---
title: "Why 40% of Healthcare Leaders Pause AI Over Unpredictable Costs, and What Fixes It"
slug: "healthcare-ai-unpredictable-costs-google-survey-pilot-to-scale"
author: "ibl.ai"
date: "2026-11-18 12:00:00"
category: "Premium"
sector: "healthcare"
topics: "healthcare AI costs, Google Cloud healthcare survey, HLTH 2026, AI pilot to production, unpredictable AI costs, tokenomics, token costs, hospital AI budget, healthcare agentic AI, AI vendor dependence, usage-based AI pricing, self-hosted healthcare AI"
summary: "A Google Cloud survey of 250+ healthcare executives found only 30% have scaled agentic AI and 40% cite unpredictable costs as a reason for pausing. The fix is a cost architecture the health system can meter, cap and move between models."
banner: ""
thumbnail: ""
linkedin: |
  A new Google Cloud survey asked more than 250 healthcare executives where AI stands.

  Only 30% say they have successfully scaled advanced or agentic AI across their operations, 13 points behind the all-industry benchmark. 40% cite unpredictable costs as a reason for pausing AI initiatives. More than 80% say cost is a key hurdle to reducing dependence on their existing AI vendors.

  Google's Aashima Gupta put it plainly: "You don't see that when you're running a 100-person or 200-person pilot."

  The pilot hides the bill. At scale, every prompt, retrieval step and agent action is a token, and a health system that cannot see, cap or reroute those tokens cannot forecast them.

  Three things fix it: per-agent cost visibility, a budget cap the health system sets, and the freedom to move each workflow to a cheaper model without a migration project.

  #iblai #HealthcareAI #HLTH2026 #AIAgents #HealthIT
---

## The Short Answer

**Healthcare AI stalls at scale because costs that were invisible in a pilot become unforecastable in production. The fix is architectural: meter every agent, cap spend, and route each workflow to the cheapest model that does the job. On ibl.ai you own all the code and the data, so the health system controls all three.**

The numbers come from a Google Cloud survey of more than 250 healthcare executives, conducted by Hypothesis Group and [reported by Healthcare Dive](https://www.healthcaredive.com/news/healthcare-struggles-to-scale-ai-past-pilot-stage-google-report/874609/) on November 16, 2026, as part of its HLTH 2026 coverage.

## What did Google Cloud's 2026 survey find about healthcare AI scaling and costs?

Google Cloud's survey found that just 30% of healthcare organizations report successfully scaling advanced or agentic AI across their operations, trailing the all-industry benchmark by 13 percentage points.

Nearly 60% of the executives said they had faced integration and deployment challenges moving enterprise AI from pilot to a live system. And 40% cite unpredictable costs as a reason for pausing AI initiatives.

Two more findings frame the cost problem. More than 80% of respondents cite cost as a key hurdle to reducing dependence on their existing AI vendors. About 40% named protection of sensitive data as an especially important consideration, nine points above the benchmark.

## Does the survey say 9 in 10 healthcare organizations are scaling AI?

No. Some social posts about the survey claim 9 in 10 healthcare organizations are scaling AI while 4 in 10 abandon it. Healthcare Dive's report of the Google Cloud survey contains no 9-in-10 figure, and its headline finding runs the other way: only 30% have scaled.

The 40% figure is also narrower than "abandon." The survey says 40% cite unpredictable costs as a reason for pausing AI initiatives, not that 40% have walked away from AI.

That distinction matters for a CFO reading it. A paused initiative is a budgeting problem that can be solved; the rest of this post is about how.

## Why do healthcare AI costs look affordable in a pilot and unpredictable at scale?

Healthcare AI costs look affordable in a pilot because the pilot is small and runs one workflow. "You don't see that when you're running a 100-person or 200-person pilot," Aashima Gupta, Google Cloud's global director of healthcare, told Healthcare Dive.

At scale, usage stops being a fixed line item. Every prompt, response, retrieval step and agent interaction consumes tokens, as Accenture's [CIO's Guide to AI Tokenomics](https://www.accenture.com/content/dam/accenture/final/accenture-com/document-fy27/q1/Accenture-The-CIOs-Guide-To-AI-Tokenomics.pdf) puts it, and agentic workflows multiply them.

Gupta's other point explains the multiplication: "Healthcare is not one workflow — healthcare is many workflows." Prior authorization, coding, discharge summaries and patient messaging each have their own volume, context length and model needs, so one pilot's unit cost says little about the next ten.

Thin hospital margins make the variance hard to absorb. Gupta went as far as saying "healthcare needs a chief token officer."

## What is AI tokenomics, and how much can it save a health system?

AI tokenomics is the discipline of measuring and managing what AI consumes against the value it returns. Accenture's guide defines it as connecting "what AI consumes to the value it returns."

Accenture projects that without optimization, token costs grow about 44% over 24 months, driven by 78% volume growth partly offset by a 19% price decline.

Organizations adopting tokenomics disciplines can avoid that projected increase, about $4.4 million for every $10 million of annual spend.

One correction to how that number has been repeated: it is Accenture's projection of avoidable growth, not a measured saving that companies have already banked.

Accenture also reports that only 20 cents of every dollar spent on tokens can today be expressed as a quantified financial outcome to a board or CFO, rising to 32 cents with chargeback and workload-level measurement.

## What cost architecture lets a health system scale AI without unpredictable bills?

A health system scales AI predictably when it can do three things itself: see cost per agent and per workflow, cap spend before it happens, and move a workflow to a cheaper model without rebuilding it.

<table>
<thead><tr><th>Cost problem the survey names</th><th>What the health system needs</th><th>How it works on ibl.ai</th></tr></thead>
<tbody>
<tr><td>Costs invisible in the pilot, unforecastable at scale</td><td>Cost per agent, per workflow, per department</td><td>Per-agent usage and spend in <a href="/docs/os/analytics/costs">Costs analytics</a></td></tr>
<tr><td>40% pause over unpredictable costs</td><td>A spend ceiling set in advance</td><td>Usage-based billing against a budget cap the health system sets</td></tr>
<tr><td>Many workflows, one model for all</td><td>Model choice per workflow</td><td>Per-agent LLM selection, model-agnostic across any provider</td></tr>
<tr><td>80%+ see cost blocking less vendor dependence</td><td>A switch that does not require a migration</td><td>The health system holds the source code and switches models anytime</td></tr>
<tr><td>Sensitive data, nine points above benchmark</td><td>Agents that read data in place</td><td>Self-hosted inside the health system's own perimeter</td></tr>
</tbody>
</table>

Google is pushing an approach that uses different models for different use cases, per Healthcare Dive, rather than one model for all of healthcare. That only saves money if switching a workflow's model is a setting, not a contract renegotiation.

## Why does per-seat AI pricing make healthcare costs harder to forecast?

Per-seat AI pricing makes healthcare costs harder to forecast because it charges for headcount while the real cost driver is workload. A clinician who never opens the tool costs the same seat as the one running hundreds of prior authorizations a week.

Seats also hide the token economics underneath, so the 20-cent visibility gap Accenture describes stays closed to the buyer. We worked the per-seat versus usage arithmetic for a 5,000-clinician system in [AI cost math for hospitals](/blog/ai-cost-math-for-hospitals-per-seat-vs-usage).

ibl.ai runs with no per-seat pricing, model-agnostic, and can deploy anywhere from the health system's own cloud to a fully air-gapped network. Usage is metered, so the token bill is visible, capped and attributable to a workflow.

## What should a health system do before scaling its next AI pilot?

Before scaling the next pilot, a health system should be able to answer four questions from its own data, because 40% of its peers paused when they could not.

First, what does one completed task cost, per agent, at the pilot's volume? Second, what volume does production imply, workflow by workflow? Third, what is the spend cap, and what happens when it is hit? Fourth, which workflows could run on a smaller model at acceptable quality?

The fourth question is where the 19% price decline Accenture projects over 24 months becomes the health system's saving rather than the vendor's margin, but only if the health system can move the workload.

ibl.ai is family-owned and operated from New York, NY, and its engineers build alongside a health system's own team, so the cost model is designed with the people who will run it.

*Sources: survey findings and Aashima Gupta quotes from [Healthcare Dive, November 16, 2026](https://www.healthcaredive.com/news/healthcare-struggles-to-scale-ai-past-pilot-stage-google-report/874609/), reporting a Google Cloud survey of 250+ healthcare executives by Hypothesis Group; token-cost projections from [Accenture, The CIO's Guide to AI Tokenomics](https://www.accenture.com/content/dam/accenture/final/accenture-com/document-fy27/q1/Accenture-The-CIOs-Guide-To-AI-Tokenomics.pdf). Healthcare Dive's report of the survey gives 30% as having scaled agentic AI and contains no 9-in-10 figure. The 40% figure is a reason for pausing, not abandoning.*

## Why does owning the AI stack matter?

**ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.**

- **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours.
- **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform.
- **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ.
- **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity.

1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.

ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.

[Explore sovereignhealthcare.ai](https://sovereignhealthcare.ai/)
