but when a company that burns more than $2 billion in six months announces it wants to become a universal platform, perhaps we should ask ourselves if we're not looking at a solution looking for a problem.

Integrated applications

The fact that these integrations aren't available in Europe yet doesn't mean the movement deserves attention: we're witnessing an attempt to reorganize the entire tech industry. OpenAI's new tool (SDK) allows developers to build applications that work directly within ChatGPT conversations. Seven applications are available as of the day of the announcement: Spotify, Booking.com, Canva, Expedia, Zillow (the equivalent of Idealista in the US), Coursera, and Figma; another 15 have been promised "later," including DoorDash, Uber, and Instacart.

Sam Altman, the owner of OpenAI, sells it as a new era of "interactive and personalized" applications. But when Casey Newton of Platformer Asked about how they'll monetize the platform, the response was predictably: "Details will come later." Naturally. First, we'll hook the developers, and then we'll figure out how to suck their money.

Apps held hostage

The Spotify integration illustrates how it works. Available in 145 countries (but not the EU), it allows you to request songs or podcasts in natural language within ChatGPT. You ask the chatbot, "Spotify, make me a dance playlist for Friday's party," and the app appears in the conversation. Free users access existing playlists like the Weekly Discovery, while paying users get personalized playlists powered by AI. Spotify's Sten Garmark says that "Spotify's vision has always been to be everywhere you are." What he's really saying is that this now means submitting to a platform that aspires to be the mandatory intermediary between apps and users.

To reassure record labels, Spotify assures that it will not share music with OpenAI. One of the Big Three, Universal Music Group, has already applauded "a monetized ecosystem." Translated: ads and commissions.

At first glance, one might think that OpenAI wants to place ChatGPT on the same level as Apple's App Store and Google's Play Store, as a new intermediary for the distribution of apps, services, and content. But the move is more reminiscent of Windows in the 1990s, when any app had to be available for Microsoft's operating system. According to analyst Ben Thompson, OpenAI doesn't want to "integrate with the web," but rather to make ChatGPT the operating system of the future. If your app isn't on ChatGPT, it won't exist for the platform's 800 million users. And the burden of making these integrations work will fall on third-party developers. This is the power of controlling user access.

Windows and the AMD deal

In 1981, IBM urgently needed a personal computer and turned to Intel for the processors and Microsoft for the DOS operating system. It also required Intel to license its processor to AMD as a secondary supplier. The real beneficiary was Microsoft: the one who controlled the software made more business than the one who made the chips.

Now OpenAI is repeating that trick, even with some of the same people involved: it has just announced an agreement with AMD to buy 6 gigawatts of computing power from its MI450 chips. The announcement comes shortly after AMD rival Nvidia invests in OpenAIThe message: ultimate power lies not with those who make the fastest chips, but with those who control user access.

800 million users burning money

OpenAI boasts 800 million weekly active users, but hides the fact that each user loses them money. The ratio between free and paid users is very low: 2.6%. With 20 million subscribers who nominally pay $20 per month, it generates around $4.5 billion annually. The computing costs for training models are around $3 billion annually, but inference costs (answering questions users ask in ChatGPT) must be added, which is another $2 billion. Before salaries and marketing, OpenAI already spends $5 billion, but earns much less. The company expects to lose $115 billion by 2029.

The AI industry as a whole has invested $325 billion in infrastructure this year, with no generative AI company profitable. An MIT study claims that 95% of AI pilot programs in companies fail to observe any clear business improvement. China's DeepSeek and Alibaba have already matched the performance of OpenAI's models at a much lower cost, making AI models increasingly less valuable. Stretching the comparison between ChatGPT and Windows, it could be said that there's also an Apple in the AI landscape: Google, which applies the same level of extreme integration between its products and services. Simply put, the AI market seems to be moving toward another duopoly: OpenAI and Google.

This is where OpenAI's new platform makes sense: if you don't earn enough from your users, let the developers pay. OpenAI can charge them commissions without assuming all the costs.

When Altman says that profitability "is not among his top ten concerns," perhaps it's because his problem has no solution. The new platform's move isn't innovation: it's survival disguised as strategic vision. OpenAI has positioned itself as the straw man in the AI investment bubble and has ensured that it will continue to receive speculative capital. Until the bubble bursts, of course. Developers would do well to remember Zynga's precedent: its game Farmville It once represented 12% of Facebook's revenue, until Facebook lost its need for it. We, the users, should ask ourselves if we need the most expensive restaurant search tool ever invented to become our operating system.