AI costs surge as companies shift to usage-based pricing, forcing researchers and enterprises to rethink strategies

AI Costs Rival Postdoc Salaries as Providers Struggle with Economics

James Zou, a biomedical-data scientist at Stanford University, has spent over $100,000 on AI tools in the past year—roughly equivalent to funding a postdoctoral fellow 1

1

. While Zou believes the investment is justified for coding, analysis, and literature summaries, this high cost of powering AI is becoming a critical issue across research institutions and enterprises. The financial burdens for researchers are mounting as AI providers fundamentally restructure how they charge for their services 1

1

.

OpenAI CEO Sam Altman revealed in January 2025 that the company was losing money on its $200-per-month ChatGPT Pro subscriptions because users consumed more computing power and electricity than anticipated 1

1

. This admission signals a broader industry challenge: the economics of flat subscription models simply don't work for AI companies facing skyrocketing infrastructure demands.

GitHub Copilot and Anthropic Lead Shift from Flat Subscriptions

GitHub announced on April 27 that it would transition GitHub Copilot from subscription-based service to usage-based pricing starting June 1, citing the higher demands of agentic AI 1

1

. The company paused new Copilot Pro signups in April while preparing for this transition, acknowledging that absorbing escalating inference costs is no longer sustainable 2

2

3

3

.

Anthropic has similarly changed its billing framework for the Claude model, moving enterprise customers from fixed fees to usage-based charges 3

3

. OpenAI has introduced a wider range of pricing tiers to better capture different usage patterns 2

2

. Unlike the software-as-a-service era where seat-based subscriptions made sense, AI agents process instructions by consuming tokens, and more tokens mean more computing costs for providers 3

3

.

Rising API Costs Force Companies to Burn Through Annual Budgets

The impact of these pricing changes is already visible in corporate spending patterns. According to The Information, Uber CTO Praveen Neppalli Naga stated in an internal memo that the company burned through its 2026 AI budget in just four months 3

3

. Multiple instances show employees exhausting a year's worth of AI budget within months as AI services becoming more expensive across the board 3

3

.

For Latentforce, an AI modernization platform, token costs skyrocketed from ₹20,000 per month six months ago to ₹2-3 lakh per month currently, even before Claude's usage-based model fully took effect 3

3

. Nvidia, which develops the majority of chips powering the AI revolution, now says that AI can cost more than human workers 2

2

.

AI Adoption Strategies Shift as Some Choose Human Capital Over AI Productivity Tools

The cost pressures are forcing companies to reconsider their AI adoption strategies. Aravind Jayendran, cofounder of Latentforce, reports that clients in the knowledge services sector are hiring people rather than deploying AI for tasks like data entry and documentation processing, as human capital is currently cheaper 3

3

. One founder posted on Reddit that his company cancelled five AI subscriptions and hired two mid-level developers instead 3

3

.

Many Indian enterprises still rely more on human capital than AI given the high costs of running and maintaining the systems, according to a founder who optimizes large language model usage for enterprises 3

3

. Research published in a March preprint suggests users trading down to downgraded AI models could end up costing more by using more tokens, even if each unit is less costly 1

1

.

Managing AI Spending Through Guardrails and Hybrid Models

Companies where AI tools remain essential are implementing strategies for managing AI spending. A Bengaluru-based unicorn founder revealed his company has put agentic guardrails in place to ensure tokens are used for approved projects and flag when usage hits upper limits 3

3

. Latentforce is exploring optimization through a mixture of open source and frontier models 3

3

.

Atomicwork, which helps clients automate employee requests and IT service workflows, now offers hybrid pricing—a flat rate with token limits and an option for additional usage with extra payment. "Enterprises need visibility on budget and fixed cost helps with that. However, they will need to pay for their consumption as the AI companies have also increased the cost of APIs for new models," said cofounder Vijay Rayapati 3

3

.

Scientific Research Faces Access Inequality and Usage Limits

For researchers, the pricing changes create significant challenges. Attila Gáspár, an economist at Central European University who uses AI to extract data from historical documents, encountered usage limits on his university-paid Claude subscription after 18 months of unrestricted access 1

1

. Geoscientist Matteo Niccoli upgraded from Claude Pro to Max and still hits limits on heavy workdays during multi-session work involving repeated back-and-forth between coding, reasoning, and analysis 1

1

.

Sebastian Baltes, a software-engineering researcher at Heidelberg University, fears that "such changes mean the gap between institutions, groups and individuals that can pay for premium subscriptions and those that can't will widen" 1

1

. Jeremy Howard, honorary professor at the University of Queensland, worries about inequality, noting that AI models have given students in parts of the world with low disposable income "the same access to the kind of intelligence that rich people do" 1

1

.

The ROI for AI investments remains under scrutiny as researchers grapple with additional overhead. Niccoli describes the bottleneck as "thinking and the discussion" around checking model outputs, noticing when context becomes overloaded, and knowing when answers drift—making AI useful but not necessarily a labor-saver 1

1

. As usage-based pricing becomes the norm, organizations must carefully evaluate whether the benefits of AI productivity tools justify their escalating costs.