THE $5.3 TRILLION HEALTHCARE CRISIS

AI Is About to Do to Healthcare What Healthcare Has Been Begging For. The Industry Won't Like How It Feels.

Preface

CitriniResearch recently published a piece called “The Consequences of Abundant Intelligence” - a scenario exploring what happens when AI disruption delivers everything the bulls expect and the economy still loses. It’s one of the best things written on AI’s second-order effects. We’d strongly encourage you to read it.

We read it and couldn’t stop thinking about healthcare.

Healthcare is $5.3 trillion. It’s 18% of US GDP. It employs nearly 18 million Americans - more than any other industry. It is the largest single line item in the federal budget. It is the number one reason Americans go bankrupt. And it is built, from floor to ceiling, on the assumption that clinical knowledge is scarce, that administrative complexity is permanent, and that the friction between a patient and their treatment is just how things work.

What if AI proves all of those assumptions wrong? What if that’s the best thing that ever happened to patients - and the worst thing that ever happened to the healthcare industry?

What follows is a scenario, not a prediction. We wrote it as a story - a narrative that traces how this could unfold, one domino at a time - because stories reveal causal chains that bullet points miss. The present-day numbers are real. The future events are imagined. The point is not to tell you what will happen but to map what could happen.

---

Part One: The Numbers That Set the Table

Everything in this section is fact. This is where we are today.

The United States spent $5.3 trillion on healthcare in 2024 - $15,474 per person, a 7.2% increase from the prior year. For context, the next-closest developed nation spends about 12-13% of GDP on healthcare. We spend 18%. The gap is enormous and it’s growing.

What do we get for this? US life expectancy was 77.4 years in 2022 - ranking 49th globally, behind Cuba, Estonia, and Saudi Arabia. Among the 38 OECD member states, we rank 30th. The average life expectancy in comparable wealthy countries is 82.5 years. We spend more than anyone, by a lot, and live shorter than almost all of our peers.

Where the money goes: Medicare consumed $1.12 trillion (21% of total spending). Medicaid took $932 billion (18%). Private insurance accounted for $1.6 trillion (31%). The federal government and households are the two largest payers, at 31% and 28% respectively.

Administrative costs now account for more than 40% of total hospital expenses. The industry spends an estimated $40 billion per year just on billing and collections. The broader administrative transaction ecosystem runs roughly $200 billion annually. Initial claim denial rates hit 11.81% in 2024 - nearly twelve cents of every dollar billed bouncing on first submission.

The median hospital operating margin in 2024 was about 4.9%. That was a good year. But 40% of US hospitals were still operating in the red. From 2017 to 2024, 62 rural hospitals closed while only 10 opened. The Chartis Group estimates 30% of remaining rural hospitals are currently vulnerable to closure.

The AAMC projects a physician shortage of up to 86,000 by 2036. The average medical school graduate carries $212,000 in debt. Medical school applications dropped to 51,946 in 2024-25 - the lowest since 2017-18 and the third straight year of declines - before rebounding 5.3% in 2025-26.

GPT-4 scored above 86% on USMLE Step 1 and above 90% on Steps 2 and 3. Specialized medical models have outperformed board-certified physicians on complex differential diagnosis in blinded studies.

That’s the setup. A system that costs more than any in human history, produces mediocre outcomes, runs on razor-thin margins, is drowning in administrative waste, is short on doctors, and is staring down an AI that can already pass the medical licensing exam better than most humans who take it.

Now here’s the story of what could happen next.

---

Part Two: The Quiet Part

The disruption didn’t start where anyone expected.

Every AI-in-healthcare story for the past decade opened the same way: a neural network reads a chest X-ray and outperforms the radiologist. It made for great headlines and great conference keynotes. It was also, commercially, almost irrelevant. Regulatory barriers, physician resistance, malpractice liability questions, and payer reimbursement rules created a moat around clinical practice that no algorithm could cross - at least not quickly.

The back office had no such moat.

Think about the claims processing system for a second. Hospitals employ thousands of people to submit claims. Insurers employ thousands of people to deny them. Hospitals then employ thousands more people to appeal the denials. Both sides use coding specialists, compliance analysts, and prior authorization coordinators whose entire professional existence is navigating a bureaucratic obstacle course that adds zero clinical value to anyone.

It is, if we’re being honest, a jobs program disguised as a payment system. And it works only because it’s complicated enough to require human labor.

AI revenue cycle management tools don’t find it complicated. They find it trivially easy. An LLM can read an operative note, assign the correct ICD-10 and CPT codes, generate a clean claim with proper modifier logic, submit it, handle the denial, draft the appeal with supporting clinical documentation, and resolve the case - not for simple office visits, but for complex multi-procedure surgical cases with payer-specific bundling rules that took experienced human coders years to learn.

In our scenario, this is where the story begins - not with a dramatic medical breakthrough, but with a quiet, boring, and devastatingly efficient automation of the billing department.

The health systems didn’t resist. They’d been drowning in administrative costs for decades. When AI tools started demonstrating first-pass claim acceptance rates above 93% - compared to the industry’s 88% baseline - every hospital CFO in America had the same reaction: finally.

Within eighteen months, the large systems cut administrative headcount by 30-40%. Rural hospitals and smaller practices, which had outsourced these functions entirely, simply cancelled their contracts. The administrative services companies - the Waystars, the R1 RCMs, the healthcare BPO divisions - didn’t shrink gradually. They contracted fast. Their revenue was the system’s waste. When the waste disappeared, so did they.

The part nobody thought about was the humans. The median salary for a medical coder was around $48,000 in 2024. For a billing specialist, roughly $42,000. These were not high earners with savings cushions. When they lost their jobs, consumption dropped immediately. And there were millions of them - scattered across every metro area and mid-size city in America, because healthcare administration is everywhere.

That was the first domino.

---

Part Three: The Diagnostic Reckoning

The second domino took longer to fall, because it required something the first didn’t: regulatory permission.

AI had been reading medical images with superhuman accuracy for years. The technology wasn’t the bottleneck. The question was whether the system - payers, regulators, hospitals, physician groups - would allow AI to replace (or significantly supplement) human diagnostic judgment in clinical practice.

In our scenario, the dam broke on reimbursement.

CMS - under pressure to slow the growth of Medicare spending, which was consuming $1.12 trillion annually and accelerating - issued guidance allowing AI-validated imaging interpretation at a reduced reimbursement rate. The math was immediate: why pay $100 for a radiologist-read scan when an AI reads it more accurately, faster, and for $8?

The American College of Radiology challenged the rule. They lost. The data was too clear - AI-read imaging had lower false-negative rates across virtually every modality. Radiology group revenues fell dramatically. Fellowship applications collapsed.

Pathology followed the same arc. Dermatology - where diagnosis is essentially pattern recognition on visual inputs - was next. Teledermatology platforms replaced the $200 specialist visit with a $15 AI assessment, and payers were happy to pay for it.

But the real seismic event was AI-powered primary care.

The AAMC had been projecting a shortage of 20,200 to 40,400 primary care physicians by 2036. In our scenario, AI didn’t fill that shortage. It made the shortage irrelevant. AI primary care systems could conduct a full patient intake, review medical history, order and interpret labs, manage chronic conditions, adjust medications, and escalate to human specialists only when warranted.

Patient satisfaction, counterintuitively, was higher. The AI never rushed. It never forgot to ask about medication adherence. It was available at 3 AM. It had access to every published piece of medical literature in real time - a knowledge base no human could match.

It turned out that what patients always meant by “the human touch” wasn’t the human part. It was the time and attention. AI provides both in unlimited supply.

The economic premium on diagnostic reasoning - the thing that justifies a decade of training, four years of residency, and $212,000 in debt - began compressing. Not disappearing. Compressing. When an AI can do 80% of what a primary care physician does, available 24/7, at a fraction of the cost, the question isn’t whether demand for physician labor falls. It’s how fast and how far.

This was the second domino: not the elimination of doctors, but the erosion of the scarcity premium that made medicine one of the last reliably high-income professions in America.

---

Part Four: The Hospital Paradox

The third domino was the cruelest, because it was built entirely out of good news.

AI started doing what healthcare reformers had been begging for since the 1990s. Diagnostic AI reduced unnecessary testing. Clinical decision support shortened hospital stays. Chronic disease management tools kept patients out of the hospital entirely. Remote monitoring caught deterioration early enough to prevent emergency admissions.

Every one of these is a clinical triumph. Every one is also a volume reduction. And American hospitals run on volume.

A hospital makes money by doing things to patients - procedures, imaging, surgeries, bed-days. When the median operating margin is 4.9% and 40% of hospitals are already in the red, even modest volume declines break the math. The fixed cost structure - the building, the equipment, the 24/7 staffing minimums, the malpractice coverage - doesn’t scale down with volume. A hospital at 80% occupancy barely breaks even. At 65%, it’s dead.

In our scenario, the rural closures accelerated first - they were already fragile, with 30% estimated as vulnerable before AI entered the picture, and 69% of closures from 2014 to 2024 occurring in non-Medicaid-expansion states. Then suburban hospitals started merging or converting to outpatient-only facilities. Then the urban community hospitals.

The political optics were impossible. AI was delivering better outcomes, but hospitals were closing. Patients were healthier, but communities were losing their largest employer. Try explaining to a town of 12,000 people that the hospital closure is good news because an AI on their phone manages their diabetes better than the ER did.

They’ll ask you: who’s going to hire us?

They won’t be wrong to ask. Healthcare is the largest employer in the majority of US states. Nearly 18 million Americans work in the sector. BLS data from 2024 projected the industry would add 1.9 million job openings per year on average through 2034. It was the one industry that always hired - through recessions, through bubbles, through everything.

In this scenario, the always-hiring industry starts always-firing. Not nurses or surgeons or physical therapists - their work still requires physical presence. But the 60% of healthcare workers who don’t touch patients - the administrators, coders, billers, diagnostic technicians, insurance processors, pharma sales reps, and much of the physician workforce itself - face a structural compression that no prior economic model accounts for.

And unlike tech layoffs that concentrate in a few coastal cities, healthcare job losses are distributed across every county in the country. When the hospital closes in a rural town, the tax base shrinks, the school suffers, the grocery store loses customers. The dominos don’t stop at the hospital door.

---

Part Five: The Insurance Inversion

UnitedHealth Group generated $400 billion in revenue in 2024. The top five health insurers collectively managed well over $1 trillion in annual premiums.

Health insurance exists because healthcare is expensive and unpredictable. Insurers pool risk, negotiate rates, process claims, and manage utilization. In exchange, they take a cut.

In our scenario, AI attacks both sides of this model.

On the cost side: the ACA’s Medical Loss Ratio rules require insurers to spend 80-85% of premiums on actual medical care. If AI-driven efficiency reduces the cost of care, insurers don’t pocket the savings. They rebate them or cut premiums. Their revenue base shrinks mechanically as the system gets more efficient. Efficiency is, quite literally, bad for insurance revenue.

On the administrative side: the same AI that automated hospital billing departments simultaneously automated insurer claims processing, utilization review, and prior authorization. Both sides of the administrative war get automated at the same time.

But the deepest threat is more fundamental. Insurance exists to manage uncertainty. What happens when AI makes healthcare predictable? When models can stratify disease risk with actuarial-grade precision, when continuous monitoring catches cardiac events weeks early, when genomic analysis can predict chronic disease onset with high accuracy - the uncertainty that justifies the pooling-and-intermediation model begins to evaporate.

The emerging model looks less like insurance and more like a subscription: AI-powered continuous care at a flat monthly rate, with catastrophic coverage for the genuinely unpredictable. The legacy insurers - with 600,000 direct employees, Byzantine provider networks, and entire corporate structures whose reason for existing is friction and complexity - don’t have a natural role in that future.

---

Part Six: The Pharma Paradox

Pharmaceutical R&D is the one corner of this scenario that’s almost purely good from a patient perspective. Drug discovery timelines compressing. Novel targets becoming druggable. Personalized regimens generated from genomic data. AI-designed antibiotics overcoming resistant strains.

The average cost to bring a new drug to market was $2.23 billion in 2024, according to Deloitte, with total ecosystem investment per approval exceeding $5 billion. AI is compressing those numbers dramatically for firms that are AI-native from the start.

The paradox is financial. The pharma business model depends on scarcity: enormous R&D cost justifies monopoly pricing during patent protection. AI collapses both the cost and the defensibility. When a lean biotech with access to the same foundation models as Pfizer can design a competitive molecule in months, the entry barriers that supported 80% gross margins start to crack.

In this scenario, drug pricing falls - not because of government mandate, but because competition explodes. Wonderful for patients. Terrible for the $200-500 billion market cap pharma incumbents priced on the assumption that their pipelines are defensible moats.

---

Part Seven: The Question Nobody in Healthcare Wants to Hear

Here’s the part that kept us up at night writing this.

What if this scenario is... good?

Not the job losses. Not the hospital closures. Not the stock drawdowns. Those are real pain. But the thing underneath all that pain - the actual transformation?

We run the most expensive healthcare system in human history and rank 49th in life expectancy. Administrative complexity eats 40% of hospital expenses. A billing dispute between a hospital and an insurer has never cured anyone’s cancer. The prior authorization process delays treatment for weeks while clerks argue about documentation.

If AI strips all of that out - the trillion-dollar back office, the $200 specialist visit that could be a $15 assessment, the 12-cent-on-the-dollar denial rate, the entire friction economy - and replaces it with care that is continuous, personalized, cheaper, and available whenever the patient needs it...

That’s the system every politician, every policy wonk, every frustrated patient, and every burned-out physician has been asking for. For decades. Well, in this scenario, here it comes. You just might not love how it arrives.

The stocks crater. The jobs disappear. The hospitals consolidate. But the patients? The patients do better. The outcomes improve. The costs fall. The system, for the first time in 50 years, starts working the way it was supposed to.

Healthcare might be the one sector where AI disruption is simultaneously the most financially destructive and the most socially beneficial. The industry breaks because it succeeds. It spent decades saying “we need to lower costs and improve outcomes.” AI does exactly that - to the industry, not for it.

---

Part Eight: The Incumbents’ Albatross and the Small Company Advantage

This is the part that matters for why you read this newsletter. And it requires us to say something that might seem counterintuitive at first:

The companies best positioned for this disruption don’t have to be AI companies.

That’s worth sitting with, because the instinct when you read a piece like this is to go looking for the “AI healthcare play” - the pure-play diagnostic AI platform, the computational drug discovery startup. Those companies exist and some of them will do very well. But the larger, more durable advantage belongs to a broader category: small healthcare companies that don’t carry the cost structure of the old system.

Here’s what we mean.

Think about what a large-cap healthcare incumbent actually is. HCA Healthcare operates 186 hospitals. It has billions in physical plant, tens of thousands of beds, and a reimbursement model that depends on those beds being full. Community Health Systems. Tenet. Same story. They are building-heavy, staff-heavy, fixed-cost-heavy organizations designed for a world where hospital volume grows every year forever.

The big pharma names - Pfizer, Merck, J&J, Lilly - employ tens of thousands of people in sales, marketing, regulatory, and commercial infrastructure built to support $2 billion drug development cycles and monopoly pricing during patent windows. Their entire organizational architecture assumes those economics persist.

These companies aren’t going to die. But in the scenario we’ve described, every one of them faces a structural headwind: the very infrastructure that made them dominant becomes a liability when the system it was designed for starts shifting underneath them. They have to restructure, and restructuring a $400 billion company is slow, expensive, and painful. The market tends to punish it.

Now think about a micro-cap biotech with 50 employees. It doesn’t have a 10,000-person billing department to restructure. It doesn’t have $3 billion in hospital real estate on its balance sheet. It doesn’t have a legacy IT stack that costs $500 million a year to maintain. It doesn’t have a 30,000-person sales force whose value proposition is being eroded by AI-informed prescribing.

It has a pipeline, a small team, and a cost structure light enough to pivot with the market instead of against it.

This isn’t about AI. A small specialty pharma company with a differentiated drug in a niche indication benefits from this disruption even if it never touches an AI tool - because the incumbents it competes against are distracted, restructuring, and weighed down by the very infrastructure that used to be their moat. When the big companies are spending two years and billions of dollars trying to transform their organizations, the small companies are just... operating. Unencumbered.

The same logic applies across every sub-sector:

• Small pharma and biotech. The large pharma companies face margin compression as AI collapses R&D costs and barriers to entry. Their response will be what it always is: M&A. They’ll buy the small companies with the most promising pipelines, the novel mechanisms, the clean cap tables. The acquirers have to restructure. The targets just have to execute. A micro-cap biotech with a Phase 2 asset in a space where AI is accelerating competitive pressure becomes more valuable as an acquisition target, not less - because it represents the fastest path for a bloated incumbent to stay relevant.

• Small-cap providers and care delivery companies. When hospitals consolidate and close, the care has to go somewhere. Hospital-at-home companies, outpatient specialty clinics, remote monitoring platforms, and telehealth providers absorb the volume. They don’t carry the fixed costs of a 500-bed hospital. They don’t need 4.9% margins to survive because their cost structure is fundamentally lighter. Even non-AI-native small providers benefit simply because they’re built for a distributed care model that the hospital-centric incumbents are too heavy to pivot toward quickly.

• Small-cap services and tools companies. Revenue cycle management is a trillion-dollar category being disrupted. The incumbents in that space - the large BPOs, the outsourcing giants - are as exposed as the hospitals they serve. But smaller companies providing next-generation billing tools, compliance platforms, or payer-provider connectivity solutions don’t carry the legacy contracts, the offshore headcount, or the organizational inertia. They can adapt, integrate new capabilities, and win business from health systems that are actively looking for leaner vendor relationships.

• Specialty insurtech and benefits platforms. The large health insurers are bureaucratic machines. A small company offering direct primary care, a lean benefits administration platform, or a specialty-focused insurance alternative doesn’t have 600,000 employees and a provider network built over 40 years. It has a clean model purpose-built for lower costs and less friction. In a world where the entire point of the industry shifts toward simplicity and efficiency, the companies that were already simple and efficient have a structural head start.

The point is this: you don’t need to find the next great AI healthcare platform to benefit from this disruption. You need to find the companies that are light while the incumbents are heavy. The disruption creates margin pressure, restructuring costs, strategic distraction, and organizational paralysis at the top of the market cap spectrum. At the bottom - where the micro-caps live - those headwinds don’t exist. The same forces that crush the large caps create space for the small caps to operate, grow, and get acquired at premiums.

---

What We’re Watching

We want to be clear one more time: this was a scenario. We are not predicting that hospitals collapse next year, or that physicians face mass unemployment, or that UnitedHealth Group posts a quarterly loss.

We are saying: the pieces are on the board. The AI capabilities are real and improving every quarter. The financial fragilities - 40% of hospitals in the red, an 11.81% claim denial rate, $212,000 in med school debt per graduate, a system that spends more than any country on earth and ranks 49th in life expectancy - those are not speculative. Those are today’s numbers, published by CMS, the BLS, the AAMC, and the insurers themselves.

The question is what happens when a technology that gets meaningfully better every quarter meets an industry that employs nearly 18 million people and was built on the assumption that clinical intelligence would always be scarce and expensive.

Healthcare has always been the “safe” sector. Recession-proof. Demographically protected. Too regulated to disrupt. Too essential to fail.

All of that was true in a world where clinical intelligence was scarce and administrative complexity was permanent. If those things change - and we think the probability is higher than the market reflects - then the safe sector isn’t safe. It’s in transition. And transitions reward the nimble over the entrenched.

That’s where we’re focused. More soon.

Nico

---

Disclaimer: The content provided in this newsletter is for informational purposes only and does not constitute financial, investment, or other professional advice. The opinions expressed here are those of the author and do not necessarily reflect the views of The Clinical Edge. Investing involves risk, including the possible loss of principal. Past performance is not indicative of future results. The author may or may not hold positions in the stocks or other financial instruments mentioned. Always do your own research or consult with a qualified financial advisor before making any investment decisions. To read our full disclaimer, click here.