As hospitals scale AI tools, consumption-based pricing models threaten budgets and ROI, forcing health systems to impose strict usage controls.
Hospital finance teams are confronting an unexpected line item: the cost of artificial intelligence tokens. What was once an obscure technical metric has become a critical budget concern as healthcare systems deploy large language models across clinical and administrative operations.
Tokens represent the basic unit that large language models use to process text. Models break down words and phrases into smaller fragments, convert them into numerical representations, and process them sequentially. According to Becker's Hospital Review, roughly every four characters translates to one token, meaning a single query can consume thousands or even millions of tokens before administrators realize the scope of consumption.
The pricing model itself is straightforward: AI vendors bill healthcare organizations based on both input tokens (the prompt) and output tokens (the generated text). However, the cumulative impact is proving problematic. Many hospitals have deployed ambient scribes, EHR copilots, and revenue cycle tools that consume tokens continuously behind the scenes, even when vendors market them as flat per-provider subscriptions.
Cost Spirals Threaten Return on Investment
Tomas Gregorio, senior vice president and chief digital information officer at Care New England, described the consumption-based model as potentially catastrophic. "It's consumption-based and that'll cripple your organization and not give you the ROI you want," Gregorio told Becker's Hospital Review in June.
The challenge intensifies as Epic Systems reports that more than 85 percent of its customers now actively use its AI tools, including Art, Emmie, and Penny applications. These tools operate on Microsoft's Azure OpenAI service, exposing health systems to unpredictable token-based expenses at scale.
The problem extends beyond healthcare. Technology companies across sectors face similar pressures. Uber exhausted its entire annual budget for agentic AI systems in just three months, prompting the rideshare platform to impose restrictions on employee AI usage. Salesforce is actively tracking whether token consumption correlates with measurable business outcomes, while Google has reported a sevenfold increase in token usage year over year.
Health Systems Implement Usage Controls
Facing escalating costs, major healthcare institutions are restructuring access to AI tools. Houston Methodist has restricted Microsoft Copilot to a select group of employees exploring how to develop custom AI agents internally. The system automatically terminates any agent that incurs costs without demonstrable financial benefit, according to Michelle Stansbury, associate chief innovation officer and vice president of IT applications.
As health systems transition from pilot projects to enterprise-wide implementations, several factors amplify token consumption:
Longer clinical documentation generates more tokens
Wider context windows in AI models require additional processing
More complex models consume tokens at higher rates
For chief information officers and chief financial officers, the token model creates a direct relationship between usage and expenditure. Unlike traditional software licensing, where costs remain relatively stable regardless of adoption rates, AI token pricing scales with every interaction.
As healthcare organizations embed AI deeper into clinical workflows and administrative processes, understanding token economics is becoming as essential to health IT leadership as evaluating subscription pricing, EHR integration, and regulatory compliance. The hidden expense of AI tokens is rapidly becoming a visible strategic consideration shaping how hospitals deploy and govern artificial intelligence systems.
This article was originally published on AI Glimpse.
Top comments (0)