The £77,000-a-Day Reality Behind 'Free' ChatGPT

The per-word cost of generative AI output sits at roughly £0.00024 per word. Multiply that across 2.5 billion prompts processed daily and 5.8 billion monthly visits, and the computational bill becomes staggering. The paradox is stark: the more successful ChatGPT becomes, the more financially precarious its free tier grows.

Infrastructure Costs Scale With Every Query

Generative AI infrastructure bears no resemblance to traditional software scaling. When a social network adds a million new users, its marginal cost is close to zero. When ChatGPT adds a million daily active users, it adds a proportionate slice of electricity consumption, cooling capacity, hardware depreciation, and network bandwidth. A single 100-word AI-generated email, created weekly, consumes approximately 7.5 kilowatt-hours of energy annually. Scaled across hundreds of millions of users, the environmental and financial footprint is enormous.

Researchers at the Alan Turing Institute in London have flagged this dynamic repeatedly, noting that the energy intensity of large language models creates sustainability pressures that no amount of algorithmic efficiency can fully offset in the near term. The institute's work on AI compute costs has become a reference point for UK policymakers reviewing the Digital Markets, Competition and Consumers Act and its implications for AI service pricing.

Meanwhile, Margrethe Vestager, during her tenure as European Commission Executive Vice-President for Digital, consistently argued that the apparent "freeness" of digital services masks real costs borne by users, society, and the environment. Her framing applies directly here: ChatGPT's zero price tag is not a product of efficiency. It is a product of investor subsidy and cross-subsidisation from paying subscribers.

The Subscription Subsidy Model Under Pressure

ChatGPT Plus subscriptions at £16 per month contribute to covering the free tier, but the mathematics are unforgiving. OpenAI reported 35 million paying subscribers generating approximately £2.1 billion in 2024 consumer subscription revenue. Set against 190.6 million daily active users, the paying cohort represents a small minority underwriting access for the overwhelming majority. That cross-subsidisation model has limits.

The company's revenue picture is broader than subscriptions alone. Enterprise API services are growing rapidly, with annualised figures pointing toward £12.8 billion. Advertising for free-tier users has been introduced carefully, with placements clearly labelled and separated from chat responses. These moves reflect a calculated recognition that subscription income cannot carry the free tier indefinitely.

OpenAI's corporate structure adds further pressure. Microsoft holds an estimated 27% stake, alongside investments from SoftBank and Nvidia. An anticipated initial public offering will bring institutional scrutiny of unit economics that, at present, depend heavily on continued investor patience. European institutional investors considering positions in any future OpenAI listing should read the infrastructure cost disclosures carefully.

European Usage and the Energy Dimension

With only 18% of ChatGPT's user base based in the United States, international markets, including significant populations across the EU and the UK, account for the bulk of traffic. That matters for European energy policy as much as it does for OpenAI's bottom line. Data centres serving European users are subject to the EU's Energy Efficiency Directive and, in the UK, to the Climate Change Act's carbon budgeting framework. As AI query volumes rise, the energy draw of serving European users will attract increasing regulatory attention.

The International Energy Agency's most recent data on data-centre electricity consumption, cited by analysts at ETH Zurich's Energy Science Center, shows that AI workloads are among the fastest-growing contributors to data-centre power demand in Europe. ETH Zurich researchers have modelled scenarios in which, absent significant efficiency improvements, AI inference costs could consume a materially larger share of European grid capacity by 2030. Those projections inform both energy procurement strategies and the political debate about where to locate AI infrastructure.

Future Access Models Taking Shape

Industry observers across Europe are coalescing around a set of plausible near-term scenarios for how the free AI access model evolves. None of them involve the status quo persisting indefinitely. The most likely trajectories include:

• Usage caps on free tiers, limiting the number of daily queries or restricting access to newer models
• Expanded advertising inventory, moving beyond clearly labelled placements toward more integrated formats
• Migration of high-value features, such as advanced reasoning modes and longer context windows, behind paid paywalls
• Tiered enterprise pricing that extracts more value from commercial users, indirectly subsidising consumers for longer
• Computational efficiency gains from next-generation hardware reducing per-query costs, buying more time for the current model

Rivals are not standing still. Anthropic's Claude has been upgrading its free offering, and Google maintains a significant AI presence through Gemini. The competitive pressure forces OpenAI to keep its free tier attractive even as costs mount, creating a tension between market positioning and financial discipline that cannot be sustained indefinitely.

For European enterprises currently building AI strategies around free-tier ChatGPT access, the implication is direct: that access is a time-limited subsidy, not a permanent infrastructure assumption. Organisations integrating ChatGPT into internal workflows, customer service systems, or product development pipelines should model scenarios in which costs increase materially within the next 12 to 18 months. Developing in-house capabilities, exploring open-weight alternatives such as those offered by Mistral AI in Paris, or negotiating enterprise API agreements with defined pricing terms are all strategies worth pursuing now rather than when the free tier changes arrive.

The golden age of unlimited free AI access is not eternal. The numbers make that plain, and European business needs to plan accordingly.