Following Microsoft's recent adjustment to GitHub Copilot pricing, billing methods for AI tools have once again become a focus of attention for enterprise users. The new plan, implemented on June 1st, fundamentally shifts from a relatively fixed subscription model to a more granular token-based billing system, widening the cost gap between different models.
High-performance models are more expensive
This adjustment has sparked discussion, primarily not just because of the shift to pay-as-you-go pricing, but because the models most frequently used and highly rated by businesses are often the ones experiencing the most significant price increases. For development teams, AI tools are gradually transforming from efficiency software expenditures into budget items requiring separate management.
Some enterprise users have reported that while they were initially encouraged to use AI tools to improve productivity, the management logic quickly shifted to controlling consumption after the new pricing was implemented. Without employee-level usage limits, a single high-frequency user could consume a large portion of the budget in a short period.
Businesses are starting to tighten credit limits
The article mentions that some large enterprises have already incorporated AI usage into their internal management. Previously, many companies emphasized expanding AI coverage, even regarding usage frequency as an efficiency indicator. However, under the token-based billing model, enterprises must also pay attention to how much was used, which model was used, and the corresponding costs.
Uber is seen as a microcosm of this trend. The report mentions that the company discovered its AI budget was being consumed faster than expected within a short period, and subsequently began setting stricter usage limits and employee restrictions. This reflects that even large enterprises with high technology investment need to reassess the return on investment of AI tools.
- Enterprises are shifting their focus from whether to adopt AI to how to control costs.
- The stronger the model's capabilities, the higher the token price usually corresponds.
- Budget management is becoming a real problem in the implementation of AI.
Industry profit pressure is being transmitted
Behind this discussion lies broader commercialization pressure. As leading AI companies like OpenAI and Anthropic move forward with fundraising and IPO preparations, market demands for revenue and profit may further increase. For model vendors, continuously subsidizing high-cost inference services is not easy.
The article also mentions that early AI subscription products were not fully based on a mature cost model. As enterprise usage scaled up, inference costs, caching costs, and model call frequency all increased, putting pressure on the original pricing system. For customers, AI is no longer just a fixed-monthly-fee tool, but more like a cloud computing resource that requires real-time monitoring and allocation.
AI-driven procurement shifts towards refined management
Some companies have begun to establish separate cost dashboards for model calls, tracking expenditures by model, token type, and call scenario. This indicates that AI usage management is expanding from an internal technical department matter to a decision-making issue involving development, finance, and management.
While the industry continues to emphasize how AI can improve productivity, businesses are now more realistically considering whether the costs can be absorbed by their operations. The changes brought about by token-based billing may not diminish businesses' demand for AI, but it will alter how they deploy it.
