Daniel Ek: Life After Spotify, Broken Healthcare Incentives, Catching Disease Early & AI's Potential
After two decades running Spotify to well over 700 million active users and more than 300 million premium subscribers, Daniel Ek became executive chairman on January 1 and returned to building. Spotify began in 2006, launched in Sweden in late 2008 and reached the U.S. in 2011; Neko Health similarly spent five years developing before launching in Sweden in 2023, followed by the U.K. and now the U.S. “Maybe I just have a high pain tolerance.”
Neko’s wedge is a vertically integrated, $499 preventive-health visit with positive unit economics. In roughly an hour, it measures 53 blood markers, captures more than 6,000 high-resolution body images, evaluates cardiovascular and other established indicators, and delivers the results through a clinician. Ek calls it “the most valuable hour you can invest in your health.” Neko co-founder Hjalmar Nilsonne is the brainchild and handles most of the work; Ek is not its day-to-day CEO.
More than 100,000 scans have produced an early health signal: roughly 1% of members had a serious, previously undiagnosed condition discovered. Ek says members with the worst starting health status have improved the most, while explicitly stopping short of claiming conclusive population-level outcomes. Neko recommends annual visits, but he says it is too early to know what broad adoption would do to healthcare budgets.
Dermatology illustrates why Neko views multimodal, longitudinal data as the product rather than a one-off checkup. The average person at Neko has roughly 950 moles; Neko indexes them all, uses AI to flag risks, layers clinician and specialist review on top, then compares every lesion across visits. Even an exceptional doctor “can’t possibly remember” how hundreds of moles looked a year earlier.
Ek’s tentative diagnosis of America’s healthcare failure is an incentive-duration mismatch. Employer-linked insurance may retain a member for only two or three years, while preventive investments can require 10–20 years to pay back; the system therefore remains oriented toward acute symptoms. His response is pragmatic rather than sweeping: lower diagnostic costs and create enough outcome data to make long-horizon ROI legible.
The data strategy extends beyond Neko’s own sensors into wearables, research and continuously upgraded diagnostics. Neko already allows Apple Health imports, has completed four clinical trials, has two underway and another four planned, and expects Gen 2 to keep expanding rather than remain a static product for a decade. Ek’s view is that making 10x or 100x more healthcare data available could enable conclusions that today’s relatively small datasets cannot support.
On AI, Ek favors both open and closed models, declines to take a strong position on pacing, and raises compute as an additional risk variable beyond static model classifications. Spotify uses frontier and fine-tuned models because cost, efficiency and control differ by task; meanwhile, a system using 100,000 GPUs may be much more powerful than an open model running on a home PC. His larger complaint is that the industry undercommunicates AI’s “crazy positive stuff,” from personalized music to earlier disease detection.
1. Spotify won by underwriting the incumbent’s downside
The interview starts with Spotify at well over 700 million active users and more than 300 million premium subscribers after the two decades Ek ran it; on January 1, he shifted to executive chairman. Ek started Spotify in 2006 at age 23, before the iPhone and YouTube and while Facebook remained college-only. Amid rampant piracy and lawsuits against individual consumers, his conclusion was blunt: “There’s no way to put the genie back in the bottle.” The product therefore had to feel like “all the world’s music at your fingertips.”
Sweden’s distressed music market created the opening. Fast broadband made piracy effortless, iTunes was unavailable, and Ek recalls the local music industry losing roughly 80% of its revenue—making the country both one of the world’s worst music markets and an unusually receptive laboratory for a legal alternative.
Ek and co-founder Martin committed their prior-startup proceeds to guarantee labels enough revenue to protect the next year’s budget and bonuses. If Spotify failed, executives still got paid; if it worked, they shared the upside. After years of negotiation, Spotify launched in late 2008, expanded to the U.K., then entered the U.S. in 2011.
Stardoll supplied Ek’s operating rehearsal. The site’s average page took roughly four minutes to load; he rearchitected it, hired a new technical team and brought loading below one second, after which traffic “exploded.” He had agreed to help for 6–12 months partly hoping Danny Rimer would fund Spotify—it did not happen—then left to build it anyway.
2. Life after Spotify means building again, not merely investing
Ek had little investable wealth before Spotify’s public listing around 2018, having spent the prior 12–13 years focused almost exclusively on the company. Once he tried investing, the preference became obvious: “I loved building way more than I loved investing.” Advising founders without operating authority proved less satisfying than tackling a problem himself.
Healthcare had occupied him since at least a 2012 or 2013 interview. His starting puzzle was why spending kept rising while outcomes deteriorated. He found no single cause, but focused on chronic disease—including heart health and skin cancer—because early discovery, he says, can make these conditions preventable while keeping treatment cost and suffering low.
Neko co-founder Hjalmar Nilsonne is the company’s brainchild and does most of the work, while Ek remains involved without serving as its day-to-day CEO. Neko’s thesis is that prevention begins with “a lot more data”—deeper within each measurement type, broader across multiple modalities and collected longitudinally. Ek explicitly connects that flywheel to Spotify: more listening data improved the prediction of the next song; richer health histories might similarly improve the prediction of future disease.
3. Neko compresses preventive medicine into one instrumented hour
The $499 Gen 2 visit is vertically integrated across facilities, nurses, doctors, diagnostic equipment and software. Members provide blood for 53 markers, enter a camera rig that captures more than 6,000 high-resolution images, and undergo heart, blood-circulation, grip-strength and other evidence-backed measurements.
Results, including blood work, are available during the same visit for uninterrupted clinician review. The full experience lasts about an hour, though someone with few questions could finish in 30–40 minutes. Ek wants it to become as routine as an annual dentist appointment, while acknowledging that “$500 is still a lot of money.”
The U.S. rollout starts in New York at 300 Lafayette. The discussion also identified Miami and Washington, D.C., as planned locations, with broader U.S. expansion targeted over the coming 12–24 months.
The hosts pressed on whether Neko meaningfully differs from previous integrated-care concepts. Ek emphasized its operating history: eight years of company development, more than 100,000 scans and a recently published third-year data survey—not simply a newly announced thesis packaged around preventive care.
Roughly 1% of members have had a serious undiagnosed medical issue discovered, while many receive guidance around stress, diet, sleep and other lifestyle variables. Most encouraging to Ek, though still early, is that members starting in the worst health condition are improving most; he cited people quitting smoking after seeing and discussing their health status.
4. Longitudinal dermatology shows where AI actually earns its keep
The average person at Neko has about 950 moles, with some having many thousands—far beyond what a normal appointment could inspect consistently. Neko indexes every mole, lesion, rash and area of redness rather than depending on whichever spot a patient or doctor happens to notice.
An AI system flags potential risks, a clinician reviews them, and multiple specialist dermatologists examine anything still concerning. Ek’s model is deliberately hybrid: “It should seamlessly be both AI and amazing clinicians in a great packaged experience,” not an autonomous algorithm substituting for medical judgment.
The larger advantage emerges at the next visit. Because every lesion has been catalogued, the system can detect abnormal change from one year to another; even “the best doctors in the world” cannot reliably remember how 950 separate moles previously looked.
5. Prevention’s obstacle is misaligned payback, not a lack of stated intent
Asked why U.S. hospital visits and drugs can cost multiples of their apparent inputs, Ek declined the invitation to pose as an expert with three sweeping fixes. His narrower diagnosis was “show me the outcome, I’ll show you the incentive”: the system was constructed around a time when infectious disease was the main concern and still rewards intervention after unmistakable symptoms appear.
Employer-linked coverage makes the mismatch worse. A person may remain with an employer—and its insurer—for only two or three years, while prevention might repay its cost over 10, 15 or 20 years. Ek’s proposed wedge is to reduce a speculative million-dollar intervention to, hypothetically, tens of thousands, increasing the chance that someone funds it.
Neko says its $499 price already produces positive unit economics and that some existing clinics are profitable. Yet Ek would not extrapolate those economics into national savings: on what universal annual screening would do to healthcare budgets, “I think it’s too early to say.”
Better measurement could eventually clarify that ROI. Ek finds healthcare datasets surprisingly small by technology-industry standards and argues that making 10x or 100x more data available could yield useful conclusions. Neko has completed four clinical trials, has two underway and another four planned, while publishing aggregate findings annually.
6. Ek wants AI judged by applications, openness and actual compute
On pacing frontier AI, Ek offered no strong view one way or the other. The host—not Ek—had framed the broader debate with examples of crises turning everyone into instant experts; Ek instead emphasized that technologies carry extreme positives and negatives and should be steered toward desired outcomes.
His positive cases are concrete. Spotify could eventually “soundtrack every moment of your life” so music makes users “feel more”; Neko can monitor hundreds of lesions over time. Ek argues the technology industry has done “a terrible disservice” by failing to foreground such broadly beneficial applications.
On model structure, Ek expects the familiar coexistence of open and closed systems. Spotify uses frontier models alongside internally fine-tuned ones because some tasks require lower cost, greater efficiency or tuning unavailable from frontier providers; the host underscored the stakes with output costs of roughly $0.13 versus $30 per million tokens.
Ek’s tentative additional safety variable is compute: a system using 100,000 GPUs may be much more powerful than an open model running on a home PC. The hosts extended that into compute-based access and certification, while Ek kept the idea tentative—“maybe it’s something I’m missing”—rather than presenting it as a finished regulatory scheme.
Full transcript
Daniel Ek is with us. You know him, the co-founder of Spotify, which he started 20 years ago and ran for those 2 decades to an extraordinary state of affairs today: well over 700 million active users and over 300 million premium subscribers. But on January 1 this year, he shifted and became the executive chairman with me and David Freeberg. Daniel, how are you, sir? Welcome to the program.
Thank you so much for having me. It’s good to be on.
It’s good to see you. We’ve known each other for 20 years. I remember the first time Shaq, your partner, leaned into me at an internet summit or something and said, “Hey, check this out.” He showed me Spotify, but this was at a time when nobody had a subscription service.
1. Starting Spotify: Beating piracy and betting his own money on the record labels
We’re going to talk today about your new startup, which is incredible in the healthcare space. But take us back to the launch of Spotify, how that crazy idea got started, and how hard it was to break through with the music industry, which is also known as the hardest partners in the world.
I usually start off by saying that whenever anyone asks me about the story: I had hair when I began this, and by the end of it, I obviously ended up being bald. You can see that it wasn’t an easy journey.
All jokes aside, I started Spotify in 2006. I was 23 at the time, and the world looked a little bit different. This was pre-iPhone. Facebook existed, but it was college-only, and YouTube didn’t exist. This was before all the things we now take for granted.
The music industry at that time was in freefall because music piracy was rampant. We had Napster, Kazaa, and all these other services. There was an organization called the RIAA that went around and actually sued individual consumers in the U.S. for illegally downloading music.
I was sitting in Sweden and thinking to myself, “This is wild and crazy. There’s no way to put the genie back in the bottle.” My co-founder Martin and I were sitting around spitballing startup ideas at the time. Martin asked me, “If you could do anything you’d like to do, what would you do?” I said, “My real passion is music, but that seems like a really stupid idea.”
He asked, “Why is that a stupid idea?” I told him it was really hard because you had to do all these deals with the record companies, figure out licensing and copyright, and do all of these different things. He said, “Well, what if you did this?” Then he kept asking these “what if” questions, and eventually I ran out of reasons to say no.
I said, “I guess you could do something, and if you did something, it would feel like you had all the world’s music at your fingertips.” He said, “Well, that sounds pretty good.” I said, “Okay, let’s give it a try.”
We started trying to license music. Originally, the idea was to try to go global from day one, and everyone was just laughing at us when we said that. But I lucked out because it turned out that one of the worst music markets in the world was my home country, Sweden.
Part of the reason it was one of the worst music markets was that we already had superfast broadband from the early 2000s. We were living in the future, where people had such fast bandwidth connections that they could download pretty much whatever they wanted. So they were doing it, and iTunes wasn’t even available in the market. There were literally no legal options.
I think the music industry had lost 80% or so of its revenue. It was this crazy coincidence, this confluence of events, where my partner and I basically took all of the money we had made from prior startups and said to the record labels, “What if we guaranteed that you’d keep your bonus for the next year and meet your budget for the next year? If this really is as bad as you think it is, then you can shut it down after a year, but you still made the bonus.”
“If this actually works, you win as well. So you win no matter what.” They said, “Okay, this seems like an interesting idea.” Eventually, after a few years of negotiation, they agreed to it.
We launched in late 2008, and it instantly became a huge hit. We then took it to the U.K. and made it a huge hit there. Then, in 2011, 15 years ago now, we launched in the U.S. The rest is history.
2. Neko Health: The Spotify playbook for healthcare, the $499 body scan, and why US healthcare waits until you're sick
Isn’t that amazing? Five years in the laboratory just to get to the U.S. market is quite a journey. That’s a lot of pain and suffering.
The crazy thing is that I’m on the same journey again.
Neko, as crazy as it is, started the company 8 years ago in 2018, and we started it in Sweden in 2023. It took us 5 years to develop the product, try it out in Sweden, and then, once we had proven it in Sweden, take it to the U.K. and do it there. From there, we’re now launching in the U.S.
It’s the exact Spotify playbook again, but this time in healthcare.
One of the afflictions of great entrepreneurs is that you’re entrepreneurial. You just can’t stop.
Or maybe I just have a high pain tolerance. I’m not sure which one.
Once things got smooth, you had to get back into it. At what point did you start thinking about doing other things? Were you actively angel investing at Spotify over the years as CEO, while the business was building, with hundreds of millions of users, a successful public company, and things growing nicely?
You had a great leadership team—we’ve obviously met a lot of your team over the years. How did you start doing these other projects? I know you’ve got a few others, and now you’re focused on Neko.
It’s kind of crazy. I had forgotten about it, but my team recently showed me that I apparently did an interview with the FT in 2012 or 2013, just after the U.S. launch of Spotify, where I was already talking about healthcare. So I’ve been thinking about healthcare specifically for a very, very long time.
A lot of people talk about all the stuff that’s going badly in the world, but my view is that most things are actually going pretty well. It’s kind of up and to the right on the technology curve, with all the things that we’re getting. I try to focus on the problems that aren’t going that well.
One of those happens to be healthcare. I’ve been interested in the question of how we’re spending more and more money on healthcare but getting worse and worse outcomes, and what we can do about it. That really got me into the space.
I think healthcare is unique and special. I honestly didn’t have any money until Spotify went public, so I wasn’t a prolific angel investor.
What year was that?
That was 2018. We went public then, right?
I probably started doing a little bit of investing in 2018, but I had no money before then. It was literally all-in focus on Spotify for the better part of 12 or 13 years before I made any investments in anything else.
You and I were talking about it at this dinner a week ago or so. I had already started Neko at that time, and I realized that I loved building much more than I loved investing.
I came to the same conclusion. It’s so painful to watch companies you invest in get mismanaged, and you can’t do it. This was our whole dinner conversation the other day: you spend time on boards, advise a CEO, and they don’t listen to you. You pull your hair out thinking, “I would have done something different. Why didn’t you do this?”
If they are listening to you and you’re effectively instructing them on what to do, that also doesn’t work because they’re not leading. I became pretty committed to the idea that you find amazing entrepreneurs and founder-CEOs, give them money, and never look at it again. You’ll generally be better off than not.
The flip side is that if that’s really what you’re passionate about and you have the activity or agency to step in and do it yourself, you should do that.
The more instruction they need, the less likely they are to succeed.
So, Daniel, maybe you could just give us an overview of Neko.
Yeah.
And how did you come to that idea? How did you come to this specific idea?
I started researching the space a very, very long time ago. Maybe, as a quick primer for the audience, the U.S. is spending 18% of its GDP on healthcare. It's the single largest line item in the budget right now.
If you take heart disease alone, it's hundreds of billions of dollars being spent on that disease group in the U.S. alone. It's just insane. It's larger than the revenues of Fortune 10 companies—just one disease group.
So it's a gigantic problem. I started looking into everything, from drugs to whether you can make drugs more efficient, to cancer, and basically started targeting what the real reasons are why it looks that way. I think the real conclusion ended up being that there are lots of reasons, not just one.
We see a few really big ones. Cardiology, or heart health, is a major one. We see skin cancer, melanomas, as another big one. There are a few major groups of illnesses that are overrepresented, but the major one is really chronic disease overall.
The crazy thing about chronic disease, when you read about it, is that if you discover these things early, they're totally preventable. The cost of dealing with them is very small, and the suffering for the person having them, and for their family and community, is very small too.
The key thing is that if you discover things early, you're actually in good shape. So then the question is: Why don't we discover things early? It happens to be that we don't have enough data.
Everyone in the entire healthcare industry agrees that we have to take healthcare from reactive to preventative health. But that's where all the disagreement starts: How do you do that?
My co-founder, Hjalmar Nilsonne, and I believe it all starts with having better data. If you have better data, you're going to be able to discover things much more easily. That, in itself, gives you a chance of discovering those things much earlier, which then creates this virtuous flywheel that can change healthcare.
So, predictive data that allows you to predict future adverse health outcomes and course-correct ahead of time?
Yeah, that's the holy grail—to get to that point. Essentially, our view is that the way you get there is that you need a lot more data than you have today, both from each modality that you gather data from and across multiple modalities. You need it longitudinally, so you need it over time.
If you had a lot more data and you had it over time, you'd have a much higher likelihood of doing those kinds of predictive analyses. Then we said, “Okay, how do we do that?”
It so happened that if you think about what's really happened in the last 20 years, one of the biggest innovations, of course, is smartphones. We've had more and more sensors packed into those smartphones, so there are more and more cheap sensors and cheap electronics available to us.
Our view was, “Okay, we should be able to utilize this to build a lot more sensors than have been built before.” The other thing is that we started before the LLMs and everything like that. It was then called machine learning, but today we talk about it as AI. We should have a lot better intelligence.
If we have all of these massive datasets, we should be able to make really good predictions. Similar to what we learned with Spotify, the more data we're able to gather, the better predictions we're able to give you about what song to play next.
This is exactly the same thing that you now need to do in healthcare, but obviously it's across multiple modalities and across longitudinal datasets as well.
Is there evidence today that some of this early observational data can affect ultimate health outcomes? I can see this working for metabolic conditions—things related to exercise depletion rates, blood sugar, and sleep—but ultimately, the correction there is typically diet and exercise.
What are some other examples of major diseases that we can identify ahead of the curve and course-correct? And maybe just explain what Neko is for people who don't know.
Yeah. Maybe I'll start there, and then we can get into it.
In a very simple way, the experience of Neko today, and in Gen 2, is $499. It is a reinvented healthcare experience from the ground up, vertically integrated. We do everything from building our own facilities, to having our own nurses and doctors on staff, to building our own diagnostic equipment, to doing everything on our own on the software side.
It's really an end-to-end solution that we've designed from the ground up, and we designed it with the goal of creating the most valuable hour you can invest in your health. That's the mission we have, and that's what we come to work to do every day.
The experience is that you come in, and the first thing you typically do is have blood drawn. We then measure 53 blood markers. You go through our skin rig, where we have a camera system that takes more than 6,000 super-high-resolution images across your body. We index every mole and lesion, as well as all of the possible rashes and redness you may have.
We then look at your heart and blood circulation, and we look at grip strength—all the traditional markers for which we have very good scientific evidence that they're valuable indicators of health.
At the end of this visit, which lasts about an hour, you get uninterrupted time with the clinician. You go through all of the results, including the blood work, during that hour. You have a complete set of results that you can go through with your doctor, and the doctor will answer any questions you may have.
You also sit down with that clinician and go through results about things you may want to improve when it comes to your health. That's the experience we're launching with today in New York in the U.S., and obviously we'll launch in many more places across the United States over the coming months.
And just the identification of things ahead of the curve—because I've heard of other attempts. There was One Medical, there was Forward. There were a few other attempts at making these integrated centers where you go in and get preventative checkups with their own instrumentation. I think I've heard the thesis a little bit before.
What have you identified that might be novel, Daniel, that we haven't seen in the past?
I think, first and foremost, the important thing to mention is that, as I mentioned, we started this in 2018. This company has been around for 8 years. We've done more than 100,000 scans already.
We publish data every year. We've just published our third-year data survey, where we go through the health outcomes we have across our members, and around 1% of our members have a serious underlying medical situation that is undiagnosed when we discover it.
The good news is that many of our members are healthy, and we're just talking about improving their current health status. As you correctly pointed out, a lot of that ends up coming down to stress, diet, sleep, and those things.
But the most encouraging thing we're finding now, as we've done these studies for 3 years in a row, is that the people with the worst health status are the ones improving the most. It turns out that having a clear visual understanding of where you are in your health journey today, and having a clinician sit down with you and explain the situation and possible remedies, is a huge factor in getting people to start improving their lifestyle.
We've seen members go from years of smoking to not smoking as part of doing Neko and prioritizing their health. We've obviously seen people with severe underlying medical illnesses discover and treat them, resulting in much better overall health.
So it's really all across the board, but I'll give you a very concrete example. Take dermatology as a great example. The average person at Neko today has 950 moles. It's pretty crazy.
The average person has 950 moles?
Yes.
That's insane. Some percentage, you can't tell, I guess, is the punch line here. They're not detectable by the—
Well, if you really went through your skin, you'd be surprised how many you'd find. It's a big dispersion. Some have many, many thousands. Some of them have—
Seems like a perfect AI application, obviously.
Exactly. So if you think about it, across a normal doctor's visit, how likely is it that that doctor will check out all 950 moles that you may have? It's not going to happen.
Normally, what happens in a normal doctor's visit is that you may say, “Look, I have a mole,” or the doctor sees one and says, “Hey, let's take a look at that mole. It looks a little bit funky.” Then they bring out their tool and inspect it.
Usually these days, they have software that checks it. They take a camera across many different frequencies to discover both the size of the mole and its texture, and so on and so forth.
We literally do that. We first have an AI system that flags possible risk factors, and then we have a human clinician who goes through and reviews those as well. Should we, even at that point, find something, we then have expert dermatologists on staff—multiple of them—who also review that result.
That’s just one indicator of what happens in one visit. But let’s now presume the scenario that we’d have multiple visits and maybe, one year, we don’t discover anything. Because we’ve cataloged and indexed every single one, we can now look at them longitudinally from one year to the next. So it may be growing abnormally. These are things that, if you think about it, even the best doctors in the world can’t possibly remember how your mole may have looked from one year to another.
So again, back to your point, this is a perfect application of AI, right? But we look at this as a perfect application of how health care should look. It should seamlessly be both AI and amazing clinicians in a great packaged experience. So that’s essentially what we’re trying to do with Neko.
That’s great. Are you actively running it as CEO, Daniel, or are you operating part-time?
Yeah, no. Similar to Spotify these days, where I have 2 great CEOs, I have an amazing co-founder and partner, Hjalmar, who’s really the brainchild of Neko. He’s the one who’s doing most of the work, even though I’m sitting here and taking most of the credit at this point. But he’s the real brain.
Great. You know what I’m struck by? The cost of this is so cheap: $499. You also have Function Health and Superpower, 2 other great services, at about $350 or $400. They just do blood work.
But if we look at the spending, right now health care is number 2, Social Security is number 1, and in fact, your favorite coming up is the interest payments. Friedberg will overtake our health care spend. You could actually have folks come to this every year or every other year for 4% or 8% of the total cost we spend on health care.
I’m curious: Is this something you expect people to do yearly or every other year? What impact do you think an investment of every American getting this done every year, every other year, or maybe every 3 years—even, which would be down to 2% of our budget—would have on the forward-looking budget and spend? Have you done those kinds of calculations?
Yeah. We obviously have a view, but I think it’s too early to say, so I don’t want to sit here and say that we know conclusively what will happen. But what we do recommend to our members is that they try to do this on an annual basis.
The way I make this analogy is that most of us today are trying to go to a dentist on a yearly basis. For some reason, most of us don’t do an annual health checkup. So for me, this should be as regular as going to the dentist or doing anything else, and it should be an amazing experience.
This should be, again, as I said, what we’re really doing as a mission here: trying to create the most valuable hour you can invest in your health. We know a lot of people are busy, so we want to be brutally efficient with people’s time. If you don’t have any questions and you’re just interested in getting the test results, you could probably get by in 30 or 40 minutes and do it. We’re not talking about a lot of time.
Obviously, $500 is still a lot of money for people, but I do think it’s a great investment if you can afford it.
That $500—do you lose money on every visit or break even, and then there’s some profit that comes from maybe the upsells? Take us through the economic model here and then how that will work with the insurance industrial complex, as dysfunctional as it is here in America.
Yeah. The benefit of being vertically integrated is that we’ve been able to build everything ourselves, which means we can also cut costs in a pretty dramatic way. Within that $500 a year, our unit economics are positive, and we have clinics today that are profitable and that we’re already operating.
We think this is a good price point to enter in because it’s a great value for people. But because we do all the things ourselves, we’re able to still make this economically viable for us—to create a great business and to be able to grow and invest in that business so that we can spread it to more places as well.
So, Daniel, are you focused exclusively on this project? Do you have other stuff you’re working on?
I’m not exclusively focused on it, but it’s obviously one of the things that’s taking a lot of my time, and rightfully so. It’s one of the biggest problems in the world.
But I’m still spending time with Spotify. I’m executive chair there, so that’s still my baby, even though I’m not involved in it on an everyday basis.
We have a few other companies with Prima Materia, which is a company I created together with Shak, who I mentioned earlier. We’re trying to really do what I mentioned before: How do we pay it forward and almost be like the greatest co-founder you could possibly find? That’s the ambition of Prima Materia.
The health care thing really perplexes me. I look at the biggest issues for affordability for people in the US right now: buying a home, dealing with educational costs, and health care costs are the top 3. I feel like we can solve education costs and home costs with certain market incentives that have been distorted. But health care is much more complex.
As you look at the US system beyond just preventative care, the cost of servicing patients in a hospital—why is it $15,000 to get stitches at an ER? Why is it $20,000 to get a drug that costs $30 to make? Why, when the doctor is making $200,000 a year, does the doctor seeing me for 8 minutes cost my insurance company $6,000?
What is going on, from your view, that extends beyond getting in front of all of the catch-up problems and so on? What’s structurally going on with the health care system in the US?
And if you were emperor of the US for a day and got to say you could do anything you wanted, what are the top 3 things we’re missing that maybe Neko doesn’t necessarily address today, but that you would recommend we fix?
I’m definitely not an expert on the US health care system, but for me, it’s, “Show me the outcome, I’ll show you the incentive.” It’s one of those classical problems.
If you really think about it, at the moment, the entire system is predicated on—and the health care system was built around—a time when we were dealing with infectious disease. That was how the health care system was built and dealt with. So all the incentives are really around that, which means we’re fixing you acutely when there are massive amounts of symptoms.
Where we have to go is toward a health care system that is preventative. That means it has to be much more long-term today. I’ll just mention one of the problems that exists: Normally, your health care is tied to your employment in one shape or form. Because the average tenure of your employment isn’t very long, it means that you may have only 2 or 3 years at one employer, and then you switch. When you switch, you switch insurers, too.
One of the questions then, if you’re that insurer, is whether you should invest in something where the payback time may be 10, 15, or 20 years. Those are just some of those incentive issues. How do we look at the ROI of some investments that may be 10, 15, or 20 years into the future, and who takes those investments?
These aren’t easy things to answer, and I don’t claim to have all the answers. But one of the things we’re trying to do is bring the cost down so that the ROI doesn’t have to be an investment of millions of dollars by the insurance industry or employers, speculatively hoping to get it back in 20 years with very little data.
Our view is that if we lower it so it may not be a million dollars—it may be, I’m making it up, tens of thousands of dollars—even if that payback time were 10 or 20 years, I think there would be a higher propensity and likelihood for a pickup on that.
The second thing we’re trying to do, obviously, is add more data to the system. If we had more data, then with these multimodal, longitudinal data sets, it’s much more likely that we could actually see the efficacy of these things over time.
This is also part of the reason why we’re investing in releasing our data every year—what we’re finding—so that people can see what happens both on an aggregate Neko population, which is also super cool, because normally health care systems are very country-specific.
This is one of the few health care companies that isn’t global but is at least a multi-country system straightaway.
So that's also super interesting. What are we finding in the UK that may or may not be similar or dissimilar to the US? What did you find between Stockholm and the UK? I can take some guesses, but I'm guessing people in Sweden were much more fit. In the UK, you probably had more cardiovascular disease and diabetes.
Those are the two, I believe. You tell me if I'm right. Aren't those the 2 biggest spends in terms of health care spending? I think diabetes is number 1. I could be wrong, but cardiovascular disease and diabetes seem to be the big 2. So what did you learn between the 2 populations?
Well, I mean, it's really early, and it's still on a relatively small base. It's 100,000 scans, as I mentioned before. But I don't think we're ready to speak about population-level health outcomes, but as we grow in Neko, I think we will. And that's going to be super cool.
We're already doing lots of clinical trials. We've done 4 already; we have 2 underway, and we have another 4 we're going to do. So we're partnering with the research community as well, deeply, around these data sets that we already have today and that we will gather more of tomorrow, too.
I think my point with all of this is really just to say, look, for people in tech, the amount of data that exists in the health care system is actually not—the data sets are not that large. That's the crazy part. If we can 10x or 100x the amount of data that becomes available, we're going to be able to draw super interesting conclusions about that. Not just Neko, but the whole health care system.
That's really the bet we're making. Putting the wearables in there is going to be really interesting with the Oura, the Whoop, and the Fitbit. You start with Apple Watches, and you start correlating that data with the blood work. I noticed Whoop offering labs now through Quest, and I don't think—maybe you can sync your Oura or Whoop with some of those, but I've never seen any data.
We have sleep, recovery, and heart rate, but I don't think anybody's put the blood work together with that, and then your system. I mean, this could really change everything in terms of health care if we can correlate that. Maybe you could speak to the impact you think wearables are going to have.
Do you think you partner with them, or do you just have a Neko wearable you eventually make part of the subscription at $9.95? It does seem like these are commodified for the basic 80% of it, I'm sure.
Yeah.
So what are your thoughts there?
Already today, we do allow you to import your Apple Health data. So if you do have wearable data, we will take that. Anyone who has a wearable—and a lot of the big ones, Oura, Whoop, et cetera, do allow you to write a lot of that to Apple Health as well. That's just one example of that.
We would love to have more data, because that enables us, again, when the clinician sits down with you, to give you a much more 360-degree perspective about your current health status. So that's one of the things that we encourage.
But I do want to say also that this is just like the service we're launching today. It is very rare in itself that, when you create a diagnostic product, the normal way of doing that in this space is you spend X amount of years doing R&D for that thing, then you literally spend the next 10 years selling exactly that thing.
With Neko, within the 3 years since we launched, and now as we're launching in the US, we're on Gen 2. So we've already upgraded our diagnostic infrastructure. I think what you should expect with Neko is that we're launching with one thing today, but we're going to keep adding more and more value, and we're going to keep adding more and more valuable diagnostics that we can find and put in at the price point.
That's kind of the goal. This is just the beginning, and the goal is obviously to make this even more valuable so that we can deliver against our mission of having the most valuable hour you can spend on your health.
3. AI: Tech's failure to sell the upside, open vs. closed models, and regulating compute
Daniel, do you want to take a step back and talk about what's going on broadly in the tech industry right now and AI? What's your view on pacing the—
I hear these questions. Yes.
Do we need to pace the frontier? I'm struck by the fact that, if we look at the world, and certainly over the last few years, I always say we're walking into COVID and every one of us became a virologist. We were all of a sudden supply-chain experts after the Ever Given got stuck in the Suez Canal. Then, with the Ukraine war breaking out, we were all military experts. Are—
Are you talking about the All-In podcast right now?
We're just talking in—
It's actually, we renamed it the All-In Laboratory Pod. It's the All-In Lab because we are just—
We're just a lab. We need to be regulated because our opinions are so dangerous, Daniel, that maybe Spotify's podcasting could regulate us because our opinions are crazy. Yeah. And my point is just that everyone expresses these things with such extreme confidence, and it turns out most of the time we end up being completely wrong, even the biggest experts.
I'm not sure I have much value to add, except to say that I believe we're still in the innings where we can alter the impact of how to use these technologies. My view is that every great technology has extreme positives and extreme negatives, and it's really up to us now to steer this toward the outcome that we want.
I think it's great that the debates are being had, but honestly, I have no strong opinion one way or the other about whether we should pace it or not. I just know that if I look at health care and I look at what we're doing at Spotify, it's truly amazing.
The perfect music product is one that I've described many times, and I think we're getting closer to it now, which is awesome. Right now, you can put together a better playlist yourself than what the system could do. But I think in the future, we can soundtrack every moment of your life in an amazing way, where you just feel more.
The power of great music is that you will feel more. If you're happy, you'll feel happier. If you're wallowing and feeling sad, you'll feel even sadder in that moment. That's thanks to AI.
Then, if I look at the example we talked about with dermatology, 950 moles being able to be tracked over time, keeping you safe—these are positive, amazing examples with AI. Yet we tend to amplify the negative. I want to see more positive examples of AI. I think we, as an industry, have done a terrible disservice by not talking about all the really crazy-positive stuff that we can use AI for that actually greatly benefits everyone.
In that sense, do you also think about the benefit and the value of open-source AI and open-weight models? You can now look at something like 13 cents per million tokens of output cost versus $30. That's an incredible difference that unleashes this technology across a broad spectrum of businesses and individuals.
There is no concentration of value or wealth among a handful of businesses or individuals if we can really make these open-weight and open-source models proliferate. But obviously, this regulatory conversation may restrict the expansion of open-source and open-weight models. Do you have a view on the importance there?
I think technology has always gone between open and closed, right? We had Windows versus Linux in the first iteration, iOS versus Android. We've seen this play out so many times, and where we tend to net out is that we tend to have both.
To the extent that I have a view about it, I've very positively advocated for open-source models before, and I will keep on doing that, because I think that brings a lot of innovation that will be really helpful for the ecosystem as well.
I completely agree with you that, if you look at these now, and certainly what we're seeing at Spotify, we're obviously using a lot of the frontier models, but we have a bunch of fine-tuned models ourselves, too. We tend to do both, and that is actually making us more innovative. It's allowing us to do certain things that we couldn't do with the frontier models, either because of cost and efficiency reasons or sometimes because we just couldn't tune them the way we would want to.
I personally think it's going to be both. But one of the things I heard that is a perspective I haven't heard before, and maybe I would encourage people to consider—not that I'm an expert on all these things—is the amount of compute.
If you think about it, for me, this isn't just about the intelligence itself of a single model, but the amount of compute you're doing can indicate something, right? If I'm using 100,000 GPUs for something, that's probably going to be a lot more powerful than if I'm running an open-source model on my home PC.
It’s probably very unlikely that I’m going to be able to do an enormous cyber thing. Certainly, if the companies have mythos class models to protect themselves, I should be able to do a lot of damage with an open-source model on my home PC. I find it fascinating to think about whether there’s a way of looking at the amount of compute as one factor in that as well.
Right. Assuming model equivalency, compute is a key metric of defensibility. That’s really where we can start to build guardrails around who can have access to the most compute and how you get certified for compute, as opposed to being certified for software. Wasn’t that the original concept, 2 years ago? One of the original ideas for limiting this was how many teraflops or how many GPUs you had?
No. What they did—that was how they wrote the regulatory laws around regulating models, right? So they basically set a model that was trained on X number of teraflops, and it was like a California—
This is static, like—
The California genius assembly got together and said, “We figured it out. We know how to classify scary models and bad models.”
How many Ethernet cables are in your data center as a proxy for how powerful—
How many watts of power went into making this model. But I do think your point of compute as a roadblock is probably a good one. This is what it used to be. You remember when Cray supercomputers came out—
And there was limited access, and they were considered a security risk. You had to get approved in order to get access to a Cray supercomputer because you could use them to design nuclear weapons, crack codes, and do all the things people worried about. Yeah. Today, a Cray supercomputer, I think, is less powerful than a washing machine or some product we probably have in our kitchen. It is a really great point, actually, Daniel, because if you’ve got strong compute installed as a defense capability, assuming model equivalency, you’re going to be hugely advantaged.
I haven’t seen it in the debate, at least, and I’m surprised, but maybe it’s something I’m missing.
4. Back to Spotify: Podcasting's open standards and the Stardoll origin story
Hey, before we let you go, I guess 2 questions for you. Take them in whichever order you want. What’s the strategy, and how does Spotify look at something we care about deeply, podcasting, which is built on open standards? You guys have become a major player in it, and open standards are kind of moving in one direction, while Spotify has a lot of proprietary stuff. I’ve talked to you about this and some of your team members: Can we keep supporting the open-source part?
The second piece people don’t know is that there was this incredible company, Stardoll. It was one of the first companies in Stockholm, Sweden, to ever get funding from American VCs. You were the CTO, if I remember, or the intern for a couple of years.
Well, I was the intern, CTO, whatever.
People don’t know this. This was an incredibly innovative company in terms of many variables, like digital goods and subscriptions and all that stuff. So maybe take us through either or both of those questions, because those are 2 personal things that I would love to hear about if we were at dinner.
Yeah, I mean, look. Again, as I said, I think the open and closed are going to be 2 things that are going to exist side by side. The interesting thing about Spotify today is that, on the one hand, of course, we’re a huge platform in podcasting, and we have a destination where there are certain things that we’re doing that general podcasting may or may not be doing. One of those will be how we’re handling comments and so on that are specific to the platform, but we actually do have tools.
What a lot of people don’t know is that we also, through our platform, aggregate and distribute podcasts onto a lot of the other podcasting platforms. There are a bunch of podcasters that are using our tools to distribute to some of the other platforms, too. We’re obviously doing that through open standards. So I don’t think the answer for us is one way or the other. We think both can coexist, but obviously, if you want to innovate within a standard, the key is to have other people agree to that standard, and that can sometimes take longer. So it might be harder to innovate. Spotify’s approach has been, “Let’s support both in tandem.”
But then there are 2 features I just wanted to make sure I ask you about. There’s one for going live. This is part of the new standard, so having that on our page. When we go live, it actually respects the RSS feed and sends that note out. Then there’s another one called value and donations, where the person who has the podcast—if they happen to be doing donations on whatever platform, or they do them through PayPal—those 2 features are what the podcast OGs in the underground really want Spotify to support. I’m just doing my job in representing as a 17-year podcaster. Please add or support those 2 if you think it’s worthy of doing.
All right. Well, it’s no longer my decision alone, I should say, but I will certainly bring—
Right to the chairman while I’ve got him on the line.
Yeah, of course. Of course.
But tell us about—
I’ll definitely bring the feature request to the team.
Tell us about Stardoll while you wrap up here. What did you learn there when you were 20 years old—21, 22, whatever it was at Stardoll?
Yeah. You know, it was a great experience. It brings me back down memory lane. I started at Stardoll just before I started Spotify. There was an entrepreneur called Mattias who was one of the original entrepreneurs in the Swedish ecosystem. He had been around since the dot-com bubble, and it was originally a site called Paperdoll Heaven in Turku, Finland, created by an old lady and her son, of all places.
Index Ventures had invested in it and taken a majority stake to help with it because they had no idea how to scale the site. Index and Mattias came in with the idea of rebranding it and building it into Stardoll. They came to me one day and asked around for people who knew tech and could be technical enough to help build it.
I had already decided I wanted to go build Spotify, but Danny Rimer is very good at convincing people to do things. As a favor to him, and sort of hoping that he would come and fund Spotify—which, by the way, didn’t end up happening—I said, “Okay, fine. I’ll help you for, I think, 6 to 12 months, something like that.”
So I went over to Finland and checked it out. It was kind of wild and crazy because I think the average page-load time at that time was 4 minutes. It took 4 minutes to render a simple page. Basically, the server was completely breaking down, it seemed. I was like, “Holy, this is crazy. I wonder how much more traffic you could do if you actually made this work in a snappy way?”
We basically rearchitected the website. I hired an entirely new technical team and brought it down to under a second of load time, and obviously the traffic exploded. Sequoia invested. I said, “Okay, well, I’ve done my part. Thank you,” and then I went off and built Spotify.
It’s amazing, these origin stories—what happened right before the thing is always interesting to me. Listen, Daniel, continued success. If people want to try Neko, where can they go?
We launched the first one at 300 Lafayette. Go sign up on nekohealth.com waitlist, and we’ll try to get to you as soon as possible.
And this is in New York, 300 Lafayette Street. Only in New York so far, but we’re expanding quickly, both the number of venues we have in New York. We’re launching in Florida as well, in Miami, and we’re launching in D.C. We’re going to try to launch all over the U.S. in the coming 12 to 24 months.
Yeah, it’s great. Right on the street. That’s back in the day when Pseudo had their big studios; they were right around the corner from you. Wow, talk about memory lane. All right, continued success. Everybody, go to nekohealth.com and sign up for the waitlist. I’ll see you next time. Bye-bye.
I’m going all in.