This text is part of a three-article series exploring the lesser-known challenges of generative artificial intelligence, beyond its apparent free access.
➤ The illusion of a cost-free digital world
The environmental impact and material challenges of these technologies
Who should pay? Ethical issues and future perspectives
Since the dawn of humanity, we have pursued the Promethean dream of creating in our own image. From Vaucanson’s automatons to Pascal’s mechanical calculators, this quest has shaped our civilization. But never, until now, had the boundary between human creation and artificial generation been so delicately blurred. I observe a revolution that, in barely two years, has completely redefined our relationship with the world: Generative AI!
ChatGPT, Claude, Midjourney, DeepSeek, Perplexity have become the new deities of a technological pantheon where creation seems liberated from human expertise. Text, image, melody: everything now merely awaits our desire, materialized in a few clicks, to transform us into artists with a simple snap of the fingers, without the shadow of a transaction.
It is this apparent free nature that constitutes the most seductive illusion of our time. Behind every dialogue with these intelligences lies a dizzying economic reality, a veritable shadow theater where puppeteers manipulate titanic infrastructures and devour colossal capital. We, enchanted spectators, only perceive the dancing silhouettes projected on the canvas of our screens, ignoring the complex and costly mechanisms that orchestrate this digital ballet.
Unlike social networks, where each new user is supposed to enrich the ecosystem without excessive additional cost (I concede that enrichment is a matter of perspective, as evidenced by the proliferation of TikTok videos each more inane than the last or the abundance of thunderous opinions from pseudo-experts on subjects they discovered two minutes earlier), each interaction with a generative AI triggers a cascade of expenses that are, to say the least, tangible.
Here is our fundamental paradox: on one side, millions of users lulled by the illusion of “all-free,” on the other, a technology whose every use turns the meter at a dizzying speed.
An essential question arises: who is currently financing this great cognitive feast?
Anatomy of these invisible costs
Remember that intoxicating sensation when, for the first time, you asked ChatGPT a complex question and in a few seconds, a structured answer appeared before your eyes. A technological miracle, we thought. A cognitive revolution at our fingertips, with no bill to pay. But haven’t we, collectively, forgotten this fundamental truth that our parents knew intuitively: nothing of value can truly be free?
This apparent free nature conceals a complex underground economy that we feed daily. Each interaction with these generative AI systems, whether ChatGPT, Claude, or Midjourney, consumes a substantial amount of resources:
- The infrastructure, these invisible foundations: In the beginning was the infrastructure, colossal, voracious, invisible. Models like GPT-4 rest on silicon archipelagos, supercomputers where thousands of processing units cluster together. The price of a single one of these computational orchestras can devour several hundreds of millions of dollars, without considering the air-conditioned mausoleums that house them or the white-coated priests who maintain them. Its perpetual operation generates a daily financial hemorrhage, mainly due to electricity consumption rivaling that of entire cities.
- The Promethean quest for training: Training the mind of a model like GPT-4 is akin to educating a digital god, an undertaking as ambitious as it is ruinous. The bill oscillates between 100 million and 1 billion dollars, encompassing raw computing power, massive data collection, and countless trial and error cycles. More troubling still: each new generation demands ten times more resources than its predecessor. This exponential progression questions the very sustainability of this technological race, even for the best-funded financial empires.
- The costly daily whisper: Each conversation with these digital oracles conceals an implacable economic mechanism. A simple search on a traditional search engine consumes about 0.3 Wh of energy. In comparison, a conversation of twenty questions with ChatGPT can require up to 50 Wh, equivalent to running a fan for several hours. The inference cost (that is, the sum of computational, energy, and financial resources consumed each time an artificial intelligence model mobilizes its parameters to generate a response from a query) of GPT-4.5 reaches $150 per million tokens, approximately 375 times higher than that of Gemini 2.0 Flash. For the professional user, this algorithmic poetry costs $75 for a million words submitted and $150 for a million words generated, sums that quickly transform the initial wonder into austere budgetary reality.
“Well” established financing strategies…
Faced with this dizzying economic reality, how do the architects of these digital cathedrals manage to maintain the illusion of free access while feeding these financial abysses? This paradox is not the fruit of chance but of meticulous orchestration, an economic choreography where the art of making people pay without it appearing so becomes the philosopher’s stone of our digital modernity. Let’s examine these financial alchemies that (temporarily) transform the lead of costs into the gold of growth.
- The calculated seduction through the freemium approach: The acquisition strategy exercised by OpenAI and Anthropic is part of a skillfully orchestrated economic choreography, where each free gesture masks an acquisition strategy with colossal financial stakes. I have observed with fascination the press conferences where Sam Altman evokes democratic access to AI while silently raising billions from investors who, certainly, have no intention of being satisfied with noble intentions. This cognitive dissonance between public discourse and economic imperatives outlines the contours of a precarious balance, as fragile as it is untenable over time.
- The art of chess: Tech giants play a different game. They finance their AI initiatives thanks to accumulated rents in their domains of predilection. Microsoft has woven Copilot into the very fabric of Windows and Office365. Google orchestrates a similar strategy with Gemini. For these titans, generative AI is not so much a direct source of revenue as a strategic lever to consolidate their empires. This cross-subsidization strategy raises troubling competitive questions, potentially erecting insurmountable ramparts for new entrants.
- The praise of precision: Faced with the economic unsustainability of generalist models, a third way emerges: the development of specialized AI solutions for specific sectors where added value justifies a premium subscription. In the legal universe, solutions like Harvey AI offer document analysis capabilities calibrated for law firms. The medical field sees the birth of specialized assistants decrypting patient files with precision. This approach directly targets segments where the willingness to pay is substantial, allowing a faster trajectory towards profitability, while requiring considerable investments in sector expertise.
… but for what perceived value?
These financing strategies, as ingenious as they may be, can only persist if they meet a sufficient perception of value among the different actors in the ecosystem. Because beyond theoretical economic models lies the tangible reality of user experience, this subjective alchemy where perceived utility must transcend the consented cost. How is this complex equation resolved in the minds of organizations, independents, and the general public? The value of generative AI proves to be as diverse as its uses, as nuanced as the contexts where it fits.
- From mirages to pragmatic analysis: For commercial organizations, the perception of generative AI follows a natural curve, from initial wonder to a sober and calculating evaluation. Large companies can explore customized solutions, evaluating these investments through the prism of return on investment. This ROI varies considerably depending on the domains, promising in content creation, more uncertain in sectors requiring absolute precision. I recently met a marketing director who, after deploying ChatGPT throughout his team, confided: “We first believed in the miracle. Six months later, we have an Excel spreadsheet that calculates where AI saves us time and where it makes us lose it. Some teams use it only for briefs, others have abandoned it. Wonder has given way to accounting pragmatism.” SMEs, constrained by narrow budgetary horizons, must precisely target use cases where AI brings immediately tangible value, favoring solutions already integrated into their daily tools.
- The economic tightrope walker: For independents, these solitary workers, generative AI represents a paradoxical companion, both liberating and budgetary predator. A designer flourishes in the possibilities of Midjourney but must integrate this subscription into his delicate financial equation. A writer dialogues with ChatGPT but must constantly ensure that this algorithmic partner generates more value than it consumes. A freelance translator recently explained to me her daily mental calculation: “20€ monthly subscription to save me eight hours of work, it’s profitable. But I have to check each technical term, because a single error would cost my credibility. AI is my silent assistant, not my replacement.” These professionals adopt a cautious approach, first exploring free versions, then gradually investing when the economic alchemy proves favorable.
- Between fascination and reluctance: For the average user, the appeal of generative AI fits more into a playful dimension or immediate utility. The data reveals a striking ambivalence: on one side, an almost frenzied adoption of free versions, testifying to an insatiable collective curiosity; on the other, an almost visceral resistance to the idea of remunerating these services. The other evening, a friend was enthusiastic about the visual creations he had generated for his son’s school project. To my suggestion to invest in a premium version for more efficiency, his face darkened: “Pay? What for? I use it every day for free, even for my professional emails!” This reaction reveals the collective amnesia that strikes our digital society – a convenient forgetting of the adage hammered since the dawn of social platforms: if you don’t pay for the service, then you yourself are the product. Our user, in his voluntary naivety, seems to have erased from his memory this fundamental equation of the digital economy, preferring to believe in the miracle of algorithmic abundance without counterpart. This paradox explains the strategic integration of generative AI into pre-existing service bundles – concealing the cost in a larger whole, like a bitter medicine coated in sweet syrup that the consumer will swallow without even noticing.
Towards a realistic economy of generative AI
This mosaic of value perceptions outlines the contours of an economic landscape in full mutation, where initial enchantment gradually gives way to a more pragmatic evaluation. As the veil of novelty dissipates, emerges the necessity for a refoundation of the tacit contract that binds the creators of generative AI to their users. Free access, this seductive but unsustainable mirage, must give way to a more sincere, more transparent, more sustainable economic architecture.
Ultimately, the economy of generative AI confronts us with this ancient truth that economist Milton Friedman had summarized in his famous formula: “There is no such thing as a free lunch.” Each interaction with these systems has a cost, whether energetic, environmental, social, or cultural.
What paths then open before us to reconcile the democratic accessibility of these technologies with their inescapable economic reality?
I am deeply convinced: the illusion of free access that nimbles generative AI is doomed to dissipate like dew under the first rays of economic sun. The real, tangible, incompressible costs, these material truths that impose themselves beyond all rhetoric, inexorably lead us towards more transparent and viable economic models.
This transition will require a re-education of our collective expectations, we who have been lulled by decades of apparently free digital services, fed at the breast of an attention economy that skillfully masked its extractive mechanisms. It will also require renewed creativity in pricing models, an economic poetry capable of translating into concrete proposals the protean value of these artificial intelligences.
Even more deeply, this metamorphosis invites us to rethink our very conception of value in the digital ecosystem. Generative AI could mark the dawn of an era where our contribution would no longer be limited to our distracted attention or our personal data, but would extend to an assumed, conscious, proportionate financial participation. An era where the user would become a customer again, where the transaction would find its place in the exchange, not as a constraint but as the foundation of a balanced relationship.
Perhaps we will discover that the true alchemy does not reside in an illusory perpetual free access, but in the transmutation of our very relationship to technology. What if generative AI, beyond its technical prowess, offered us the opportunity to reinvent our social pact with digital innovation? A pact where cost transparency would engender a sharper awareness of our usages, where financial contribution would become the guarantor of ethical and sustainable evolution, where democratization would no longer mean “free for all” but “accessible according to each one’s means.”
The fundamental question is therefore not to determine if generative AI has a cost, this evidence imposes itself with the force of a physical law, but how this cost should be distributed among the different stakeholders. How to orchestrate this distribution to allow both continuous innovation and democratic access to these augmented capabilities? It is on this condition, and on this condition only, that we will be able to transcend the mirage of digital free access and build an AI economy that is both sustainable and equitable, an ecosystem where the value created nourishes the value to come in a virtuous and sustainable cycle.
This demanding path, less seductive than the promises of abundance without counterpart, could well be that of a technological maturity finally reconciled with our economic humanity.