One surprising thing about life sciences investing is how vividly it confronts you with the many ways the human body can fail. In practical terms, that means many of the companies you see are developing treatments for grave diseases that are far from household names.

The other surprising thing is how tiny the teams can be.

The small markets are not the mystery. Neither are the small teams. A drug for thirty thousand people can still command a price that makes the math work, through orphan-drug incentives, pricing power, and the rest of the apparatus behind it. And a five-person team can move a program forward by contracting with organizations running the same sequence of milestones for dozens of others at once. Humanity has, in effect, industrialized its response to some of those thousand doors. The machinery exists; you just have to run it. The real puzzle is how drug development became one of the few places where a clinically valuable asset can be advanced without building a full company around it.

Now compare that with companies building devices, diagnostics, decision-support software, or research equipment. The underlying situation is usually similar. There is a promising invention, a credible clinical case, and a real patient need. Some supporting pieces exist, including contract manufacturers, regulatory consultants, and the occasional specialized CRO. But the vendor system is thin and fragmented, organized around too many different development routes to generate the concentrated demand that makes shared infrastructure self-sustaining. Downstream, the pull is weaker too. There are fewer large acquirers with business-development functions built to value and absorb assets in development, and fewer investors who know how to price a milestone they have seen a hundred times before. Each team is advancing a program by hand that looks as if it should work, through infrastructure that almost exists, toward a market not quite ready to receive it. “Come back when you have FDA clearance,” say investors. “Come back when you have revenue,” acquirers add. It can be a grinding, undercapitalized march toward a finish line that keeps moving.

Assume drug developers are not smarter or more disciplined. What created the conditions for a different organizational form to emerge, and how did that arrangement become so self-reinforcing that it came to seem natural?

Nobody planned it. The FDA organized drug review a certain way, for its own reasons, and the result was that suddenly everyone needed the same kinds of work done at the same stages (ie, IND, Phase 1, Phase 2, Phase 3, NDA) across hundreds of simultaneous programs. Predictable demand at predictable stages is what makes shared infrastructure economically viable, and the FDA created it without intending to.

Contract research organizations grew up to run preclinical studies and clinical trials because enough drug programs needed those services at those stages. Contract development and manufacturing organizations grew up to handle process development and scale-up for the same reason. Regulatory consultants who had shepherded dozens of IND applications became available by the hour. The fixed costs of world-class science (eg, equipment, expertise, facilities) could be spread across enough clients that each paid only for the work actually done.

That infrastructure wasn’t designed to enable lean drug companies. It emerged because the demand was there, and then made them possible as a side effect. The virtual pharma model — in its purest form, a company with no employees that converts capital directly into development work — did not precede the infrastructure. Nobody sat down in 1985 and decided to build a system that would let a tiny company advance a billion-dollar program. That approach only followed once the supporting infrastructure existed.

Ryan Avent, speaking about his time as a senior editor at The Economist, described his firm as not so much a business but as “a way of doing things consisting of an enormous set of processes. You run that programme, and you get a weekly magazine at the end of it.”[1] Drug development works much the same way. Run the program, and, if the bets are right, you get a drug. The company is almost incidental. What matters is whether the process is managed well enough to carry the asset forward.

A virtual pharma company is a handful of people and a lot of vendor contracts. That can look like a cost-cutting measure. It is really an answer to a more specific organizational question about how little structure an asset actually requires. In this case, the asset is a drug program — defined by its IP, development data, and regulatory route — with value independent of the organization built around it. It can become valuable long before the company does. It can be sold at almost any stage (eg, post-preclinical, post-Phase 1, post-Phase 2) as it accumulates evidence and value. Those exits do not require a complete company, just a credible asset and a minimal coordination layer.

The work itself is narrower than you might think. It sets direction, especially around indication and endpoints. It manages milestones and makes go or no-go decisions as results arrive. It selects and manages vendors. It talks to potential acquirers early, sometimes very early, to understand what an eventual buyer will need to see. And it allocates capital with enough discipline to reach the next decision point without building organizational overhead that will outlive its usefulness. What it does not do is equally important: employ bench scientists, operate manufacturing facilities, own equipment, or run clinical trials in-house. A Phase 2 biotech might have a handful of people.

This works in practice because the people who run these programs share a professional culture as much as a skill set. There is a cadre of experienced drug developers — scientists and operators who have worked across multiple programs, in startups and in industry — who can assemble around a new asset and be productive almost immediately. Everyone knows their role and expects their colleagues to know theirs; they speak an occasionally arcane lingua franca. The playbook is embodied in the people, not just the infrastructure.[3]

At the other end, large pharmaceutical companies maintain business-development functions whose job is to value and absorb mid-development assets. They can look at a Phase 2 program, understand what it is worth, and move quickly. That acquisition readiness is part of what makes the early-exit model work. In medtech, the function is weaker and less systematic, which is one reason devices don’t have the equivalent early-exit window.

Virtual pharma shows that advancing a clinically valuable asset does not require building a full company in the traditional sense. What it requires is the right coordination layer and the shared infrastructure to support it. Healthtech has neither, in the concentrated, self-sustaining form that model depends on, and not by accident.

Medical devices don't move through a single development route. There is no device equivalent of Phase 1, 2, and 3. A PMA, reserved for novel high-risk devices, might seem closest to what drugs go through, but the FDA doesn’t impose a standardized sequence. Two PMA programs can look completely different from each other. Specialists exist (eg, device CROs, regulatory consultants, contract manufacturers) but they are fewer and less specialized than their drug development counterparts, organized around too many different configurations to generate the same concentrated demand. The consequence is that device companies have to internalize more of the work that drug developers can contract out. The company grows to fill the gap the infrastructure can't supply. In that sense, every device startup is partly a workaround.

Reimbursement makes the problem worse in a specific way. For drugs, the route from approval to patient access is frustrating but at least roughly linear. For devices, the commercial path is often a separate maze: bundled payments, facility fees, professional fees, distinct coding for the device and the procedure, Medicare coverage that varies by administrative contractor, payer-by-payer contracting with commercial insurers. A 2023 study found that among novel devices and diagnostics requiring a new Medicare reimbursement route, the median time to achieve at least nominal coverage was 5.7 years after FDA authorization, longer, on average, than it took to get FDA clearance in the first place. More than half were still waiting when the study was published. That delay keeps asset value low until very late in development, which closes the early-exit window that virtual pharma depends on.[4]

But the stranded innovations in university tech transfer offices are not just devices. Some are diagnostics, which may have access to contract labs but not to anything like the combined clinical, regulatory, and reimbursement support that drug developers can buy off the shelf. Some are lab equipment and research tools, which may be manufacturable on contract but still have to build quality systems, field service, and customer support for themselves. Some are research software and algorithms, for which there is barely a specialist ecosystem at all. Others are lower-risk products — training aids, monitoring tools, consumer health applications — that face no major regulatory gauntlet but still have to figure out commercialization on their own. The details differ, but the pattern is the same. There is too little volume in any one category to support shared infrastructure, and no organization willing to take responsibility across categories. So each new program ends up building its own solution, not because the people involved are unsophisticated, but because no one has made it their job to assemble what already exists.

Novonate began inside Stanford’s Biodesign Innovation program, where a team spent years making prototypes and gathering clinical feedback before spinning out. They had a specific and well-documented problem in sight. Umbilical catheters in neonatal intensive care units were secured with tape and adhesive, a method associated with high infection rates and “catheter migration” in one-third to one-half of cases. The solution was LifeBubble, a dome-like device that gave nurses a consistent, reliable way to secure the catheter. It went through dozens of design iterations before it was ready for market.[5]

NICUs are not a venture-scale market. A few hundred high-acuity units in the United States, a patient population most investors dismiss. But the concentrated customer base that looked like a liability turned out to be a feature; the entire market was reachable with a small commercial effort. And the team had asset discipline, letting what the product needed determine everything else, without accumulating organizational overhead the asset didn’t require. Small raises, surrogate endpoints instead of expensive pivotal trials, a path to commercial viability that stayed visible throughout. CEO Eric Chehab recalls: “We felt like we might have a shot to make the economics work.” Novonate was acquired by Laborie Medical Technologies in February 2023.

Not a unicorn. As James Wall, a Stanford pediatric surgeon who advised the company, put it: “No one’s retiring off this technology or off this company.” But it was a successful outcome that returned value to investors and put a product in the hands of nurses who needed it. What Novonate doesn’t tell us is what to do when the asset can’t reach a natural acquirer and the early-exit window never opens.

That was exactly the problem Duke Rohlen set out to solve. After a career of building and selling medtech companies at unusual speed, he concluded that the early-exit model that therapeutics relies on simply does not work in medtech. Rohlen explains:

“Single product companies are highly risky...Single product companies are inefficient...Single product companies are beholden to a buyer universe that has all the power.”

So he went in the other direction entirely.

Maverix Medical, formed with backing from KKR and Hologic, is building a portfolio of diagnostics and devices across the lung-cancer care pathway, from early detection through biopsy and interventional technology. It acquires assets where they exist and develops them where they don’t, using Hologic’s diagnostic capabilities and commercial infrastructure and KKR’s capital as supporting resources. The exit is not the standalone sale of any one product but the clinical and commercial value of an integrated lung-cancer platform.[6]

This is not virtual pharma. It is a different organizational answer to the same structural problem of how to advance assets when the environment is too harsh for them to survive on their own. Rohlen’s answer was to expand the unit of organization from the product to the clinical domain. Asset discipline — the same logic Novonate applied to a single product — now operates across a portfolio, with infrastructure, relationships, and institutional knowledge accumulating across programs rather than being rebuilt for each one. Rohlen had the track record and the capital relationships to organize at that scale. Not everyone does. Maverix demonstrates that the coordination logic is portable, even if the organizational form has to change entirely.

The Cystic Fibrosis Foundation didn’t have a commercial rationale or a strategic backer. It had a disease community watching the median survival age sit at 32, and an impatient president, Robert Beall, who believed that “we have a responsibility for our patients’ destiny.”[7] Here is Beall on why the Foundation had to act:

“We had the gene. We had some targets. We understood the basic underlying defect... All these things could come together in a test tube, but it was taking academic scientists too long to do it. We had to accelerate the pace and bring industry in to that process, which meant we had to de-risk it. We had to take the early risk to draw them in.”

Beall again: “I approached several groups. Not many people returned my phone call.”

So the Foundation supplied patient capital, took on the coordination role the market had abandoned, and organized the effort around a defined clinical need rather than market size. The result was the first genuinely effective CF treatments and, eventually, a royalty stream that returned $3.3 billion to the Foundation when it monetized its Vertex-related royalties in 2014, funding further research.

The CF Foundation model is not virtual pharma either. CF affects roughly forty thousand Americans, which meant the usual commercial model was not available: the patient population was too small, the commercial interest too weak. The coordinating role that shared infrastructure usually plays in drug development was not going to emerge on its own. The Foundation stepped in, bringing together the clinical knowledge, development resources, and capital the market would not have assembled by itself. Capital, development strategy, and execution all mattered, but they were downstream of a more basic decision. Someone had to decide it was their job to make this happen, whether or not the market agreed it should.

Three teams, working in unrelated clinical domains with resources that aren’t comparable, each found a different way to supply what the standard model could not. That work can be lean when asset discipline is in place and the market is contained enough to reach. It can expand to the level of the broader clinical problem when no single product can bear the economics alone. It can be supplied by a stakeholder organization willing to act as coordinator of last resort.

Novonate worked because a particular team stayed disciplined and found a natural home in Laborie’s portfolio, a company organized around urology and gastroenterology workflows and built to acquire and commercialize exactly this kind of asset. Maverix works because Rohlen had the track record and the capital relationships to focus on an entire clinical domain. The CF Foundation worked because a disease community had both the motivation and the organizational capacity to take on the role the market had walked away from. None of these examples produced a replicable system. Each depended on people who saw the gap and had the means to do something about it.

Virtual pharma is an accidental demonstration of what happens when transaction costs fall far enough: you stop needing discrete companies and start needing coordination that the market won’t supply on its own. Healthtech keeps building companies around inventions not because companies are the right answer, but because no other credible model exists. Similar, improvised solutions keep appearing elsewhere, but only as one-offs. A repeatable alternative only becomes possible when you stop organizing around the company and start organizing around the output.

For the inventions this series has been describing — too small for venture, too useful to abandon, too numerous to keep treating as exceptions — that’s the problem. The pieces exist in healthtech. What’s missing is an organization whose job is to assemble them.

Endnotes

[1] Ryan Avent, quoted in Tim O’Reilly, WTF? What’s the Future and Why It’s Up to Us (2017). Avent was describing The Economist’s print operation. The drug development parallel holds well, with one difference: Avent’s concern was about how embedded processes resist change.

[2] The term “virtual pharma” dates to the biotechnology boom of the 1970s and 1980s. Genentech operated as a virtual company from its inception in 1976 until one of its founders established an in-house lab two years later. The model is older than it looks and more standard than it sounds.

[3] The medtech equivalent of this cadre exists, but lightly. There are experienced operators who have built and sold device and diagnostic companies and could in principle bring that knowledge to bear on a portfolio of programs. The problem is throughput. Drug development runs enough simultaneous programs to generate a continuous pipeline of people who have done it before. Medtech doesn’t, at least not at the same density. You can’t build a professional culture around a model that hasn’t been practiced enough to have practitioners. The virtual pharma model also creates an unusual entry point for faculty inventors, who can take advisory or part-time roles in programs built around their work without leaving their academic positions. In short, a meaningful stake in the outcome without a full career transition. That flexibility is structurally absent in most healthtech domains, where commercialization has historically required someone to go all-in.

[4] Sexton ZA, Perl JR, Saul HR, et al. “Time From Authorization by the US Food and Drug Administration to Medicare Coverage for Novel Technologies.” JAMA Health Forum. 2023;4(8):e232260. The study examined 64 novel technologies authorized through premarket approval and de novo pathways between 2016 and 2019 for which no Medicare reimbursement route already existed. FDA clearance isn’t the finish line, just the start of another, different race.

[5] Stanford’s Biodesign program describes the Novonate team working through dozens of design iterations before spinning out. Eric Chehab and James Wall quotes are from “Innovating for Niche Populations,” Medtech Talk podcast, July 29, 2025. Wall served as advisor and board member to Novonate and is currently at Intuitive Surgical.

[6] Maverix Medical was announced November 30, 2023; see “KKR, Hologic and Ajax Health Create New Platform to Accelerate Medical Device Innovation,” Business Wire, November 30, 2023. Cirrus Bio acquisition announced March 29, 2024; see PR Newswire, “Maverix Medical Closes Acquisition of Cirrus Bio as Foundation for Diagnostics Platform in Lung Cancer.” For Rohlen’s account of the underlying model, see transcript of “The Middle Path to Innovation,” Duke Rohlen keynote with Mano Iyer, LSI Europe ‘24.

[7] Robert Beall quotes are from Nature Medicine, “Straight Talk with… Robert Beall,” March 2012, interview by Elie Dolgin.

Imagine a grad student who discovers a better way to detect sepsis from a drop of blood. She discloses it to her university’s tech transfer office, who reviews her filing and calls her up: “Great work. Really promising. Let’s turn this into a company. Who’s going to be CEO?”

The grad student, thinking of loans and promises to parents, and the post-docs she’s been interviewing for, protests: “I’m less than a year away from finishing my PhD. I already have plans to work in a lab. Can’t we just license it to a diagnostic company?”

TTO again: “We tried. They want to see more data and another validation study. You need to de-risk this first. We’re going to need to get a company together and raise a seed round to fund that. We can help. Meet us next week and we can get the process rolling.”

Some version of this conversation happens at research universities across the country day after day. It’s also the moment when a lot of promising inventions stall.

No matter how good the invention, the only way it’s allowed to grow up is to become something it may or may not be: the founding technology of a scalable, venture-backable company.

This post, this series even, isn’t an argument against startups. Some of these inventions absolutely warrant them, and the ones that do deserve a system that serves them well. It is an argument about what happens to the ones that don’t, and about why the system can’t tell the difference.

You build a company when organizing things internally is cheaper than coordinating through markets. Hire the expert rather than contracting with one, own the equipment rather than renting it. That’s economist Ronald Coase’s famous answer to why companies exist at all. And for a long time it was a pretty good description of why university spinouts made sense too.

For many early-stage university innovations, that logic has largely stopped applying. Specialized capabilities have never been more accessible in the market. What once required building a company (research, manufacturing, regulatory and reimbursement expertise, software engineering) can now be suitably contracted out as needed. And in many cases, the costs to develop healthtech (and other) products have fallen dramatically.[1] The economic case for company formation is weaker than it’s ever been.

So, while nobody makes a conscious decision to push every invention toward a startup, the machinery surrounding inventions on university campuses is built squarely around company creation. Tech transfer offices, accelerators, incubators, pitch competitions, venture funds. Each has its own processes, metrics, and incentives, and every one of them points toward a startup as the natural outcome.[2]

On the ground, even university programs purpose-built to foster entrepreneurship awkwardly manifest the same patterns. At UW, the “primary organization tasked with driving entrepreneurship on campus” is called Discovery to Product, or D2P for short. (Sounds like exactly what this series is arguing we need.) But D2P’s own description of its goals is to mature projects “so that they become competitive for the funding needed to form a company.” Since 2014, it has helped launch or grow more than 140 startups. What it hasn’t built is a route for inventions that shouldn’t be startups. Recent university efforts to strengthen campus entrepreneurship have made the move explicit, excluding licensing and industrial partnerships from their mandate, effectively drawing a circle around company formation and calling it entrepreneurship. The underlying assumption, that every invention should become a company, goes unexamined.[3]

This is the best example I’ve found that Professor Coase’s logic has inverted. The startup infrastructure has today become so dominant that the cost of doing anything other than forming a company is higher than the cost of doing it the familiar way. You could say that what was supposed to solve a market coordination problem has itself become one.

If company formation is the default, licensing should at least be the obvious fallback. In practice, it rarely is. (At least for things that aren’t drugs.) And how that usually fails, which you can hear in the tech transfer dialogue, is exactly what pushes offices toward company creation in the first place.

So, why doesn’t licensing work? First, established companies aren’t interested in early-stage university research. WARF’s public affairs director Kevin Walters explains the incumbent’s perspective plainly: large medtech and device companies “don’t often want to take risks on basic research patents, what we call ‘deep tech.’ To take the massive risks, you’ve got to create a new company.”[3] Most strategics are hunting incremental improvements to products they already sell. An early-stage university invention asks them to absorb a kind of risk they’re not built for.

Second, most of what arrives at a tech transfer office isn’t ready to license anyway. Greg Keenan, who runs WARF’s Accelerator and Ventures programs, describes most disclosures as “early-stage technologies, sometimes just lab bench experiments, that aren’t products in any meaningful commercial sense.”[3] A licensing deal requires a licensee who can see a plausible route from here to something they can sell. That’s a hard case to make for something that hasn’t left the lab.

Since licensing rarely works for non-pharma innovations, doing nothing (ie, not starting a company) means the invention sits in a catalog nobody browses. That’s part of why the institutional default is so hard to dislodge.

Underneath all of this is a simpler problem: where does the money come from? An inventor has a compelling idea, and to develop it, she needs funding. But the only financing available at meaningful scale is venture capital, and it requires a compelling story about market size, because venture returns depend on a small number of large outcomes. So the innovation needs not just a company, but a company with an addressable market big enough to support that return profile. Otherwise, the capital simply isn’t there, regardless of the clinical or commercial merits of the invention.

The human cost is most visible in pediatric medicine. “A lot of companies avoid making things for kids,” one ICU physician told me, “just because it isn’t worth their time.” After all, that’s a market with smaller patient populations, lower reimbursement, and higher regulatory complexity. As a result, clinicians routinely adapt adult devices for pediatric use, off-label and at some risk, because the right-sized product simply doesn’t exist.[5]

It isn’t only the smallest-market innovations that fall through, leaving real gaps in patient care. There’s a whole category of outcomes in the tens to hundreds of millions that are achievable, valuable, and still, eventually, orphaned. They’re objectively not failures; they’re the right size for what they are, even if venture needs them to be bigger and private equity wants them to be already profitable. The acquisition market that used to bridge that gap has mostly closed its doors. Large medtechs rarely acquire pre-commercial companies anymore, which means there’s no buyer waiting at the end of the runway for a mid-sized outcome either. A substantial portion of healthtech innovation quietly disappears between those positions.

The same logic distorts the route for innovations who have a clear path to profitability. Immuto Scientific, a WARF spinout from UW-Madison, developed a faster, cheaper method for determining protein structure. It had obvious value to pharmaceutical companies and a natural, straightforward path to profitability in a few years through recurring revenue from their per-protein analysis fees. “We knew right away that there was a service business here,” CEO Faraz Choudhury has said. “And we could make some money on services.” But the logic of the capital structure pointed them elsewhere; to justify venture investment, the company needed a different “North Star.” As Choudhury put it: “How do we make this into a high-growth business? We’ve got to make drugs of our own with this platform.” So the business pivoted toward making drugs and walked away from a profitable model to chase a venture-scale bet.[3]

Universities apply their own pressure. They behave as if they were licensing a de-risked drug through a well-understood regulatory pathway, when what they’re actually doing is asking someone to build a company from scratch to take an unproven product into an uncertain market. Calibrating their equity demands to the invention rather than to the full cost of building a company around it gets the pricing wrong.

Demands for substantial equity extract a disproportionate share of a venture that hasn’t been built yet. Past a certain point, the terms make the deal not worth doing for exactly the people you’d want to attract. There have been serious international efforts to reform and standardize academic licensing terms, but the conversation has stayed focused on making spinout deals faster and the equity curve fairer, rather than helping to define when a spinout is the right vehicle in the first place.[6] European universities on the wrong side of that curve restrict supply through onerous terms, then try to compensate by subsidizing demand through accelerators and innovation hubs, which is a somewhat expensive way to avoid asking whether the terms were right in the first place.

And then there are the companies that form anyway, on the strength of early enthusiasm, friends-and-family capital, maybe an angel round, with big promises about markets that don’t quite pencil out. Eventually the company hits the institutional funding market and stalls. The market isn’t large enough, the team doesn’t have the right track record, the capital needs are too high for what’s on offer. Many of these companies don’t fail cleanly. They drift, cycling through accelerators and pitch competitions, collecting small grants, accumulating just enough forward motion to stay alive. The inventor stays attached long after the point of rational exit, spending time on fundraising instead of research, patient care, or finding a buyer with the resources to take it forward. There is no defined endpoint, no moment at which the system declares the bet lost and lets everyone go home.

Think about what that grad student actually needed. The tech transfer office pushed her toward a startup because that was the available mechanism. What would have really helped her is someone with commercial instincts and domain knowledge who could take the invention, do the development work, find a licensing partner or an industry buyer, and shepherd the product to market without building a standalone company around it. It’s easy to imagine that person and yet that’s the one role with no institutional home in the current system. Tom Chapman, co-founder of a medical device commercialization company, describes the situation directly: “The gap isn’t ideas. It’s pathways for business-oriented operators to step in early, explore viability, and lead.”[7] It’s not just that capable operators are scarce or that incentives are misaligned but that we have no defined positions for them to fill.

This is a problem of selection, too. When Norway transferred IP ownership from faculty to universities, startup formation among science and engineering academics fell dramatically.[8] The reform didn’t change who was still trying, but faculty with the most valuable research had the most options and they exercised them.

The inventors I’ve spoken with in preparing this series confirm the picture from the other side. One physician-researcher described what most of his colleagues would actually want as “something like a side hustle” — passive income from an invention without having to become a founder. The financial appeal, he said, is real: most faculty “are taking a pretty substantial pay cut” to stay in academic medicine, and the idea that an invention could generate some additional return is genuinely attractive. What makes a startup unrealistic isn't interest; it’s “bandwidth.” The skill acquisition required, he said, “is rather steep. Networking, fundraising, pitching” are things that, to pursue seriously, would require “a leave of absence or a big percent reduction” from a traditional faculty role. That's not a realistic ask for someone trying to keep a lab running and a chair happy.

Another was more direct. “I am not interested personally in starting a company,” he said. “I don’t have time to be honest with you.” He actively wants his inventions at the bedside — “that’s the reason,” he said, “why I am looking into a solution like this.” But the route through company formation doesn’t fit his career or his life, and there is no other path on offer in which he can move the invention forward without a high-stakes career detour. Forcing him to retrain as a startup CEO is a rookie misallocation.

Every spinout builds from scratch the operational capability to function as an independent company. Sure, the specific questions differ: regulatory strategy for a diagnostic differs from regulatory strategy for a surgical device. But each founder-CEO figures out how to manage a cap table, negotiate a manufacturing contract, engage a regulatory consultant, build a board, while also trying to develop a product. The institutional knowledge accumulated by one spinout doesn’t flow to the next one down the hall. The overhead serves one program and disappears.

Economists have a name for what’s missing here: economies of scope. Economies of scale says you get more efficient as you do more of the same thing. Economies of scope says it’s cheaper to produce two different things through the same platform than to produce each one independently. The savings come not from volume but from variety. Shared infrastructure serving different programs simultaneously costs less than building separate infrastructure for each. This is, I should say, a pretty basic idea.

Procter & Gamble doesn’t build a separate company for every product in its portfolio. Tide, Crest, Gillette, and Pampers share legal, regulatory, manufacturing, and distribution infrastructure; the marginal cost of adding a new product to that platform is a fraction of what it would cost to stand it up independently. But the current startup-per-invention model does exactly the opposite. Every new spinout rebuilds operational capability from scratch, and when the program ends, what was learned evaporates with it.

But universities and accelerators are rewarded for creating new companies, not for efficiently developing realized inventions. Startup numbers feature in rankings, press releases, grant applications, and annual reports. And the incentive to keep doing what’s countable is more powerful than the incentive to explore alternative ways to get gadgets to market.

Universities also know how to get paid from the current model. Decades of deal-making have produced templates for how equity stakes and royalty streams flow back to the institution from a spinout or a license. The legal infrastructure exists, the precedents are clear, the finance office knows where to put the income. A product-not-company model has none of that. No established deal structure, no standard royalty framework, no clear accounting treatment for returns from a product commercialized through another mechanism. For a tech transfer office trying to justify its budget, the known deal is always easier than inventing a new one.

The venture studio, a professional company-creation platform that uses shared legal, regulatory, and operational infrastructure to systematically create startups, is the most direct response to the newco-for-every-invention overhead problem. A growing number of universities have attached these directly to their tech transfer efforts. That means having operators on staff so the inventor doesn’t have to become a CEO, and focusing on active development instead of passive licensing. Venture studio advocates report better financial returns and faster paths to subsequent funding rounds than traditional approaches.[9]

It solves the overhead problem completely, for companies. But each of those still needs to build a team, raise venture capital, and clear the market size hurdle. It’s a more efficient expression of the institutional default, not an alternative to it. The innovations that were never right for the company form don't get rescued by a more efficient company-creation machine. They get processed more quickly to the same outcome.

Anastasia Gamick, of Convergent Research, gets closer to the right diagnosis. There’s a class of companies that should get built but don’t, she argues, sitting in a valley between capital categories. The problems they address are “well-characterized, urgent, and solvable” — she specifically names diagnostics, antibiotics, and others with roots in healthtech — but the companies they require are “Too slow for venture. Too early for PE or growth equity. Too capital-intensive for bootstrapping. Too profitable for charity.” Her answer is more patient capital and better-structured funds. That will help. But it’s still a company-creation answer when the problem runs upstream of capital structure. This series is asking whether companies are the right unit in the first place.[10]

The system is very good at producing companies. What it cannot do is develop a product without building the typical startup around it. Post 3 is about different organizational logics that already exist in corners of the life sciences and elsewhere, which turn out to be much better at getting inventions to patients.

Endnotes

1. Getting a medical device through clinical validation and early commercialization remains expensive and is getting more so. Development costs are rising, particularly at the clinical validation stage. Pivotal trials for novel devices now routinely run into the tens of millions of dollars, IDE study requirements have grown more demanding, and post-market surveillance obligations have expanded significantly under updated regulatory frameworks on both sides of the Atlantic. Meanwhile, the commercialization gauntlet has lengthened. A 2023 study found that among 64 novel devices and diagnostics that required establishment of new Medicare coverage after FDA authorization, the median time to achieve at least nominal coverage was 5.7 years. That figure applies only to the 44 percent that achieved coverage at all. The majority were still waiting at the time of analysis, meaning the true average is likely considerably longer. (See Sexton ZA et al., "Time From Authorization by the US Food and Drug Administration to Medicare Coverage for Novel Technologies," JAMA Health Forum 4, no. 8 (2023): e232260. doi:10.1001/jamahealthforum.2023.2260). FDA clearance, long a viable exit milestone for acquirers, no longer reliably triggers reimbursement, leaving companies to navigate the full coding, coverage, and payment process on their own. This is one of the structural reasons the drug development analogy, while instructive, doesn’t port directly to healthtech. Post 3 is specific about where it holds and where it breaks down.

2. SBIR and STTR programs, the federal grants marketed as non-dilutive alternatives to venture capital, sound, on the surface, like an off-ramp from the company form, or at least a small-business alternative to the venture model. In practice they require a legal entity to receive the grant, reward the same milestones venture does, and run through the same tech transfer machinery. They reinforce company formation rather than offering a way around.

3. “Empowering the Wisconsin Idea: The Future of Entrepreneurship at the University of Wisconsin–Madison.” Report of the ad hoc working group commissioned by Chancellor Jennifer Mnookin, September 2024. The report is notable for what it excludes as much as what it recommends: the working group explicitly scoped its charge to entrepreneurship resulting in company formation, excluding industrial partnerships and licensing to established companies. That framing is the institutional default made visible.

4. John Allen, “WARF Seeks the UW’s Next Big Thing,” On Wisconsin, Fall 2025. The Walters and Choudhury quotes are drawn from this article. The Keenan quote is also sourced here.

5. Pathak et al., “High-risk Therapeutic Devices Approved by the US Food and Drug Administration for Use in Children and Adolescents From 2016 to 2021,” JAMA Pediatrics 177, no. 1 (2022). Of 124 high-risk therapeutic devices approved by the FDA during the study period, 80 percent were approved for adults only. Estimates of off-label device use in pediatric patients range from 60 to 75 percent. See also Sutherell et al., “Pediatric interventional cardiology in the United States is dependent on the off-label use of medical devices,” Congenital Heart Disease 5, no. 1 (2010): 2–7, cited in Haller et al., “Experience With Pediatric Medical Device Development,” Frontiers in Pediatrics (2020).

6. The USIT Guide was developed by TenU, an international collaboration of leading technology transfer offices including MIT, Stanford, Oxford, Cambridge, Imperial College, and others. Its focus on deal structure, equity terms, royalties, and licensing mechanics reflects the consensus view that the problem with university spinouts is how deals are constructed, not whether spinouts are the right vehicle. That’s meaningful reform, but not the reform this series is describing.

7. Tom Chapman’s LinkedIn post

8. Hvide, Hans K., and Benjamin F. Jones. “University Innovation and the Professor’s Privilege.” American Economic Review 108, no. 7 (2018): 1860–1898. The pattern is worth sitting with: the reform designed to increase commercialization activity reduced it most sharply among exactly the faculty most likely to produce commercially valuable work.

9. Matthew Burris, “The University Venture Studio: Unlocking the Innovation Potential of Higher Education,” Venture Studio Perspective (Substack), August 5, 2025, citing Global Startup Studio Network / Morrow & Co, “Disrupting the Venture Landscape.” The data comes from industry sources with an interest in the conclusion and has not been independently validated. The directional claim, that studios outperform traditional approaches, is widely accepted among practitioners even if the specific figures are contested.

10. Gamick’s piece is worth reading alongside this series. Post 4 returns to this distinction when examining what an alternative infrastructure might actually look like and how it might be capitalized.

    Anastasia Gamick

    Companies that should exist

    There is a class of companies that should exist but don’t because the economics don’t fit neatly into any existing capital structure…

    Read more

    a month ago · 136 likes · 22 comments · Anastasia Gamick

Here is something that happens every day at research universities. A scientist, or an engineer, or a physician has an idea for a product. Could be modest: a small improvement to a piece of lab equipment, an update to a diagnostic algorithm. Could be significant: a device that makes a common surgery meaningfully safer, a test that catches a disease earlier, a piece of software that gives a clinician needed information at the right moment. The ideas range from mundane to transformative; most of them share one characteristic: they never go anywhere.

Not because they don't work. Not because there are no customers. Because they're misfits in an ecosystem designed for a very specific kind of product.

I’ve spent much of my career watching valuable innovations get quietly stranded in this way. This is the thing I want to explain. And, eventually, do something about.

The Three Inventions

Let me give you some examples, drawn from patterns I've seen repeat across universities and clinical settings. The details are composites, but the problem is not.

A biomedical engineer builds better equipment for tissue culture. Research labs will pay $50,000 per unit. You could sell five hundred units a year at a healthy profit. The product works, the customers exist, the economics are fine, and if you described this situation to a normal person they would say “Ok, so build the product and sell it to the labs.” But there’s no recurring revenue. No platform potential. No network effects. No odds to achieve the kind of returns that attract venture capital. So it sits.

A computer scientist develops clinical decision support software for a specific surgical procedure. Hospitals need it. It demonstrably improves outcomes. It could save millions of dollars and, more importantly, meaningfully improve patient care. But it isn’t an AI platform with applications across fifty procedures. The sales cycle into health systems is long and grinding. This is a good product for a real market, and a good product for a real market is, we’ll come back to this, not actually what the available funding mechanisms are designed to support.

A physician-scientist creates a better diagnostic test for a condition affecting a hundred thousand people a year. The existing commercial lab networks could pick it up and sell it profitably tomorrow. What they need is someone to develop it far enough that it’s a real commercial asset. But the venture investors who might fund that development want a platform for dozens of tests, not one really good one. (The one really good one is, financially speaking, not very interesting to them. More on why in a moment.)

Notice what these three examples have in common. They aren’t ideas that failed to find customers. They’re working ones that can’t find a route forward. They’re orphaned innovations. The product works. The market exists. Someone would pay for it. And yet.

How Venture Capital Works (And Why That’s A Problem)

To understand why this happens, you have to understand what venture capital is actually for.

Venture capital is not a general-purpose mechanism for funding good ideas with real customers. It’s a specific financial instrument with specific mathematical requirements. A venture fund raises money from limited partners (eg, endowments, pension funds, family offices) with the promise of outsized returns. “Outsized” means something specific: five to ten times the invested capital, in seven to ten years, across a portfolio where most investments fail. The structure demands that the winners be very large, because the losers are total losses, and the math only works if enough winners are large enough to return the whole fund.

Mike Partsch, who runs WARF Ventures, the venture fund of the University of Wisconsin’s tech transfer office, explains this directly. “If an entrepreneur said ‘If you invest in us and we’re wildly successful, you could get a 25 to 30 percent return,’ we’d be like, ‘No, not interested.’” He’s not being unreasonable. He’s describing the actual constraints of his fund’s structure. A 25 percent return is an extraordinary result in most investment contexts. For a venture fund, it’s disappointing. The math requires something different.

So when we say a company’s market is “too small for venture capital,” we don’t mean the market is small in any ordinary sense. A business that generates $30 million in annual revenue at healthy margins, helping a hundred thousand patients, is not a small business. It’s a good one. It’s just the wrong shape for the only growth-capital mechanism that touches early-stage health technology. The problem isn’t the size of the opportunity. It’s that the opportunity is the wrong size for the only financing model available.

(The alternative, licensing the technology to an existing company, works well in exactly one industry: pharmaceuticals, where large global businesses have dedicated teams actively scouting for new assets. In most of healthtech and medtech, the licensing market is thin, the buyers are fragmented, and the transition from university IP to commercial product is murky enough that it rarely happens in practice.)

How Universities Think About This

Here’s where it gets interesting. Universities know this is a problem. Or at least some of them do.

Tech transfer offices, and the accelerators and entrepreneurship programs that have grown up around them over the past two decades, are almost universally organized around a single output: new companies. This makes a certain kind of institutional sense. Startups are legible and easy to count. You can report them to your board, feature them in your alumni magazine, compare your number to peer institutions. When the University of Wisconsin’s working group on entrepreneurship released its 2024 report, Empowering the Wisconsin Idea, it noted that UW ranks eighth in the country in research expenditures but thirty-first in startup formation. The gap is framed as the problem to solve.

A better startup pipeline would be valuable. UW has since appointed Lewis Sheats as inaugural director of a new Wisconsin Entrepreneurship Hub, and the language around it is deliberately ambitious.

But here’s the thing: the same report explicitly sets aside other commercialization pathways (eg, licensing to existing companies, product development outside the startup model) as beyond its mandate. The most self-aware reform effort currently underway at one of the country’s leading research universities has defined success as producing more startups. The solution to the problem of innovations not reaching patients is, per the working group, more and better startups.

This would be fine, or at least less problematic, if most faculty innovations were good startup ideas. Many are not. Starting a company requires a particular kind of founder commitment. You are, typically, leaving or substantially reducing an academic career. And it requires a particular kind of commercial trajectory: A good product that could profitably serve a modest market is not, in most cases, a good startup idea. It doesn’t have the growth dynamics that justify bringing on investors, hiring aggressively, and building toward an exit. Pushing it through the startup model doesn’t make it a better startup. It just adds a layer of expensive institutional overhead to what might have been a straightforwardly buildable product.

The Norway Problem

For a long time, watching innovations stall at exactly these points (eg, promising product, no path, wrong shape for the only available capital) I assumed the explanation was local. Bad luck. Weak management. An inventor who didn’t work hard enough. Then I came across a natural experiment that suggested something more structural was going on.

In 2003, Norway changed its IP laws to transfer ownership of university research from individual professors to their institutions, basically adopting the American approach. The theory was that centralizing ownership would professionalize commercialization and increase spinout activity. Give institutions control over the IP, build out the infrastructure, produce more startups.

What actually happened: university startup formation fell fifty percent. Among science and engineering researchers, startup formation fell sixty-three percent on a per-worker basis. Patenting rates fell by a similar margin, and patent quality declined alongside.

The system designed to encourage commercialization suppressed the commercialization instinct instead. Not because Norwegian professors stopped having ideas, but because the institutional process replaced a decentralized, faculty-driven impulse with a centralized administrative one that most faculty found alienating and most innovations couldn’t survive. And the institutional machinery that replaced it was built, as ours is, around a single output: new companies. It didn’t just produce fewer startups. It substituted an organizational reflex for the real question: whether a startup is the right vehicle for a given innovation in the first place.

You can read this as a specifically Norwegian story about institutional culture. Or you can read it as a canary: a visible version of a dynamic that operates more slowly and less visibly in the American model, where it’s a little harder to observe because the change happened gradually rather than all at once.

I tend toward the second.

What Gets Left Behind

The University of Wisconsin receives roughly three hundred fifty to four hundred invention disclosures per year. About one per day, as Greg Keenan, a senior leader of WARF, describes it. They patent roughly half. A small fraction become companies. The rest find another path, or none at all.

The true cost of this isn’t visible in any annual report. It lives in the distance between what could have reached patients and what actually has. Universities measure patents filed, startups formed, venture capital raised. They don’t measure how many innovations actually reached patients, how many profitable products never got built because the market was deemed too small, or how many faculty inventors gave up after the tech transfer-to-startup process proved too slow or too discouraging to navigate. The metric and the thing you actually care about, clinical impact, have decoupled almost entirely.

Ben Reinhardt, who runs Speculative Technologies and thinks carefully about where innovation gets stuck, calls this the missing middle: the space between what philanthropy will fund and what venture capital finds interesting. Most promising faculty innovations fall into the gap between them. Too commercial for philanthropy, which wants pure research with no financial entanglement. Not commercial enough for venture capital, which needs billion-dollar exits to work. Too early for strategic acquirers. Too dependent on makeshift licensing channels not built for this category of innovation.

It’s not a theoretical construct. It’s an actually crowded space of working innovations with real clinical value and no institutional home or way forward. The device that helps tens of thousands of patients, profitably, at modest scale. The diagnostic test that works and has customers and needs someone to build the business-shaped wrapper around it, just not a venture-backed company. The software that hospitals need and will pay for, just not at the growth rates that justify a Series A.

These aren’t failed ideas. They’re working ones that we’ve forgotten, or never learned, how to handle. Once the Bayh-Dole Act established the university-to-startup pipeline in 1980, we lost the habit of asking what an innovation actually needs, rather than what the system is designed to produce.

In the next three posts in this series, I take up why the system defaults to companies when it should sometimes just make products, why drug development figured out a model for this that healthtech hasn't adopted, and what it might actually take to create something new. Nobody has built that machinery yet. That is what this series is about.

Part 2 is now up. Read it here.

Citations

Greg Keenan quotes and WARF disclosure figures: John Allen, “WARF Seeks the UW’s Next Big Thing,” On Wisconsin, Fall 2025; and Brittney Kenaston, “Taking Ideas to Market,” In Business Madison, August 2025.

Mike Partsch quote: John Allen, “WARF Seeks the UW’s Next Big Thing,” On Wisconsin, Fall 2025.

WARF Ventures and Accelerator timeline: Brittney Kenaston, “Taking Ideas to Market,” In Business Madison, August 2025.

UW rankings and report: Empowering the Wisconsin Idea: The Future of Entrepreneurship at the University of Wisconsin–Madison, UW–Madison Working Group Report, 2024.

Lewis Sheats appointment: Rodee Schneider, “Lewis Sheats named inaugural leader of UW’s Wisconsin Entrepreneurship Hub,” UW–Madison News, January 20, 2026.

Norwegian IP reform data: Hvide, Hans K., and Benjamin F. Jones. “University Innovation and the Professor’s Privilege.” American Economic Review 108, no. 7 (2018): 1860–1898.