AI companies are moving up the value chain and looking at more complex uses of their core technologies.
Midjourney, best known for turning text prompts into striking images, has announced its first hardware product — and it has nothing to do with image generation. At an event in San Francisco on June 17, founder David Holz unveiled the Midjourney Scanner, a full-body ultrasound machine built under a new division called Midjourney Medical. The company says the scanner can produce a 3D internal map of the human body in about 60 seconds.
How The Scanner Works
The mechanics are specific. The person being scanned steps into a shallow pool of water and is slowly lowered through a sensor ring at five centimetres per second. Water is essential to the design: ultrasound travels cleanly through it, allowing the sensors to image the body from every angle simultaneously rather than pressing a probe against one spot at a time. The system uses 40 Butterfly Ultrasound-on-Chip modules and 8,960 individual transducers, processes 17 gigabytes of data per second, and requires roughly 40GB of raw data per reconstructed cross-sectional slice. Total compute is around two petaflops.
The output is closer to a CT scan in its structure than a conventional ultrasound — a full 3D reconstruction of muscle, fat, bone, and organs — without the ionizing radiation of CT or the powerful magnets required by MRI. Midjourney is calling the technology “Ultrasonic CT.”
Holz has described the prototype as roughly ten times cheaper and sixty times faster than a conventional MRI, where scans currently take one to two hours and cost between $400 and $4,000. Whether those comparisons hold at equivalent image quality remains to be seen independently.
The Hardware Partner
The most substantive detail in the announcement is who built the hardware. Midjourney signed a co-development and licensing agreement with Butterfly Network in November 2025, securing exclusive rights to Butterfly’s ultrasound-on-chip technology for a defined field of use. The deal involved a $15 million upfront payment and $10 million in annual licensing fees, with potential milestone payments up to $9 million — part of a five-year arrangement that Butterfly has disclosed could total $74 million. Butterfly’s Q4 2025 results confirmed the partnership contributed $6.8 million in revenue that quarter alone.
Butterfly Network is not a startup with an idea. It has FDA-cleared ultrasound products and a long track record in semiconductor-based ultrasound imaging. That partnership lends technical credibility to Midjourney’s claims in a way that a purely internal engineering effort would not.
The hardware lead is Ahmad Abbas, who joined Midjourney in late 2023 after working on the Vision Pro at Apple.
What It’s Actually For — At Least Initially
Midjourney is not claiming diagnostic capabilities for the scanner, and the framing here is deliberate. Diagnostic imaging devices in the United States require FDA clearance, and the company has none. The stated early use case is body composition mapping: repeatable scans that show changes in muscle mass, fat distribution, bone, and organs over time — the kind of longitudinal data that appeals to people tracking fitness or diet, not patients seeking a disease diagnosis.
Holz’s own comments reinforce that the technology is earlier than the announcement might suggest. “We’re not even using any AI in this yet, just really cool hardware and software,” he said. The company uses AI to segment and label scan outputs, but the imaging itself runs on ultrasound physics and signal processing. About a dozen people have been scanned so far.
The regulatory path, if Midjourney eventually pursues diagnostic claims, would require narrowly scoped FDA applications built on clinical validation data the company has yet to generate.
The Spa Model
The commercial plan is unusual for a medical device company. Midjourney is building what it calls the Midjourney Spa — the first location planned near Union Square in San Francisco by end of 2027. The facility will house around ten scanners alongside hot tubs, saunas, cold plunges, and a gym. The framing is wellness infrastructure, not a clinic.
That positioning keeps Midjourney below the FDA’s line for now, while letting the company accumulate scan data, refine the hardware, and build a user base. A third-generation machine using custom silicon, aimed at substantially higher image quality, is planned for 2028, when the company also intends to pursue broader city expansion.
The longer-term targets Midjourney has published are striking: 50,000 scanners deployed globally within six years and one billion full-body scans per month. Those are company ambitions, stated on a day-one announcement, with roughly twelve people having used the device. They say more about the size of the vision than the current state of the product.
The Bigger Picture
What Midjourney is attempting sits at the intersection of several things happening in AI right now. Compute built for generative model training is being redirected toward different kinds of dense tensor workloads — in this case, ultrasound reconstruction and segmentation. Meta’s recent licensing of Midjourney’s aesthetic technology for its own models shows one dimension of how AI companies are expanding their reach; Midjourney Medical shows another. The company is betting that medical imaging can be repositioned from a rare, expensive clinical event into a routine longitudinal record that people accumulate over years.
The gap between that bet and a proven product is still wide. Midjourney has no regulatory clearance, no published clinical validation, and a hardware track record that starts today. What it has is a credible hardware partner, a specific technical approach, and a founder willing to build something with no obvious connection to the company’s existing business.
That last part is either the most interesting thing about this announcement or the reason for the most skepticism, depending on how you weight ambition against execution history.