Midjourney's Pivot to Medicine Is the Most Surprising Announcement in AI This Year

Midjourney's Pivot to Medicine Is the Most Surprising Announcement in AI This Year

On June 18, 2026, Midjourney founder David Holz stood in front of an audience in San Francisco and announced a full-body ultrasound scanner he said was "in many ways superior to even MRI machines," able to complete a scan in 60 seconds, with no radiation and no magnetic fields, at an eventual cost of a few dollars. The company framed the announcement as the beginning of a new relationship between people and their own biology.

By Midjourney's own description, the current prototype takes roughly 20 minutes per scan and has been used on about 12 people. The team building it has nine members. The device has no FDA clearance for diagnosing anything.

That gap between the presentation and the reality is the central fact of the Midjourney Medical launch. It does not make the underlying technology implausible, but it clarifies how early the company's medical ambitions actually are.

What Was Actually Announced

The Midjourney Scanner is built around a technique called ultrasound computed tomography, or USCT, in which the body is submerged in water and surrounded by a ring of ultrasonic transducers. Each transducer fires sound pulses; the ring records echoes from every angle simultaneously; software reconstructs a 3D volume from the resulting data. The principle has been studied since the 1950s, and the FDA cleared a USCT device specifically for breast cancer screening in 2021. What Midjourney is attempting, whole-body USCT covering the full torso and legs, has not previously been cleared or independently validated at that scale.

The experience Midjourney designed is meant to minimize clinical anxiety. A user steps onto a motorized platform in a shallow pool. The platform descends into warm water at about five centimeters per second, passing the body through a ring of sensors that emit ultrasonic pulses and record the returning echoes from every angle. The company says the result is a sub-millimeter resolution, full-body 3D anatomical map comparable to an MRI. That comparison, like Holz's superiority claim, has not been assessed by any independent radiologist or medical institution.

The Chip That Makes It Work

The scanner's hardware rests on a licensing deal, not proprietary invention. In November 2025, Midjourney signed a co-development and exclusive licensing agreement with Butterfly Network, a publicly traded medical device company, for rights to its Ultrasound-on-Chip technology. According to an SEC Form 8-K filing, the deal commits Midjourney to a $15 million upfront payment and $10 million per year in licensing fees over a five-year term, with additional milestone and revenue-sharing payments totaling up to $74 million. Butterfly disclosed that the arrangement contributed $6.8 million in Q4 2025 revenue.

Butterfly's underlying technology is a genuine engineering achievement. The company replaced the piezoelectric crystals used in conventional ultrasound probes with capacitive micromachined ultrasonic transducers, or CMUTs, built directly onto standard CMOS semiconductor chips, the same fabrication process that produces computer processors. Each chip contains up to 9,000 individual elements, each capable of both emitting and receiving sound waves across a programmable frequency range of 1 to 12 megahertz.

Because the transducer elements and their signal-processing electronics sit on a single silicon die, the system eliminates the hundreds of analog wires that connect a conventional probe to its processing hardware and enables mass production at semiconductor-scale economics.

The Midjourney Scanner uses 40 of these chips per machine, approximately 8,960 active transducer channels in the current configuration. The system is paired with about two petaflops of computing power. During a scan it produces roughly 17 gigabytes of raw acoustic data per second. Reconstructing a single cross-sectional slice requires approximately 40 gigabytes of data. Midjourney describes the total data volume as equivalent to 500 hours of high-definition video for every second of scan time.

Where the Numbers Stand Now

The gap between the 60-second target and the 20-minute current reality is a data transfer bottleneck. The acoustic data the transducers produce arrives faster than the reconstruction computing cluster can receive and process it. Midjourney's engineers have not yet moved data quickly enough to complete the computational reconstruction in 60 seconds.

The company's roadmap targets a second-generation hardware design over the next 12 months. A third-generation scanner, planned for 2028 and led by Ahmad Abbas, who joined Midjourney in late 2023 after working on the Vision Pro at Apple, will use fully custom silicon designed by Midjourney rather than licensed from Butterfly. The company says that is when scan times and image quality will change substantially.

One data point from the announcement is traceable to an independent source. AuntMinnie, a medical imaging trade publication, noted that the system demonstrated a 0.93 correlation coefficient with MRI-based proton density fat fraction, a specific body composition metric. That figure comes from Midjourney's own research and has not been independently replicated. It suggests the device's composition estimates are in a reasonable range for that one measurement, though it does not constitute broad clinical validation.

The Wellness Framing

Midjourney is navigating FDA regulation by making no diagnostic claims. The agency requires clearance or approval for any device intended to diagnose, treat, or prevent disease. Midjourney is launching the scanner as a provider of "detailed body composition maps," outputs that describe muscle, fat, bone, and organ volume without asserting clinical diagnoses. This approach is consistent with the FDA's general wellness guidance for low-risk non-invasive physiologic measurement devices. Prenuvo and Ezra have operated whole-body MRI services for consumers using the same regulatory lane.

Holz said at the launch event that Midjourney has begun discussions with the FDA and intends to submit test results to incrementally pursue clearance for diagnostic capabilities over time.

The company's go-to-market plan bypasses hospitals and clinical settings entirely, at least for now. Midjourney is building a chain of branded facilities it calls Midjourney Spas, designed as wellness destinations that happen to contain scanners. The first location, planned for Union Square in San Francisco, is expected to open at the end of 2027 in roughly 25,000 square feet, housing 10 scanners alongside hot tubs, saunas, and cold plunges. The company describes the scan as "a side effect" of a spa visit rather than the reason for the trip.

By positioning scans inside wellness spaces, Midjourney keeps the product outside the clinical regulatory framework while building the scan volume and longitudinal dataset it would need to pursue diagnostic clearance later. The company has no outside investors and describes itself as a community-backed research lab funded by its image-generation revenue.

What Comes Next

By 2031, Midjourney aims to operate more than 50,000 scanners worldwide with a claimed capacity of one billion scans per month. That projection extends far beyond what the company has demonstrated. As of the June 18 announcement, the scanner had been used on roughly a dozen people. The company has never operated a regulated medical device or a consumer hardware product at any scale.

Independent radiologists and medical institutions have not assessed the device's resolution, artifact characteristics, or anatomical coverage across a full-body scan population. The literature on incidental findings from whole-body imaging in healthy people, including false positives that lead to unnecessary follow-up procedures, is substantial, and Midjourney's wellness framing does not address it.

What the announcement does establish is that the physics of whole-body USCT are achievable with commercially available semiconductor technology, that Butterfly Network's chip design is capable of the transducer density the approach requires, and that Midjourney has the resources to license credible hardware and build a working prototype. Whether a nine-person team with no hardware track record can close the engineering, regulatory, and operational distance between that prototype and a commercially deployed scanner remains an open question. The timeline the company has set for itself leaves very little room.