India overtook the US. The AI center of gravity is shifting to the Global South

The biggest surprise of the analysis is the geographic redistribution of AI traffic. While just two years ago ChatGPT users from the United States dominated, today the clear leader is India with 13.2 billion prompts per month. The US follows with 11.9 billion. Among the most active markets are also Brazil, Indonesia, and the Philippines. This shift reflects a broader trend: large populations in developing countries, which skipped the desktop computer era straight into the mobile age, are adopting AI tools surprisingly fast. In contrast to previous waves of internet growth — where the West dominated — AI adoption is much more evenly distributed. This will have a fundamental impact on where infrastructure investments flow and where regulations will be shaped.

Europe and Czechia: High usage intensity

In Europe, AI usage is more concentrated but all the more intense. Great Britain stands out: 37.1 million monthly visits, 2.44 billion prompts per month, meaning roughly 35 prompts per person per month — about one query per day for every resident. For comparison: Google Gemini has an estimated 750 million users, but the average frequency of use is lower than ChatGPT. Anthropic's Claude scores well with developers, but its market share in general search remains below 4%. The Czech Republic can observe a similar trend to Britain, just on a smaller scale. ChatGPT is available in Czechia both for free and in paid versions Plus ($20 per month, approx. 460 CZK) and Pro ($200, approx. 4,600 CZK). Czech is fully supported by the GPT-5 model and the quality of responses in Czech has improved dramatically since the arrival of GPT-4. For Czech companies and freelancers who use ChatGPT for writing, translations, or programming, it is now an everyday work tool — much like Google was years ago.

AI as a power plant giant: 22 billion kWh per year

And now the numbers that make your head spin. According to researchers from the AI lab at the University of Rhode Island, GPT-5 consumes an average of 18.9 watt-hours of electricity per one medium-length prompt. That is roughly 50× more than a standard Google search, which requires about 0.3 Wh. Converted to global traffic, this means:

• 60.7 GWh daily — equivalent to the daily consumption of Puerto Rico or Croatia
• 1,908 GWh monthly — more than Slovenia's annual consumption (1,431 GWh)
• 22,150 GWh annually — equivalent to powering 2.11 million American households for an entire year

For perspective: ChatGPT's annual query consumption could charge every electric vehicle in the US 36 times, or 4.23 billion iPhones every day for an entire year. And we are only talking about the inference phase — meaning answering queries. Training models like GPT-5 is even more energy-intensive. Training GPT-4 consumed roughly 62 GWh, which is 48× more than training GPT-3.

The energy bill: $8 million per day

At the average US commercial electricity price of $0.136 per kWh (as of December 2025), the daily electricity bill for ChatGPT comes to over $8 million. Annually, that's over $3 billion just for electricity. In the European context, where energy prices are often double those in the US due to regulatory fees and emission allowances, the bill would be even more striking. This is one of the reasons why major AI infrastructure is concentrated primarily in North America and Asia. The good news is that OpenAI is certainly not losing money on operations. With 50 million paying subscribers (of which an estimated 32 million are Plus at $20, 8 million Go at $8, and 2.5 million Pro at $200), the company generates roughly $1.3 billion per month from subscriptions, according to BestBrokers. Annual electricity costs are thus covered in less than three months.

The rebound effect: More efficient models, higher consumption

The paradox of AI is the classic rebound effect: the more efficient models become, the more people use them — and total energy consumption rises despite technological progress. While GPT-5 is significantly more efficient per token than GPT-3, thanks to massive adoption, multimodal features (images, voice, video), and reasoning capabilities that require longer model "thinking," overall energy intensity is not decreasing, but rather increasing. Compared to GPT-4, GPT-5 can be up to 8× more energy-intensive in certain tasks.

Competition isn't sleeping — and also consumes electricity

ChatGPT holds a dominant position with a 60.4% share of the AI search market, but competition is growing. Google Gemini has 15.2%, Perplexity AI 5.8%, Microsoft Copilot benefits from integration with Office and Windows. A separate chapter is Anthropic's Claude, which wins over developers and companies with its emphasis on safety and long context. Moreover, Anthropic recently overtook OpenAI in market valuation ($965 billion) and its model Claude Opus 4.8 now surpasses GPT-5.5 in benchmarks. Each of these models has its own energy footprint, and together they form an industry whose energy intensity will be one of the key topics in the coming years — including the EU AI Act, which since 2025 requires operators of large-scale AI systems to report energy intensity.

What this means for the average user

For the average user in Czechia, nothing dramatic changes — ChatGPT remains available for free and OpenAI is not raising subscription prices in the foreseeable future. Quite the opposite: the cheaper ChatGPT Go plan at $8 per month (approx. 185 CZK) shows that the company wants to expand among less affluent users, especially in Asia and Africa. In the long term, however, AI energy intensity will manifest in two things: first, AI service prices will have to reflect real operating costs — which is why OpenAI has already admitted that the "unlimited" subscription model may not be sustainable. Second, AI sustainability will become a regulatory topic that will also affect the availability of certain features in the EU.