CVS Health Plans to Split Itself Up, and It Tells Us a Lot About Where AI in Healthcare Goes Next

What happens when one of the biggest names in American healthcare decides it might work better as several smaller companies? That is the question now sitting over CVS Health, which has plans to split itself up. Instead of one giant machine, the idea is to break a sprawling operation into more focused pieces.

Most people will read this as a money story. A story about structure, debt, and what shareholders want. But there is a bigger story underneath it, and it is about software. A breakup changes how data moves. It changes who owns that data and who gets to build on top of it. And that changes almost everything about how artificial intelligence actually gets used inside a national healthcare network.

So let's look at the CVS Health split through an AI lens: what it signals, what it puts at risk, and what other companies should copy right now.

What Is Actually Happening

CVS Health plans to split itself up. That is the headline. It is not a small reshuffle of departments or a new logo on a slide deck. The intent is to separate a business that has grown to touch many different parts of healthcare at once.

Think about what a company like CVS Health holds together. There is a retail pharmacy counter filling prescriptions. There is a benefits business that negotiates drug prices and handles claims. There is an insurance arm that carries risk. There are walk-in clinics and virtual care. Each of those pieces has its own customers, its own regulators, its own profit math, and its own data.

Those pieces have been sitting inside one company. A split says: maybe they shouldn't be.

Why a Breakup Could Make Sense Now

Big usually wins. Scale gives you buying power, brand trust, and the ability to spread fixed costs across more customers. But scale has a cost too, and that cost has been rising.

First, there is the focus penalty. When one company owns five very different businesses, every big decision has to survive five sets of priorities. That slows everything down. In a world where a competitor can ship a new model or a new workflow in weeks, slow is expensive.

Second, there is regulation. Pharmacy rules, insurance rules, and clinical care rules do not look alike. They do not move at the same speed. Bundling them means the strictest rule sets the pace for everything else.

Third, there is capital. Investors like clean stories. A company that does five things is harder to value than five companies that each do one thing well.

And fourth, the one nobody puts on the slide, there is AI.

The AI Reason Nobody Is Saying Out Loud

Here is the uncomfortable truth about enterprise AI: it is not really a technology problem anymore. It is a plumbing problem.

Most AI projects fail not because the model is bad, but because the data is stuck. It sits in a system owned by one team, governed by one legal contract, locked behind one permission model. When you ask an AI system to answer a question that crosses those boundaries, the answer is usually "no", or worse, a wrong answer delivered with total confidence.

Now imagine that problem inside a company that does pharmacy, insurance, benefits, and clinics. A single question like "is this patient likely to stop taking their medication?" needs data from all four. The model is easy. The permissions are brutal.

Splitting a company up does not automatically fix that. But it does something important: it makes the boundaries explicit. Instead of one giant tangled data estate where nobody is fully accountable, you get clean lanes. Each business owns its data. Each business owns its models. Each business owns its risk.

That is a much easier place to build from.

Why Focused Units Ship AI Faster

Small, focused teams have a real advantage with AI, and it comes down to four things.

1. Clear ownership of data

When one business unit owns a dataset, that unit can decide what to do with it without a six-month negotiation. It can label it, clean it, and train on it. It can measure results and stop what isn't working.

2. Narrower, cheaper models

A focused business rarely needs a giant model that knows everything. It needs a model that is very good at one thing. Predicting refill timing. Flagging a claim that needs review. Matching a patient to the right clinic slot. Smaller, targeted systems are cheaper to run, easier to test, and far easier to explain to a regulator.

3. Faster feedback loops

A pharmacy team can see within days whether an AI tool is helping the counter. A clinic team can see it within hours. When the people who feel the pain also own the fix, improvement gets fast.

4. Cleaner accountability

This one matters more than people think. If an AI system makes a bad call, someone has to answer for it. In a mega-company, that conversation gets fuzzy. In a focused company, it doesn't. That clarity is what lets risk teams say yes instead of maybe.

The Real Risk: Losing the Data Flywheel

There is a counter-argument, and it is a strong one.

The whole promise of a company like CVS Health is the loop. A prescription gets filled. A claim gets processed. A clinic visit gets logged. When all of that lives in one place, you can learn things no single piece could learn alone. That loop is a moat. It is also the exact raw material that makes the most valuable AI possible.

Split the company badly and you break the loop. You get five businesses that each see a slice and none that see the whole. You get worse AI, not better.

So the split only works if three things travel with it:

• Data-sharing agreements that are real. Not vague promises. Specific contracts that say what moves, how fast, and under what rules.
• Shared identity and interfaces. If a patient, a member, or a prescription has a different ID in every new company, nothing will connect.
• Portable models. Any AI built before the split should still run after it. That means standard formats, standard evaluation, and clean documentation.

Companies that plan a split without planning for these three things usually lose a year of progress. The ones that plan for them can come out faster than they went in.

What This Means for the Future of AI

Step back from CVS Health and look at the pattern. This is not one company's story. It is a preview of a much larger shift.

AI becomes the connective tissue between separate companies

For the last decade, the big enterprise bet was consolidation: buy everything, put it in one data lake, and build from there. That era is cooling. What comes next looks different. Instead of one giant data lake, we get many well-run data ponds, connected by software.

AI will be that software. Expect a lot more investment in clean interfaces, shared standards, and privacy-preserving techniques that let models learn across companies without moving raw data at all. The winning architecture will not be "one company owns everything." It will be "many companies that can talk to each other safely."

AI agents will start crossing company lines

Once separate companies share standard interfaces, software agents can act across them. An agent handling a prior authorization could check a plan's rules, pull clinical notes from a clinic, and confirm stock at a pharmacy, three different companies, one automated workflow. That is genuinely new, and it only becomes possible when the boundaries are clean.

Governance gets easier inside and harder outside

Smaller units are easier to govern. But you now need rules for what happens at the seams, where one company's model touches another company's data. Expect contracts to start looking like software. Expect "data terms" to become a real engineering discipline.

Buy versus build flips back to buy

Smaller, focused companies rarely have large AI research teams. That pushes them toward vendors, out-of-the-box tools, and partners. The result is a market where a few strong AI platforms get embedded deeply into healthcare, which raises a new set of concentration questions a few years down the line.

A Practical Playbook for Business Leaders

Whether or not your company is splitting, the same lessons apply. Here is what to do now.

• Map your data boundaries before you need to. Know which datasets live where, who owns them, and who can approve access. If you cannot answer that in a week, you have a problem.
• Give every AI project a named owner. One person, one business outcome, one budget. Shared ownership is the fastest way to kill a project.
• Prefer narrow over giant. Start with a model that does one job well and can be measured. Expand only after it proves value.
• Write your data contracts like code. Specific, testable, and versioned. Vague language about "sharing insights" helps nobody.
• Keep identity consistent. Invest in a shared way to identify customers, patients, and transactions. Every AI system you build will lean on it.
• Plan for model portability. Document your models, track your data sources, and avoid lock-in to a single toolchain.
• Measure business results, not model scores. Fewer missed refills. Faster claim turnaround. Fewer no-shows. Those are the numbers that matter.

What to Watch Next

Three things will decide whether this split becomes a strength or a stumble.

One: the data terms. If the new entities announce real, specific agreements for sharing data across the seams, the AI story gets better. If the language stays vague, start worrying.

Two: leadership continuity. The people who built the AI systems need to land somewhere with a mandate to keep going. When key people leave during a split, the institutional knowledge goes with them.

Three: the customer experience. Healthcare is one journey across many companies. If a patient suddenly has to explain themselves three times, the split has failed on the most human test there is.

The Bottom Line

CVS Health planning to split itself up is being sold as a financial story, and it partly is one. But the deeper signal is about how AI will be built over the next several years.

The age of "buy it all, pile it up, and let a giant model figure it out" is fading. What is replacing it is quieter and more practical: focused businesses, clean data boundaries, standard interfaces, and AI that connects separate organizations instead of swallowing them.

That is a harder architecture to build. It is also a far more durable one. The companies that get this right will not just survive a split, they will come out of it faster, cheaper, and smarter than the giant they used to be.