The Centers for Medicare and Medicaid Services released its national health expenditure projections this week, and the headline number is the kind that makes everyone nod and no one think: US health spending is on track to reach $9 trillion by 2034, climbing from roughly 18 percent of the economy today toward 20.6 percent.
Spending is projected to grow 5.4 percent a year against 4.1 percent GDP growth. When the thing you spend on grows faster than everything else, its share of the whole keeps rising. That is the entire story of the chart, and it is not, by itself, a crisis.
Here is the question the number does not answer, and the one that actually matters: what did we buy?
A bigger health bill is not the same as a sicker country. Some of the increase is demographic; an aging population uses more care, and that is a feature of success, not failure. Some of it is genuine medical progress that costs money and is worth it. But a meaningful share is something else entirely: a system that pays for volume and complexity rather than outcomes, and that has gotten very good at generating both.
Sit in an exam room for a week and the macro number resolves into something concrete. Patients are paying more every year, through premiums, deductibles, and out-of-pocket costs, and they still cannot get a same-week appointment. They get more imaging, more referrals, more portal messages, and more administrative friction. The spending is real. The experienced value, from the patient's side of the table, is the open question.
There are three places the money actually goes, and boards tend to scrutinize only two of them.
This is where the AI conversation belongs, and where it usually goes wrong. The promise of clinical AI is that it attacks category three: it removes administrative drag and lets clinicians spend more time on the care that works. The reality, too often, is that it gets pointed at category three and used to run the broken process faster. A model that speeds up a prior-authorization denial has not improved care. It has industrialized the friction. CMS just put a prior-authorization AI vendor on a corrective action plan for missing turnaround requirements; that is a preview of what happens when speed is prioritized over oversight on a clinical-adjacent decision.
"The $9 trillion number is not the problem to solve. It is the receipt. The real work is making sure that what it pays for is care, and not the cost of the system getting in its own way."
The boards that will navigate the next decade well are not the ones that pledge to spend less. Spending more on health is a defensible choice for a wealthy, aging society. They are the ones that get specific about category three, and that hold every new tool, including every AI tool, to a single test: does this remove the complexity, or does it just run the complexity faster?
The $9 trillion number is not the problem to solve. It is the receipt. The real work is making sure that what it pays for is care, and not the cost of the system getting in its own way.
Surgery-trained, currently practicing internal medicine (charity care). Advisory practice focused on clinical AI governance, vendor evaluation, and implementation strategy for health systems and health-tech companies. If you are on a board, an operating team, or a clinical-AI committee trying to draw that line, that is the work I do.
Weekly clinical reality on AI, governance, and what actually works inside health systems.