Midjourney Can Fund a Scanner — But Can It Fund a Clinical Business?

Midjourney’s move into medical hardware is easy to admire from a distance. A profitable, self-funded AI company wants to use its own cash flow to build a whole-body scanner, avoid VC pressure, and control its roadmap. That sounds like the kind of autonomy founders say they want.

But clinical deep-tech is not image generation. It lives under different economics, different regulation, and different failure modes. The question is not whether Midjourney can afford to start. It is whether a software-margin business can carry the full cost of proving, clearing, supporting, and defending a medical device long enough for it to become real.

What Midjourney is actually building

The company’s medical push, described in 1 Minute Signal coverage of Julia McCoy and other creators, is a whole-body scanner backed by Midjourney’s own revenue rather than outside capital. The setup is unusual: a 25,000-square-foot luxury spa in San Francisco is slated to open in 2027, and the scanner technology is licensed from Butterfly Network. ^{1, 2}

Midjourney’s pitch is ambitious. It has talked about a scanner that could eventually be 10x cheaper and 60x faster than MRI, while current prototypes reportedly take about 20 minutes, have only been tested on roughly a dozen people, and do not have diagnostic approval. The company says it is starting with body composition rather than clinical diagnosis, which is a sensible regulatory narrowing — but also a reminder that the strongest claims are still ahead of the evidence. ^{1, 3, 4}

"The gap between the promised 60-second scan and the current 20-minute, non-diagnostic prototype is immense."

— Fireship, via 1 Minute Signal coverage ³

That gap matters because in medical hardware, “working” is not the same as “approved,” and “approved” is not the same as “adopted.”

Capital autonomy solves one problem, then exposes three others

Midjourney’s independence is real. David Holz has framed the company as a “self-funded research lab,” one that can tolerate losses without outside investors forcing short-term optimization. ⁵ That can be an advantage in frontier research. It lets founders ignore the usual venture clock and keep iterating when the product category is still undefined.

But autonomy changes the burden, it doesn’t remove it. In clinical deep-tech, the money you do not raise has to be replaced by something else: product margin, time, or patience. And the hardware case is especially unforgiving because the company is not just building a model; it is building a device, a regulatory path, a clinical validation story, and eventually a support and reimbursement machine.

That is the first hidden cost: revenue-funded hardware still has to pay for a “regulated wrapper.” Innolitics’ MedTech analysis is blunt that the wrapper around software — strategy, validation, quality systems, cybersecurity — takes more than half of a capital-efficient round. It also notes that clearance is not adoption, because payer evidence often forces a second round before commercial scale. ⁶

"The regulated wrapper around the software (regulatory strategy, validation, the quality system, cybersecurity) takes more than half the round."

— Innolitics ⁶

Midjourney is trying to avoid VC dilution, but it cannot avoid those costs. The FDA’s own materials make clear why: AI-enabled devices are expected to change over their lifecycle, while the traditional device regime was built around systems that remain largely unchanged. ^{4, 7} That mismatch is manageable for well-funded device companies. It is much harder if the project is supposed to be financed out of a software business.

The regulatory path is the real burn

This is where the Midjourney story leaves the comfort zone of “we have revenue, so we can self-fund.” Clinical devices have a pre-revenue period that is structurally longer and more expensive than software. Zechmeister Solutions argues that MedTech startups need roughly 3x to 10x more capital than comparable SaaS companies because there is an 18 to 24 month regulatory path between prototype and sellable product, with real costs and zero revenue in the middle. ⁸

That claim fits the broader FDA picture. The agency has added Predetermined Change Control Plans, more explicit cybersecurity guidance, and a risk-based framework that still expects a lot of documentation before a product can move. ^{4, 7} For AI-enabled devices, especially those drifting toward diagnostic claims, the compliance burden is not a side quest. It is the product.

Midjourney’s current choice to start with body composition, not diagnosis, is strategically smart because it sidesteps some of the hardest clearance questions. But that also means the company is not yet proving the most valuable version of its thesis. It is proving an easier, narrower one first.

The problem is that easier versions do not always finance harder ones. A business model built on spa visits and consumer wellness may help distribution, but it does not erase the cost of clinical validation if the company later wants to claim early disease detection. The more it leans into “medical” rather than “wellness,” the more its economics begin to look like MedTech rather than a subscription software company.

Wellness distribution is not the same as clinical adoption

The spa is clever because it lowers friction. People already expect to spend money there, so the scanner can be framed as a premium experience rather than a hospital instrument. But that distribution choice should not be confused with clinical adoption.

Technology Org’s coverage of the scanner warns that routine whole-body scans of healthy people often surface ambiguous findings that trigger more tests, more cost, and more anxiety, while most of those findings never need treatment. ² That is where the commercial model starts to separate from the clinical one. A spa customer may buy a scan because it feels like prevention. A health system or payer asks a different question: does the scan reduce downstream cost, change care, and fit a reimbursement pathway?

That is the part consumer enthusiasm cannot solve.

And it is why liability matters so much. If a product is marketed as wellness, the company has more room to experiment. If it drifts toward diagnosis or screening, it inherits the burden of clinical follow-up, false positives, and the cost of proving that a detected abnormality was worth detecting in the first place. Midjourney’s early body-composition framing buys time. It does not settle the harder question of who pays when the device becomes part of actual care.

The physics problem is real, but it is not the whole problem

Some skepticism about Midjourney’s scanner is about physics, not finance. 1 Minute Signal coverage of Midjourney’s medical imaging move highlights criticism from Hank Green and others that ultrasound physics imposes real limits, especially around brain and lung imaging where bone and air make the full-body promise much harder to sustain. ^{3, 9}

That distinction matters. Midjourney’s ambition may be to reconstruct around those limits with software, but the company’s current diagnostic clearance does not let it market that ambition as a solved clinical reality. The present device is still a prototype, still non-diagnostic, and still far from proving that software can overcome the physical limitations involved in imaging through bone and air. ^{1, 2, 3}

Why revenue-funded hardware looks elegant until utilization arrives

There is a version of the argument for self-funding hardware that sounds compelling: if the company is already profitable, why raise capital? Why not buy or build the next layer of the stack and keep the upside?

That logic is strongest when workloads are predictable and capital can be amortized quickly. Presenc AI’s 2026 pricing snapshot shows that on-prem H100 economics only work well for sustained workloads, with utilization and infrastructure overhead doing most of the work in the model. ¹⁰ ValueAdd VC makes the same point more directly: founders often focus on the apparent per-token advantage of self-hosting while ignoring the engineers, uptime, and utilization required to make the economics real. ¹¹

Clinical hardware is even less forgiving than AI inference. It is not enough to own the machine. You have to validate it, maintain it, support it, insure it, and keep it aligned with changing regulation and clinical practice. That is why Goldman Sachs’ infrastructure note matters here: the useful life of silicon is one of the biggest variables in total AI infrastructure spend, because hardware depreciation and replacement cycles keep reappearing. ¹²

In other words, revenue-funded hardware is not a shortcut around capital intensity. It is a way of internalizing capital intensity.

What the Midjourney case gets right

To be fair, Midjourney is not behaving like a typical hardware startup. Its position is stronger than most because it already has revenue, brand, and a large user base. Contrary Research estimates the company at $200 million in revenue in 2023 and says it reached profitability early, while other coverage pegs it closer to $500 million ARR by May 2025. ⁵ That is a meaningful war chest for a small team.

Its lean structure is real leverage. A company with 40 employees and a profitable subscription engine has more freedom to choose a hard frontier than a traditional startup burning VC capital. ^{5, 13} That matters because some clinical deep-tech ideas do require patient iteration and long horizons that public-market or venture timelines can distort.

But the distinction is between freedom to explore and freedom to scale. Exploration is plausible. Scaling is where the math gets ugly.

McKinsey’s data center forecast puts AI infrastructure needs at trillions of dollars by 2030, while Goldman Sachs argues that cumulative AI CapEx through 2031 depends heavily on chip replacement cycles and build-out bottlenecks. ^{12, 14} Those figures are for the broader AI build-out, not Midjourney specifically. Still, they make the structural point: physical infrastructure is expensive, slow, and recurring. A profitable software company can subsidize a pilot. It usually cannot casually absorb a full medical-device lifecycle at scale.

The real test is not whether it can launch

The test is whether Midjourney can move from prototype to approved, clinically credible product without letting the revenue engine starve the hardware roadmap or the hardware roadmap distort the core business.

That is why the current framing feels unstable. The company is trying to do three things at once:

1. Fund hardware from software profits.
2. Avoid outside capital pressure.
3. Enter a regulated market where development, validation, and adoption all take longer than software founders want to believe.

Those goals are not impossible together, but they are in tension.

If Midjourney stays in wellness-adjacent body composition tools, spa distribution, and consumer scanning, self-funding may be enough to get a product out the door. If it wants to become a clinically relevant screening platform, it will face the same reality that most MedTech founders do: the expensive part is not getting the first machine working. It is paying for the long tail of evidence, regulation, support, and trust.

That is the hidden cost of capital autonomy. It buys strategic independence up front, then hands the company the entire burden of proving that independence can survive contact with medicine.

What builders and investors should watch

For founders, the key question is whether the product is actually a software-margin business with some hardware attached, or a hardware business wearing software branding.

For investors, the useful diligence questions are more basic than the hype cycle suggests:

• What exact clinical claim is the product making today?
• How much of the roadmap depends on diagnostic use versus wellness use?
• What is the validation budget after prototype success?
• What happens when reimbursement, cybersecurity, or incidental findings become the dominant cost center?
• If the product works, who pays to deploy it at scale?

Midjourney may still prove that a profitable AI company can bootstrap a medical device category. But the burden of proof is now on the company, not the skeptics. The physics, regulation, and economics are all conspiring to make sure of that.