AI Tools Drive Major Scientific Advances as Usage-Based Pricing Replaces Subscriptions

Researchers are confronting higher costs for artificial intelligence tools as providers hike prices and tighten usage limits in response to mounting computing expenses. James Zou, who leads the AI for Science Laboratory at Stanford University in California, has spent well over US$100,000 on artificial intelligence in the past year.

He says the AI fees are in the same ballpark as the cost of supporting a postdoctoral fellow at Stanford.

Zou says we are entering into a new golden age of science with AI assistance that enables fundamental scientific advances because of their increasing capabilities of these AI scientist agents. @Nature reported that AI providers have been hiking up prices and tightening usage limits after struggling to make the economics work on subscription plans.

In January 2025, Sam Altman posted on X that OpenAI was losing money on its $200-a-month ChatGPT Pro subscriptions.

Sam Altman said people were using the chatbot more than OpenAI expected, driving up computing power and electricity costs. GitHub announced on 27 April that it would move GitHub Copilot from a subscription-based service to usage-based billing effective 1 June. GitHub cited the higher demands of agentic AI as the reason for changing its Copilot pricing policy.

Those shifts are reaching scientists who have grown reliant on the tools. Attila Gáspár, an economist at Central European University in Vienna, has been using AI to extract data from historical documents. For around 18 months, Attila Gáspár used his university-paid subscription to Claude without restrictions.

In late April, Attila Gáspár received the message "You have hit your limit" from Claude. Zou maintains the expense remains justified. He says the models are very useful for researchers, for coding, for analysis, for literature summaries.

The changes reflect broader pressure on AI companies facing higher-than-anticipated demand. OpenAI's experience with its premium tier illustrated how usage could outpace projections and inflate electricity and computing costs. GitHub's policy shift, set to take effect next month, marks the latest adjustment tied directly to the computational load of more advanced agentic systems.

Gáspár's experience shows the practical effect on individual researchers. After 18 months of unrestricted access funded by his university, the sudden limit arrived without warning in late April. His work pulling structured data from historical texts had come to depend on the AI chatbot's availability.

Zou's laboratory at Stanford continues to invest heavily even as bills climb. The biomedical-data scientist views the outlay as comparable to standard personnel costs in academia. His optimism centers on the expanding role of AI scientist agents in driving core discoveries that would otherwise move more slowly.