GitHub Copilot: The Token Era Disrupts the AI Economy

The economy of artificial intelligence (AI) is undergoing a significant transformation with the adoption of a pay-per-use billing model by GitHub Copilot starting June 1, 2026. This change marks the end of unlimited flat-rate subscriptions, forcing companies to rethink their use of AI. Every token consumed will now be measured and billed, imposing a more rigorous management of resources.

This turning point comes at a time when the use of generative AI is becoming increasingly costly. On May 22, 2026, a report from Fortune, based on internal sources at Microsoft, revealed that in some cases, generative AI costs more than the work of a human developer. Companies like Amazon and Meta have even coined terms for this phenomenon, respectively "toxenmaxx" and "Claudeonomics." Uber, for example, has already exhausted its annual budget for AI tools in just four months.

Goldman Sachs predicts an exponential increase in token consumption, which could reach 120 quadrillion per month by 2030. Microsoft, for its part, is investing heavily in AI, with an annual budget of $150 billion, of which $25 billion is dedicated to rising costs of memory and chips. The idea of zero marginal cost AI is now a thing of the past.

A study from Microsoft Research highlights that agentic tasks, such as automated coding, consume up to a thousand times more tokens than a simple code chat. Moreover, the costs of these tasks are unpredictable, varying by a factor of 30 for similar executions. Anthropic, for instance, maintained its nominal prices upon the release of Opus 4.7, but the new tokenizer generates 35% more tokens for the same input text.

The environmental impact of AI is also concerning. The International Energy Agency estimates that energy consumption by data centers could reach 945 TWh by 2030, accounting for more than 1.7% of global electricity production. In France, data centers already represented 46% of the digital carbon footprint in 2025, up from 16% in 2020. ADEME and Carbone 4 emphasize that the carbon footprint of generative queries is 10 to 15 times higher than that of traditional web searches.

In light of these challenges, companies must now evaluate their AI projects not only in terms of licensing costs but also considering the carbon footprint and the human resources required. Google Cloud's DORA 2025 report indicates that while 90% of developers use AI, productivity gains do not always translate into organizational improvements. Gartner anticipates that more than 40% of agentic AI projects will be abandoned by the end of 2027 due to a lack of clear return on investment.

To maximize the benefits of AI, companies must ask key questions about token usage, carbon footprint, and team efficiency. This includes assessing the necessary volume to make the investment profitable, choosing between frugal or flagship AI models, and the need for agentic orchestration. Organizations must also measure the cost per session against the business value created to determine the true return on investment of AI.

Ultimately, the token economy imposes a new discipline in the use of AI, transforming what was once a choice of ambition into a strategic decision based on measurable economic and environmental criteria. Companies that can formalize this trade-off will emerge from the fog, while others risk discovering, like Uber, that they can consume in four months what was planned for twelve.