Companies have also joined the bandwagon - especially those involved in software development. The end of 2025 is generally considered a major turning point in software development, when Claude Code made such a huge leap in AI-assisted programming that it managed to win over even those coders who had previously been skeptical of AI's programming skills.
But the hype has hidden an unpleasant truth that AI companies have barely dared to speak aloud.
Almost all current AI services are, in fact, loss-making businesses for the companies offering them. This means that consumers are currently able to use AI solutions that they might not be able to afford - or be willing to pay for.
Information about this financially unsustainable situation is hidden in companies' financial figures. And digging out precise data is made more difficult by the fact that the industry's largest independent players, OpenAI and Anthropic, are not publicly listed companies.
But information can still be found. Advertising giant Google has been accustomed to being seen as a money-making machine, generating enormous profits year after year. But even Google's profits could be at risk as the company is forced to pour hundreds of billions into building its future data centers that provide AI computing. Similar indications can also be found in the financial figures of another profit engine: just like Google, Microsoft may also have to compromise on its profits as it too builds or leases data centers around the world with long-term contracts.
The biggest question mark is, of course, OpenAI, which, unlike Google and Microsoft, has no other business besides AI to finance the enormous investments required by AI (paywalled article). The company is practically at the mercy of its investors and lenders as to whether it can build enough computing capacity, while simultaneously continuing to incur huge losses month after month.
Someone has to pay
In practice, we are now approaching the stage where end-users will be forced to start paying the true price for AI. And that reality can be very harsh, whether the payers are consumers or companies using AI in their business.
Many signs of this change have already been seen in the last couple of months.
Anthropic, the developer of Claude Code, has been continuously tightening the billing criteria and usage rights for its AI models over the past couple of months. A developer's Max-level license, costing $200 per month, is no longer sufficient for almost everything; on top of that, one must dutifully start paying for the costs incurred from actual usage. At the same time, functionalities included in fixed monthly prices have also been drastically cut.
OpenAI, strong in the consumer market, knows that average consumers are not willing to pay money for using ChatGPT, but money still needs to be made somehow. So, the method by which even free online journalism for consumers has been paid for until very recently is coming to the rescue: advertisements. And ads have already arrived in ChatGPT in some countries. In the future, it is likely clear that more and more ads will be seen in AI services intended for consumers: Google is also already considering bringing ads to the company's Gemini AI and there are already ads in its search engine's AI summaries.
Another clear sign of this growing cost awareness was seen early in the year when OpenAI discontinued its Sora service, which creates AI videos and had been publicly launched only a few months earlier. The company likely calculated that it would not be able to make Sora profitable in any way.
Next, we will likely see a situation where reasonably affordable AI model programming interfaces start to become more expensive, so that their price truly reflects their production costs. This could severely impact thousands of companies that have integrated AI functionalities into their own products and services.
The rise of local AIs..?
Of course, something sensible might also emerge from all of this. Local, free, and entirely user-run AI models on one's own computer (or server) have also been developing at a furious pace. Already, the cleverest coders have adopted solutions where an AI running on their own computer handles a kind of pre-processing for AI requests - and the paid AIs humming in data centers are only used for solving the most demanding tasks.
Using local language models is, in a way, very smart, because each of us currently uses a computer whose power we only utilize a small fraction of, most of the time. Harnessing that unused computing power for one's own use is already quite clever in itself. And by using AI running on one's own machine, all information fed to the AI is naturally kept entirely in one's own hands, within one's own computer.
The near future
This year, it appears two significant turning points will occur in AI development and the assessment of its financial risks. Both OpenAI and Anthropic are most likely going public already during 2026. At the same time, the companies' financial figures will become available for everyone to evaluate quarterly - and investors will have to make assessments after each earnings release as to whether they believe AI computing can truly be turned into a profitable business.
In itself, the building frenzy for AI-demanding data centers is unlikely to calm down in the very near future, as so much money has been poured into it and so many expectations invested that both investors and AI companies will likely be able to wait a few more years. But even if AI companies are not necessarily required to have profitable years just yet, they must soon be able to prove to investors that the direction is at least correct.
Written by: Petteri Pyyny @ 19 Apr 2026 12:58
