The Fix

Posted December 2, 2025

In October of 2024, I wrote a blog called “The Black Hole” which described the escalating costs of healthcare that at the time were taxing the US economy with total costs exceeding $4.8 trillion annually. Unfortunately, the black hole has only gotten bigger.

Today, total healthcare costs in this country are estimated to exceed $5.7 trillion, which consumes almost 20% of our entire economy. If you want some visuals around how bad the problem is, I recently spoke to some of our engineers at an offsite. Video’s here.

Not only are healthcare costs out of control, but they continue to grow at an obscene rate of ~7%. Given that our economy grows much more slowly, if we don’t do something and do it quickly, healthcare costs will become an anchor around our country’s neck that we can’t escape from.

Just do the math. If healthcare continues to grow at its current rate, within 25 years it will become a ~$30 trillion spend. To put that number in perspective, that is the size of our entire economy today.

Worst of all, despite the cost (we spend roughly as much on healthcare as the next largest 15 countries), we aren’t producing better outcomes. Average life expectancy in the US is ~78 years, which is well below many other nations whose citizens live on average well into their early 80’s.

Ok, so we have a problem. But is there a solution?

You’ve likely heard many people blame particular segments of the healthcare industry. For example, it’s common rhetoric to attack drug companies. But even if every single pharmaceutical company in the US made zero profit, it would only save ~$112 billion annually, or 2% of our costs. Likewise, it’s common to attack insurance companies. But their profits pale in comparison to drug companies’, so eliminating their profits would save us even less.

The real culprit is waste (which encompasses inefficiency and errors). By most estimates, waste accounts for ~20-30% of total healthcare spend, with some estimates as high as 40%. Even taking the lower end, if we could avoid errors, mistakes, and other waste, we would save ~$1.5 trillion annually.

Everything else is too small to focus on. We need to go after waste.

The challenge as you can imagine is that waste isn’t easy to root out. We have 1.1mm doctors, 5mm nurses, and over 20mm people working in healthcare who wake up every day trying to do what’s best for their patients. But the system is stacked against them.

We have too many middlemen, too many rules and regulations, too much process, too little data, etc. We ask care teams to make data driven decisions without sufficient data and then when something goes wrong, there is a line of personal injury lawyers waiting to go after them.

To bend the curve, we need tools that help physicians and care teams avoid mistakes before they occur. We need tools that help them become more efficient. Those tools also have to be fully embedded within their workflow or they will never gain adoption.

These tools must be smart, data driven, and ubiquitous. Capable of reading and reacting to medical records and complex multimodal healthcare data in real time. Hmm – where would we find such a tool?

AI, and in particular large multimodal models, have the potential to solve this problem, which I’ve also written about.

To encourage companies to invest in bringing AI to healthcare, and to gain mass adoption, we need to align incentives. So here is a possible solution:

FDA is the governing body of AI solutions
- Companies looking to deploy complex AI solutions into the market to inform treatment decisions need to go through the FDA. The FDA is the governing body that already oversees medical device efficacy and safety, which includes many AI products and algorithms.
If FDA approved, provide a 2-3 year provisional CMS reimbursement period
- Once approved, bundle AI products with guaranteed, but provisional, CMS reimbursement, which would cover a significant percentage of patients given the reach of Medicare.
- The rate of reimbursement should be based on both the cost of developing the AI and its estimated value to the system based on information provided by the AI developer, with oversight and the final rate being set by CMS.
- Once a reimbursement rate is established, the AI should be broadly covered for a period of 2-3 years, which will encourage adoption – let’s call this the provisional reimbursement period.
Put the burden for demonstrating clinical and economic utility on the AI provider during that period
- At the end of the provisional reimbursement period, the AI provider should be required to submit data to CMS demonstrating two things: (1) the device worked as intended, as in it saved lives or improved care and (2) the device was cost effective for CMS, as in even after paying the AI provider, Medicare saved money while improving outcomes, which is the main focus of value based care.
- The burden to demonstrate both elements using real world evidence should fall squarely on the AI provider. If they are successful, CMS will make reimbursement permanent and add it to their fee schedules or national coverage determination (NCD) policies. If the AI provider fails, their reimbursement is automatically terminated and they can’t reapply for that AI for some period of time, say a year.
- This will naturally encourage AI providers to set a “fair” price for their AI as they have to defend it, with real world data, in order to maintain coverage.
Ensure the program is minimally cost neutral to administer
- And in order to pay for the administrative burden of this program, we add a significant CMS user fee for AI applications that is material enough to discourage poor quality applications and allow FDA and CMS to manage this program effectively.

This system naturally encourages 3 things:

Investments in AI. Companies will now have a path to reimbursement.
Adoption of AI. Without reimbursement, the system has no incentive to change behavior
Value from AI. We force companies to bear the burden of proving that their AI works – that it’s improving care and saving the US healthcare system money.

Only AI has the potential to catch mistakes early enough in the system where they can be rooted out. The same analogy applies to spacecraft. Catch a faulty ring early in the manufacturing process and you can replace it for $5. Catch it too late and you have a $65 million rocket exploding into pieces.

If we can hypercharge our investments in AI with a clear path to adoption, we have a real chance at letting the free market propagate technology, early enough and broadly enough, to reduce waste and error.

And that might be “the fix.”

Sources