Hardware costs and availability aren’t the only critical factors
The AI Factory has become a hot topic in healthcare. As a refresher, an AI Factory is infrastructure built to manage the entire AI lifecycle, from data ingestion through training and fine-tuning to high-volume inference.
When you consume a public model, someone else owns the guardrails, the evaluation, and the audit trail. When you invest in an AI Factory, you own all of it. It’s no surprise that hospitals and health systems welcome that level of ownership and control.
Any time a healthcare organization starts looking seriously at an AI Factory, questions typically focus on hardware costs, quantities, and lead times. Those are fair questions, but there’s another one to ask first: Does your organization have a working AI Center of Excellence (CoE)?
If the answer is no, it suggests your organization isn’t ready to make the most of AI, and the AI Factory is unlikely to deliver the value you hope. It’s a common scenario.
Is adoption outrunning readiness?
Research from the Center for Connected Medicine at UPMC and KLAS Research found that over 90% of health systems have deployed third-party AI, mostly in clinical documentation and revenue cycle. But only 44% have dedicated testing environments, and 63% describe their strategy as developing or ad hoc, with limited resources, time, and specialized talent cited as the leading barriers. UPMC’s Chief Medical Information Officer Rob Bart, MD, put it plainly: “Implementation is only the first step.”1
The governance picture is thinner still. In a survey of 182 hospital leaders, Black Book Research found that only 22% were highly confident they could produce a complete AI audit trail within 30 days for regulators or payers, and just 29% had implemented and enforced policies covering AI model inventory, lineage, and signoffs.2
If an organization can’t yet produce an audit trail for a vendor-supplied documentation tool, it is not positioned to own the full lifecycle of a model it trains itself. An AI Factory doesn’t reduce that burden. It moves every bit of it in-house.
Defining what an AI Center of Excellence is
An AI Center of Excellence is a dedicated, cross-functional unit that centralizes AI expertise and standards to drive responsible adoption across the enterprise. It sets strategy, writes guidelines, supports compliance, and owns risk.
Published frameworks recommend a steering committee or CoE with real C-suite representation (including a chief AI officer or chief data scientist alongside the CIO, CMIO, and CMO) and built on a vision articulated at the executive level.
In healthcare, I would widen that further. Legal, compliance, HR, revenue cycle, nursing leadership, and practicing physicians all belong in the room. After all, they are the people who will either use these models or answer for them. And no organization should spend millions building infrastructure to train its own models before it has identified which departments, workflows, and decisions get better because of these investments.
A functioning AI CoE owns three questions that no vendor can answer for you:
-
Which use cases are real, and which are pilots that will never scale?
-
What happens to patient data — where it lives, who reaches it, how it’s classified?
-
What proves the model behaved the way it was supposed to?
The reality is that an AI Center of Excellence takes months to stand up. It pulls in people who don’t report to IT. And it doesn’t result in a quick demo. But it’s the difference between an AI Factory that compounds value and a very expensive cluster running three pilots.
The AI CoE’s first assignment is your data
Once the AI CoE exists, the work starts with data because an AI workflow is only as good as the data and the process underneath it. If your systems are broken, an agent inherits the breakage.
The scale of that problem in healthcare is well documented. An estimated 80% to 90% of healthcare data is unstructured — clinical notes, imaging, radiology and lab reports, patient communications — and it stays hard to operationalize because it lacks consistent formats and metadata. The same analysis reiterates that governance can’t be retrofitted. Access controls, tagging and confining sensitive data, auditing, and encryption must be architected into the pipeline from the start.
That breaks into real disciplines, including clearing duplication and data rot, classifying data so access boundaries are enforceable, and building immutable records so there’s a clear trail. Each merits greater discussion, which is why our CTO Mark Giles is going deeper on these topics at the Midwest Healthcare Innovation Summit 2026 hosted by Bamberg Health. He’ll be part of a panel entitled From Data to Discipline: AI, Analytics & Enterprise Execution on Wednesday, September 30.
How we help
If your organization doesn’t yet have an AI Center of Excellence, we can help you design one, including who sits in it, what it owns, and where each workload should live. If you already have one and it has done its work, we can help you architect, build, deploy, host, and operate the infrastructure. CBTS is a Dell Titanium Partner, the highest tier in Dell’s channel program, and we work across every major OEM platform.
Before the budget conversation and before the GPU count, ask the question that predicts the outcome. It isn’t a matter of what you can afford. It’s whether you're ready to own it.
Citations:
1. https://www.upmc.com/media/news/080626-ai-use-in-healthcare-systems-upmc-research
2. https://www.newswire.com/news/u-s-hospitals-underfund-ai-governance-as-adoption-accelerates-22674017