AI Companies Are Cutting Hospitals Out of Healthcare

American healthcare has long operated on a fixed sequence: a patient enters a healthcare institution, a physician orders imaging, a radiologist reads it, and the results flow back to the institution that ordered it. AI is breaking that sequence apart.

Leading AI companies are inverting the order, reaching the consumer first and leaving the clinical system to manage what comes next.

Over the past three years, Amazon, Apple, Google, and Microsoft have each built consumer health products that sit between patients and the traditional clinical system, bypassing hospital procurement entirely. In June, AI image generation company Midjourney took that shift one step further. It announced a full-body ultrasonic scanner, a physical device in a physical room, that it plans to deploy in consumer wellness venues called Midjourney Spas.

</> Raw HTML Block — preview on published page

The AI companies are looking to bring costs down drastically. An MRI that can cost up to $4,000 may now cost just $100 with the advanced imaging solutions. 

2 Models, One Problem

Radiology is facing a supply and demand mismatch. The American College of Radiology reported in February that radiologist workforce shortages ranked as the single biggest structural threat to the field for the third consecutive year, with rising imaging volumes ranked second.

Two distinct responses have emerged, and they point in opposite directions.

Incumbent imaging hardware manufacturers are solving the problem through institutional integration. GE HealthCare acquired imaging software company Intelerad for $2.3 billion in November 2025, positioning itself as what analysts called a "digital diagnostics consolidator" by building AI-enabled workflows that it sells to hospitals and health systems.

At the Radiological Society of North America's annual meeting last December, GE unveiled its Signa One MRI platform, which weaves AI across pre-patient planning, scanning, and image reading, all within the clinical procurement channel. Siemens Healthineers presented comparable AI-enabled radiology services at the same conference, specifically targeting staffing and workflow pressures at health systems.

Consumer AI companies are solving the same problem by removing the hospitals from the equation. Midjourney's scanner, built by a nine-person team led by a former Apple Vision Pro engineer, submerges users in a shallow pool of warm water and fires sound waves through their bodies using 358,000 transducers. Founder David Holz has set a target of 1 billion scans per month, with the first Midjourney Spa planned for San Francisco by the end of 2027.

</> Raw HTML Block — preview on published page

The Price Compression and What It Unlocks

The consumer imaging market has been compressing costs at a pace institutional players cannot match.

Preventive healthcare startup Prenuvo (the market leader in Whole-body MRI) charges $2,499 for a comprehensive scan. Ezra, acquired by Function Health in May 2025, dropped its base scan price to $499 after the deal. Holz, without specifying the exact price, has described Midjourney's target price as “extremely reasonable,” while noting the raw scan has "almost no price."

Each price reduction is enabled by the same regulatory mechanism. Under the FDA's general wellness policy for low-risk devices, companies that market imaging output as a "body composition map" rather than a clinical diagnosis operate entirely outside medical device regulation, in the same category as fitness trackers. No diagnostic clearance is required. No insurance reimbursement is involved. The company captures the scan revenue; the healthcare system absorbs what follows.

What follows is the more expensive part. A hospital diagnostic MRI of a single body part costs $1,000 to $4,000, depending on insurance. A hospital CT abdomen with contrast runs comparable figures. Prenuvo covers the full body for $2,499. But those hospital costs exist within a reimbursable, physician-ordered, clinically indicated framework. Consumer scans do not, and the follow-up they generate enters the system without that framework in place.

Who Inherits the Follow-Up

The clinical term for an abnormal finding in an asymptomatic patient is an incidentaloma. Research published in Radiology Advances identifies downstream follow-up costs as the central economic burden of whole-body screening programs, with a significant share coming from findings that ultimately prove benign. In a large population study of asymptomatic individuals, 30.3% had at least one incidental MRI finding; among those who underwent biopsy, 62.1% showed no malignancy.

Prenuvo and Ezra have already established the handoff model: the company delivers the scan report, and follow-up care is directed to the patient's own healthcare providers. At the scale Midjourney is targeting, that handoff becomes a structural transfer of clinical and financial burden to health systems that had no role in generating it.

Steve Forthuber, Eastern Operations CEO of outpatient imaging company RadNet, described conditions in the system that would receive that burden. "There's a severe shortage of technologists in radiology while there's an increased demand for imaging services, particularly the outpatient imaging," Forthuber had said in a conversation with AIM Media House in February. 

 Read: “We Want to Utilize AI to Bridge That Gap,” Says RadNet Eastern Operations CEO 

The AI That Isn't There Yet

Meanwhile, at the June launch event, Holz acknowledged that no AI is involved in the scanner's image reconstruction. "We're not even using any AI in this yet, just really cool hardware and software," he said, observing that the prototype takes approximately 20 minutes per scan, and it plans to bring it down to a minute. 

The contrast with validated AI imaging tools is significant. Viz.ai's stroke detection software, which received FDA De Novo clearance in 2018, identified large vessel occlusions faster than neuroimaging specialists in more than 95% of cases, saving an average of 52 minutes per stroke case. RadNet's Enhanced Breast Cancer Detection (EBCD) program improved cancer detection rates by 21% across all races and ethnicities, more than double the 9% improvement the industry recorded when mammography shifted from 2D to 3D. Both results came through FDA clearance, peer-reviewed research, and integration into existing clinical workflows.

Forthuber, whose company signed a letter of intent with GE HealthCare in May to expand AI-powered imaging across modalities, says the gap between what is technically possible and what is cleared for clinical use is a known condition of the field. He said that the technology is available today in some cases, but emphasized that they are not yet there due to regulatory and reimbursement hurdles, which means progress won't happen overnight.

Midjourney's own roadmap places AI-assisted reconstruction in later hardware generations and treats its third-generation device, projected for around 2028, as the point at which it pursues proper regulatory submissions. The hardware being built now is the distribution infrastructure for the AI that the company intends to layer in later.

GE HealthCare and Siemens Healthineers are watching that trajectory from inside the institutional channel, with billions in acquisition spend and decades of clinical relationships behind them. The question is not whether one model will replace the other, but which reaches a meaningful scale first. Currently, the health systems sit at the intersection of both, where consumer-generated health data meets clinical decision-making.