How a temporary demo became a billion-dollar business

When OpenAI launched ChatGPT in November 2022, it was, according to Turley, a temporary demo they planned to shut down after a month. But the chatbot became a viral phenomenon — it gained a million users within the first five days, and OpenAI quickly realized they had a real product on their hands. "We literally stumbled into subscriptions," Turley admitted. The paid plan model emerged as an emergency solution to capacity problems — servers couldn't keep up and the company needed a way to manage overwhelmed demand. ChatGPT Plus at $20 a month, later Pro at $200 — all of these were steps that weren't the result of a carefully crafted business strategy, but real-time improvisation.

Unlimited AI is like unlimited electricity

Two and a half years after launch, however, the situation is radically different. Models are increasingly powerful — and also exponentially more expensive to run. Each new GPT model requires orders of magnitude more computing power than the previous generation. While GPT-4o was optimized for speed and low cost, newer models like GPT-5.5 and its variants are substantially more demanding on infrastructure. "There is no scenario in which pricing wouldn't change significantly when technology is evolving this fast," Turley emphasized. The comparison to electricity is no coincidence — Sam Altman at the BlackRock Infrastructure Summit in March 2026 described a vision where AI is sold as a utility: "We see a future where intelligence is like electricity or water — people will buy it from us by the meter and use it for whatever they want."

Pay per token: the new standard in the AI industry

OpenAI isn't the only one moving away from traditional subscriptions. Anthropic and Google already charge based on usage — their API services operate on a pay-as-you-go basis, where customers pay for each processed token (a unit of text that the model processes or generates). This model is more transparent for demanding enterprise users, but for average consumers it could mean the end of predictable monthly bills. Microsoft is going even further. CEO Satya Nadella hinted on the Dwarkesh Podcast that the company is considering charging "per agent" instead of "per user" — meaning the price will depend on how much autonomous work the AI does, not on the number of people using it. Consulting giant Globant, meanwhile, is experimenting with token packages and so-called "AI Pods" — monthly subscriptions that include a predetermined volume of tokens. Once exhausted, customers purchase more.

Ads in ChatGPT: an alternative for those who don't want to pay

Alongside considerations of pricing changes, OpenAI is rolling out an advertising pilot program that launched in February 2026. According to data from Sensor Tower, over 100 brands have joined the testing, with 44% of them being retail companies. Among the first advertisers are Best Buy, Expedia, and Target. Ads appear as clearly marked cards with a "sponsored" label — for example, when asking for a smartphone comparison, a Best Buy offer appears. Turley described ads as a way to make ChatGPT accessible even to those who can't afford paid plans. "Our north star is accessibility," he emphasized. Free users could thus be funded by advertising in the future, while paying customers get a premium, uninterrupted experience.

What this means for Czech users

In the Czech Republic, ChatGPT is massively popular — according to Similarweb data, it ranks among the most visited AI services on the Czech market as well. It is currently available for free with limits, Plus for $20 per month (approximately 460 CZK), and Pro for $200 per month (about 4,600 CZK) — the official pricing is on the OpenAI website. The Pro model offers unlimited usage and access to the most powerful models. If OpenAI switches to a metered consumption model, Czech users — especially the more demanding ones who use ChatGPT daily as a programming assistant or creative partner — may feel a significant price increase. On the other hand, occasional users could pay less than the current $20. Also important is the context of the EU AI Act, which has been gradually introducing new obligations for operators of large AI models since 2025. OpenAI will have to ensure transparency and safety in the EU, which could further increase operational costs — and be reflected in end prices for European customers.

Competition pushes prices down

While OpenAI considers price increases, the competition is heading in the opposite direction. Google recently lowered Gemini subscription prices and launched the Gemini 3.5 Flash model with aggressive pricing. Claude by Anthropic offers competitive plans and its Sonnet model is comparable to GPT-5.5 in many benchmarks. Moreover, open-source models like Meta's Llama or models from Mistral are entering the market, which technically skilled users can run for free on their own hardware. This pressure from below means OpenAI can't raise prices indefinitely — it must find a balance between covering rising costs and maintaining competitiveness.

What exactly is OpenAI planning?

Specific plans have not yet been made public. However, several likely scenarios can be inferred from Turley's and Altman's statements. The first is introducing limits even on higher-tier plans — for example, instead of unlimited Pro for $200, a plan with a specific number of tokens or queries per month would emerge. The second option is metered consumption, where users would pay based on how much computing power they actually use — similar to cloud services like AWS or Azure. The third path is a hybrid model that would combine a base subscription with an additional charge for above-standard usage — a certain volume "included" and everything beyond that at an extra charge. This model is common in the tech world (used by GitHub Copilot, for example) and would be the easiest for users to swallow. One thing is certain: the era of unlimited AI for a flat fee is coming to an end. And it's not just about ChatGPT — the entire industry is moving toward a model where intelligence is paid for like water or electricity. Metered, drop by drop, based on actual consumption.