The Hidden Price of Free: OpenAI Spends £77,000 a Day Keeping ChatGPT Accessible

The Maths Behind the Model

Each word that ChatGPT generates costs OpenAI roughly £0.00024. Multiplied across 2.5 billion prompts processed daily and 5.8 billion monthly visits, the aggregate is enormous. With only 35 million paying subscribers out of 190.6 million daily active users, the cross-subsidy is eye-watering: the paying minority funds the free majority, and the ratio is worsening as adoption accelerates.

This creates what economists would call a perverse scaling dynamic. Unlike traditional software, where onboarding an extra user costs close to nothing at the margin, generative AI demands real compute for every single query. A single 100-word AI-generated email, if produced weekly, consumes approximately 7.5 kilowatt-hours of energy annually. Scale that across a continent of corporate users and the environmental and financial pressures compound quickly.

Anna Shedletsky, head of manufacturing AI at a major European industrial group, has publicly noted that organisations treating AI inference costs as negligible are building on sand. The same logic applies to the infrastructure providers themselves.

European Exposure Is Real

With only 18% of ChatGPT's user base located in the United States, international markets, including the United Kingdom and the EU27, account for the overwhelming majority of traffic. European enterprises, public sector bodies, and academic institutions have integrated ChatGPT into everything from legal drafting to code review. The IEA's 2024 analysis on AI and energy noted that data centre electricity demand in Europe is projected to double by 2026, driven substantially by inference workloads from large language models.

Researchers at ETH Zurich have modelled the energy intensity of transformer-based inference at scale, finding that per-query costs remain stubbornly high even as hardware improves. Their work underlines that efficiency gains from next-generation chips, while real, are unlikely to offset volume growth in the near term.

OpenAI's revenue picture does show momentum. The company reported £2.1 billion from consumer subscriptions in 2024, and its enterprise API business is now annualised at £12.8 billion. But revenue and profitability are different things, and the infrastructure cost curve is steep.

Revenue Streams: A Fragile Stack

OpenAI's current monetisation relies on three pillars, none of which is yet self-sustaining in isolation:

• Consumer subscriptions: ChatGPT Plus at £16 per month generates £2.1 billion annually from 35 million subscribers, but this must subsidise the far larger free-tier population.
• Enterprise API services: Annualised at £12.8 billion, this is the growth engine, but it is also the segment most exposed to competition from European and open-source rivals.
• Advertising revenue: Recently introduced for free-tier users, ads appear clearly labelled and separate from chat responses. This is an early-stage revenue line, not a lifeline yet.

Microsoft's estimated 27% stake, alongside capital from SoftBank and Nvidia, provides a buffer, but investors expect returns. An anticipated initial public offering will intensify scrutiny of the unit economics further.

Competitive Pressure From Within Europe

OpenAI does not operate in a vacuum. Mistral AI, headquartered in Paris, has built a credible challenger offering on far leaner infrastructure, partly by prioritising smaller, more efficient models. Mistral's approach has attracted backing from European institutional investors and has prompted serious questions about whether the American hyperscale model is the only viable path.

Anthropic's Claude is also upgrading its free offering, forcing OpenAI to maintain feature parity at the free tier even as costs rise. The competitive dynamic is uncomfortable: pulling back the free tier risks ceding users to rivals, but maintaining it deepens losses.

The EU AI Act, which entered into force in August 2024, adds a regulatory cost dimension. General-purpose AI model providers, including OpenAI, face transparency and systemic-risk obligations that require compliance investment. Andrea Renda, senior research fellow at CEPS in Brussels and one of the EU AI Act's most closely followed analysts, has argued that compliance costs will disproportionately pressure providers whose monetisation models are already strained. That observation lands squarely on OpenAI's free-tier economics.

What Changes Next

Industry observers are coalescing around a set of probable near-term scenarios for ChatGPT's access model:

1. Usage caps on free tiers: Monthly or daily query limits that nudge heavy users toward paid plans.
2. Feature migration behind paywalls: Advanced reasoning, longer context windows, and multimodal capabilities reserved for Plus or higher tiers.
3. Expanded advertising: More prominent ad integration for free-tier users, following the playbook of search and social media.
4. Tiered enterprise pricing: More granular API pricing to capture value from the highest-volume corporate users.
5. Efficiency-driven cost reduction: Hardware advances and model distillation reducing per-query costs, buying time without requiring access changes.

Key factors that will determine which path dominates include computational efficiency improvements, the pace of hardware advances at ASML and its chip-manufacturing customers, revenue diversification through enterprise channels, user tolerance for advertising, and the competitive behaviour of European and open-source alternatives.

For European organisations, the practical implication is straightforward: build contingency into any AI procurement strategy. Assuming that current free-tier access terms are permanent is a planning failure.