Midjourney CEO David Holz announced the Midjourney Scanner at a San Francisco event on June 18, 2026, the company's first hardware product and its first venture outside generative imagery. The device is a liquid-immersion ultrasound tank with 8,960 transducers — a user stands on a platform that descends into water — and Holz pitched it as a full-body scan in roughly one minute with no radiation. The current prototype runs about 20 minutes per scan, has been used on roughly a dozen people, runs no AI in the imaging pipeline today, and has no FDA clearance. Holz said the goal is a fleet of 50,000 units and plans to debut the device commercially inside a "Midjourney Spa" in San Francisco at the end of 2027.
Strip the pitch back and the announcement is a hardware company in stealth dressed as an AI company in production. None of the impressive numbers — transducer count, target scan time, fleet size — are shipping today, and the imaging pipeline that would make this an AI story doesn't yet exist. What does exist is a generative-imagery firm with no medical-device experience announcing a Class II-equivalent diagnostic concept aimed at consumers without a clearance path. The FDA's De Novo and 510(k) pathways are not where image-generation engineering culture excels.
The pivot lands the same week OpenAI reported a 71% factuality drop on health prompts with a physician-led evaluation network, and Anthropic named LG, Samsung, and NAVER as Korean enterprise customers. Consumer AI brands are converging on health: OpenAI for triage and explanation, Midjourney for capture and imaging, Anthropic for enterprise workflow. Holz's bet is that a hardware moat is more defensible than another foundation model — a bet that requires Midjourney to learn medical device manufacturing, regulatory affairs, and clinical evidence at a pace it has never had to operate at.
Takeaway for learners: when an AI company announces hardware, separate the spec sheet from the timeline. A one-minute scan with 8,960 transducers in 2027 is a different company than a 20-minute prototype in 2026. The most useful question to ask of any AI-into-physical-world pivot is which regulators it has to convince, and whether the team has ever convinced them before.