Midjourney Built a Medical Scanner. The Real Story Is What It Means for AI.
An AI image company just announced a full-body ultrasound scanner that lives inside a spa. The scanner is real. The 60-second scan is not. Here's what the pivot actually tells us.
You know Midjourney. Type some words, get a picture. Maybe you've used it to generate a hero image, a weird album cover, or that one portrait of your dog as a Renaissance nobleman. It's an AI image company. That's the whole thing.
So when Midjourney announced its first hardware product this week, nobody — and I mean nobody — guessed what it would be.
Custom inference chips? AI cameras? AR glasses? A pen?
No. Midjourney built a full-body medical scanner.
You step into a shallow pool of warm water. You sink slowly through a ring packed with half a million sensors. Sound waves fire through your body from every angle. Sixty seconds later, you have a 3D map of your insides. No radiation. No claustrophobic tube. Just water and ultrasound.
And it lives inside a spa.
I had to read the announcement twice. Then a third time. Then I checked the date to make sure this wasn't an elaborate prank. It's not. This is real, it's ambitious, and it tells you something important about where AI companies are going — even if this particular bet is half-brilliant and half-delusional.
The Scanner: Half a Million Dolphins
Here's the simplest version of how it works.
The Midjourney Scanner uses a technology called Ultrasound Computed Tomography (USCT). It's not new — researchers have been studying it since the 1950s, and it recently got FDA clearance for breast cancer screening. What's new is the scale.
Midjourney's ring contains about 500,000 tiny elements. Each one is the size of a grain of sand, and each can do two jobs: emit a sound wave and listen for the echo. Midjourney compares each element to a dolphin using echolocation. A full scan is like being surrounded by half a million dolphins, all pinging you at once.
The reason for the water is pure physics. Sound doesn't pass easily from your skin into air — it bounces off. That's why a regular ultrasound involves cold gel squeezed onto your skin. The gel lets the sound slip into your body. A full-body scanner just scales that up: put the whole person in water, and the sound enters smoothly from every side.
The data volume is staggering. Midjourney says the scanner produces about 17 gigabytes of raw acoustic data per second. Reconstructing a single cross-sectional slice needs roughly 40 GB. Their comparison: if you converted that data into HD video, you'd need to watch 500 hours of footage for every one second of scan time. A cluster of thousands of computers processes it all into a 3D image.
The Reality Check
Here's where the story gets interesting — and where you need to pay attention to the gap between the announcement and the actual machine.
The "60-second scan" is a goal, not a fact. The current prototype takes about 20 minutes. The team that built it is roughly nine people. Only about a dozen humans have been scanned so far. The first Midjourney Spa — where you'd actually use one — isn't scheduled to open in San Francisco until the end of 2027. And the very big numbers, like 50,000 scanners worldwide doing a billion scans a month? That's a 2031 ambition.
There's a second surprise that barely got mentioned: there is no AI inside the scanner yet. Midjourney is an AI company. This product has no AI in its imaging pipeline. David Holz, the founder, said so plainly. The actual imaging technology isn't even Midjourney's invention — they licensed it from Butterfly Network, a medical device company that makes ultrasound sensors on silicon chips.
The deal, signed in November 2025 and disclosed in an SEC filing: $15 million upfront, $10 million a year for five years, up to $74 million total. So a clearer description of this launch is: an AI image company took another firm's ultrasound chips, built them into a large scanner, and put it inside a spa.
That's not a dunk. It's actually the interesting part.
What Ultrasound Can and Can't Do
"Ultrasonic CT" and "like an MRI" sound impressive, but they mash together three machines that work in completely different ways.
A CT scanner takes X-ray photos from many angles. It uses radiation. It's excellent for lungs and bone.
An MRI uses powerful magnets. No radiation. Brilliant for soft tissue and the brain — but slow, loud, and expensive.
Ultrasound sends sound waves and reads the echoes. Cheap, safe, fast, no radiation. Best at soft, watery tissue near the surface: thyroid, breast lumps, kidneys, gallbladder, blood flow, a baby in the womb.
But ultrasound has two enemies, and both are physics: air and bone. Sound bounces off both. Your lungs are full of air, so ultrasound can't see deep inside them — which matters, because catching lung cancer early usually needs a CT. Your brain sits inside a bony skull, so a sound-based machine won't give you a brain map the way an MRI does.
A clever AI might someday read messy ultrasound data better than a human can. But AI can only find patterns in data the machine actually collected. It can't recover information the physics never let the scanner capture.
And no scan — not this one, not any — can reach inside you and fix a problem. A colonoscopy can spot a precancerous growth and remove it on the spot. That's prevention, not just detection. A full-body scan adds to the toolkit. It doesn't replace what's already there.
The Spa Strategy Is Actually Brilliant
Here's the part that made me sit up.
Midjourney's go-to-market plan avoids hospitals entirely. No insurance codes. No FDA approval for every diagnostic use case. No sales cycles that take years. Instead: build a spa, put scanners in it, and let people pay you to generate medical data while they relax in a hot tub.
The first Midjourney Spa is planned for Union Square in San Francisco — 25,000 square feet, 10 scanners, hot tubs, saunas, cold plunges. Open 24/7. The scanning is designed as a side effect of visiting. You barely think about it, but you accumulate a library of health data over time.
This is a data acquisition strategy disguised as wellness.
Building a scanner and selling it to hospitals means fighting through regulatory hell at every step. Building a spa means people hand you their bodies voluntarily, happily, repeatedly. You get longitudinal data — the same person scanned over months and years — which is the holy grail of preventive medicine. Most healthcare compares you against a population average. A cheap, repeatable scan compares you against you. Did anything grow? Shrink? Move?
That's genuinely valuable. It's also the part that could turn into a nightmare — cheap scans for everyone will surface countless harmless oddities, leading to anxiety, extra tests, and medical bills for things that were never going to hurt you. We don't have the data yet to know how often that happens.
Midjourney is threading the regulatory needle carefully. The FDA requires clearance for any device that diagnoses disease. So Midjourney is launching with "body composition maps" — muscle, fat, bone density, tissue changes. No diagnosis. That's the same wellness lane that companies like Prenuvo and Ezra already use for whole-body MRI services. It sidesteps the strictest FDA enforcement. Over time, they plan to submit results to unlock diagnostic capabilities.
What This Means for AI Companies
Here's the bigger story, and the reason I'm writing about this on a blog about AI automation.
"AI company" is becoming a meaningless label.
Midjourney started as a text-to-image tool. Now it's building medical hardware. OpenAI started as a research lab — now it's reportedly losing $20 billion a year while building consumer apps, enterprise tools, and custom silicon. Anthropic makes a chatbot and also lobbies governments about national security. xAI builds a chatbot and also launches rockets (okay, that one's cheating — that's just Elon).
The point is that AI is no longer a product category. It's infrastructure. It's a capability that gets bolted onto everything — scanners, coding tools, browsers, terminals. The interesting companies aren't asking "what AI product should we build?" They're asking "what industry becomes possible when you assume AI is free, fast, and embedded?"
Midjourney looked at medical imaging and asked: what if scanning was so cheap and easy that people did it while getting a massage? That's not an AI question. It's a systems question. AI is just one component — eventually — of the answer.
This is the same shift happening in software. The question isn't "can an AI write code?" anymore. It's "what does a development workflow look like when an AI agent runs your terminal, reads your files, and ships features while you sleep?" The companies winning right now aren't building chatbots. They're rebuilding entire workflows around the assumption that intelligence is cheap.
The Honest Take
Midjourney just pulled off one of the most unexpected product reveals in recent memory. The scanner exists. The images are real. The engineering blog reads like actual documentation, not a press release.
But the gap between the announcement and the reality is enormous. A nine-person team that has never built a regulated medical device is promising to reinvent preventive healthcare. The 60-second scan is a 20-minute scan. The "superior to MRI" claim is unverified by any independent radiologist. The billion-scans-per-month target is five years away. The custom silicon that would supposedly make it all work doesn't exist yet.
And yet — the spa strategy is clever enough that it might not matter whether the scanner reaches its most ambitious specs. If Midjourney can get people into the water cheaply and repeatedly, the longitudinal data alone is a genuine innovation. The company is privately held, profitable, and has no investors to answer to. That gives them the runway to be patient.
I was wrong about what Midjourney was building. So was everyone else. And maybe that's the real lesson: the companies that surprise you are the ones worth watching. The ones doing exactly what you expect are usually the ones running out of ideas.
If you want to see what happens when AI stops being a chatbot and starts being a tool that actually does things — browse the web, run terminal commands, automate your work — check out CopperRiver. It runs on your Mac, uses open-source models, and doesn't require a 25,000-square-foot spa. Plans start at $9/month.