その AI economics crisis is becoming increasingly visible as companies shift from flat subscription pricing to usage-based models. The recent move by GitHub to introduce token-based billing for Copilot signals a broader industry correction: generative AI services cost far more to operate than users have historically paid.
For years, AI tools were offered at fixed monthly prices that obscured the actual cost of computation. As models evolve into more complex, agent-like systems capable of multi-step tasks, their demand for processing power has increased significantly. This has exposed a fundamental mismatch between pricing and actual infrastructure costs.
Providers such as OpenAI and Anthropic rely on expensive GPU infrastructure and large-scale data centers, often powered by hardware from NVIDIA. These systems require continuous investment and high utilisation rates to remain economically viable. However, subscription models allowed heavy users to consume far more resources than they paid for, forcing providers to absorb losses.
The transition to usage-based pricing reflects the need to correct this imbalance. Under the new model, costs are directly tied to consumption, making AI spending more transparent but also more unpredictable. For enterprises, the shift introduces a new challenge: managing AI as a variable operational expense rather than a fixed cost.
At the same time, the return on investment for AI remains uncertain. While organisations continue to invest heavily in automation and productivity tools, many struggle to measure tangible business outcomes. Rising costs without clear returns are creating pressure on both vendors and buyers to justify spending.
Ultimately, the AI economics crisis is not about the failure of the technology itself. It is about aligning cost, value, and expectations. As pricing becomes more realistic, organisations will need to focus on high-impact use cases, optimise consumption, and ensure that AI investments deliver measurable results.
ソース：
