Daniel Ek - Enabling Creators Everywhere - [Invest Like the Best, EP. 242]
What this covers
This is the fifth-anniversary episode of Invest Like the Best, with host Patrick O'Shaughnessy interviewing Daniel Ek, founder and CEO of Spotify. It starts well outside music, with Bits Versus Atoms. Ek explains why digital ideas attract a hundred competitors and winner-take-most outcomes, while atoms companies often have no Minimum Viable Product and must commit capital years before reaching market. He also describes how he learns an unfamiliar field: he explains it aloud to people who ask questions, and he counts it as understood once he can argue the opposite.
The middle of the conversation ranges across large problems. On climate, Ek calls the challenge a Coordination Problem, citing green steel, which he says has existed since the 1960s and 70s. On healthcare, he says people rather than drugs drive US costs, and he argues for predicting disease instead of reacting to symptoms. Both claims are flagged as contested. He then turns to Europe, which he sees as underperforming in tech partly because it is spread out, whereas US output has leaned on San Francisco and Seattle. He credits Sweden's sustainability-reporting mandate for European leadership on climate and diversity. He also forecasts machine-driven parallel experimentation in biotech.
The Spotify material covers Bad Bunny topping the platform's listening, podcast moments pulling unpopular music genres upward, and the claim that new platforms like TikTok can still emerge. Ek also gives personal frameworks: patience, Charlie Munger-style Inversion, assigning made-up values to unknown variables, and what he learned from missing Instagram, Uber, and Skype as an investor.
Ek argues that the largest unsolved problems (climate, healthcare, and frontier atoms industries) are coordination and sequencing problems with no established playbook, and that the winning move is to start executing early, enable creators and users rather than control them, and make decisions by classifying their reversibility and simplifying their variables.
- Atoms industries lack the cheap experimentation, standardized capital, and distribution rails that made digital companies scale, so they require patient, large upfront bets.
- Enablement (free access plus monetization of specific mechanics or experiences) tends to outperform control, whether for creators, labels, or platforms.
- Good decisions come from identifying one-way versus two-way doors, reducing variables, and inverting problems to find what must be avoided.
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Ek's single biggest virtue is extreme patience, since change takes a long time and people overestimate short-term impact while underestimating long-run impact, a pattern attributed to Bill Gates that Ek says he repeatedly makes himself.
“in the short term, people tend to overestimate the impact and in the long run, they tend to underestimate it”
Green steel technology has existed since the 1960s and 1970s, and at scale it could be produced at roughly the same price as current steel, which matters because steel accounts for about 7 to 8 percent of world emissions.
“you're talking about seven or 8% of the world's emissions, that's totally possible to change. And you can even, as an output, create it at pretty much the same price as the current steel, when it's at scale”
The vast majority of the technology needed to drastically reduce carbon emissions already exists, so the core climate challenge is a massive coordination problem in which fossil-based alternatives are cheaper than scaling clean options fast enough.
“the vast majority of the technology we need to drastically reduce our carbon footprint, exists. It's not about innovating”
Digital experimentation is so cheap that for any internet idea there may be around a hundred competitors, and the winner then gains economies of scale, global reach, and rapid growth on top of existing platform rails, producing winner-take-most dynamics.
“For any idea now on the internet, they're already today probably a hundred different people that are trying to compete in that thing.”
Gaming moved from paid games to free games with monetized mechanics, and giving away the core experience while charging for boosters made more money, showing that enablement can outperform control.
“maybe the best thing is to just give away as much of the game as possible, and then there are certain mechanics inside of the game that if you really enjoy the game, you'll probably want to pay for that thing”
Music is globalizing: artists from places like Korea break through internationally, and Bad Bunny, a Latin artist, was the most-listened artist on Spotify last year, which ends the era in which only Anglo-American artists could reach a global stage.
“the most listened to artists as of last year was Bad Bunny, which is a Latin artist in the global world”
In US healthcare the main cost driver is not drugs but the large number of people required to deliver care, which is why automation and efficiency are plausible solutions.
“it's not really the drugs that is the major cost impediments in healthcare in the US. Instead to cut to the chase, it's really all about all the amount of people that's needed”
Charlie Munger, working as a meteorologist for the military, inverted the question from improving weather forecasts to keeping fighter pilots alive, focusing only on catastrophic weather risks like snowstorms rather than average conditions.
“What will kill the pilots and how can I avoid that from happening?”
Atoms companies often have no minimum viable product, so they must commit far more capital upfront and wait many more years before reaching market than digital companies do.
“there isn't a minimum viable product oftentimes, or that possibility. So you have to bet much more capital upfront”
Centralized aggregator platforms and decentralized models like blockchain will both have a place, just as Apple's locked ecosystem and Android's open ecosystem both succeed with different customers.
“the centralized aggregator model works. But what is blockchain? It's certainly not a centralized aggregator model. That seems to work too.”
Europe has over half a billion people, roughly the same GDP as the US, and strong science and universities, yet it has underperformed in technology over the last two decades, largely because it is decentralized rather than concentrated into a few clusters.
“It's got roughly the same GDP as the US, so relatively well-developed as a region as a whole”
The common view that platform evolution is mostly complete is wrong; TikTok's rapid growth out of China shows that new platforms can emerge even after Instagram and Facebook seemed to have won.
“the common view was that there wouldn't be any new platforms. The Instagram and Facebooks of the world had already won, and then out of China comes hype dance with TikTok”
Government investments made decades earlier, such as UK initiatives in the 1980s and 1990s that now make Cambridge and Oxford prolific, and US DARPA's role in the internet, produce breakthroughs only after a very long lag.
“we're seeing Cambridge and Oxford right now is on fire because of initiatives that the UK took, I believe, in the eighties and nineties”
Healthcare should move from reacting to symptoms to predicting and preventing disease before it appears, because costs are dramatically higher once a problem is present and symptoms appear late in the disease cycle.
“the actual solution to healthcare in the end, has to be that we have to go from reactive, which is the whole system, to proactive”
Personalized inference depends on first building systems that collect data and find reasonable proxies for structuring it, because simple inference comes only after a data engine exists, as Spotify's own evolution from one homepage to full personalization demonstrates.
“the first step was create an engine that collects the data, find a reasonable proxy for how to structure that data and only much later you'd make simple inference”
Multi-variable decisions are hard usually because two or three variables are unknown, so a leader can temporarily assign made-up values to reduce the problem to a minimal set, map the outcomes as a decision tree, and pick an option when none is expensive, researching only where the outcomes are painful.
“what I typically end up doing is I invert down to the minimal amount of variables that you can imagine. So just remove the other ones and make numbers up”
Spotify observed that winning podcast moments changed when and how people use the platform, and this pulled previously unpopular music genres upward because music started playing right after podcasts.
“because we were winning in some of these podcast moments, we're starting to see some of those musical genres then coming up too”
If you remove San Francisco and Seattle, the US tech economy is not in great shape, so US technology output has been highly concentrated in two clusters, while Europe is more spread out, which may explain part of the performance gap.
“if you take away San Francisco and Seattle out of the US and you start comparing now the US versus Europe, you would see that US actually is not in a great shape”
The platforms that enable digital marketing and distribution (Google and Facebook for promotion, app stores, AWS) do not exist for atoms products, and atoms companies often sell B2B or enterprise-style even when the end user is a consumer.
“the AWS, a lot of that infrastructure just don't exist for how you promote and market and distribute a atoms based product”
Platforms historically used one-size-fits-all models because they needed users to learn to interact reliably, but the next internet phase will let every type of content exist and be monetized, moving toward ecosystems where creators control their audience relationships.
“It's really about moving from one size fits all to this ecosystem approach where there will be more control ultimately, to whoever creates the content”
Venture funds have shortened fund cycles from three years to one year and make more bets, which conflicts with the much larger, longer-duration bets that atoms and healthcare transformation require.
“given how their model works today with fund cycles going from three year to one year and making more and more bets”
As an investor Ek missed Instagram, Uber, and Skype, and the lesson is that investors should look for highly driven, learning-machine founders rather than fall in love with the idea, since pattern recognition about founder traits was available in hindsight.
“what I should have realized in all of those examples is that there were highly unique people that were learning machines and incredibly driven”
Ek learns best by explaining concepts orally to people who ask questions, and he considers a subject grasped when he can argue the opposite is false, using inversion to check understanding.
“I can invert whatever I know and explain why the opposite is not true, I usually have grasped the subject well enough”
Even within European countries there are multiple mini-clusters at different maturity levels, such as London versus the rest of England, and Stockholm punches above its weight partly because of Spotify and Klarna and earlier dot-com failures that taught entrepreneurs.
“Stockholm, even though it's not a very big city, it's disproportionately punching above its weight”
Sweden's government mandate, over a decade ago, requiring large employers to publish sustainability reports covering climate, diversity, and gender pay gap, created public debate, benchmarking, and transparency that now underpins European leadership in those areas.
“the Swedish government, I believe more than a decade ago, mandated that if you were over a certain amount of number of employees, you had to have a sustainability report”
The smartest record labels already act as service providers to artists, and a label run by Ek would push further toward enablement over control, because artists now run their own social media and labels no longer coordinate messaging.
“The point being is the best labels have realized that they are in service of the creator. Their job is almost to try to enable the creator”
Podcast discovery is a bigger problem than music discovery because a podcast can require 20 to 30 hours of commitment from an unknown host, whereas a song is a three-minute decision, so Spotify expects automated merchandising and trailers to yield a 10x to 20x discovery improvement.
“it may be 20,30, even an hours worth of your time. So you're going to pick that very carefully”
Cars now warn drivers of problems early and plug into diagnostics that pull data directly, while doctors still rely on questions and visual checks; healthcare should be turned into the car model of data-driven diagnostics.
“That is what we have to turn healthcare into. And it's so obvious when you say it like that”
The best record labels recognize they serve creators, and a label that embraces enablement rather than control would expand what artists can do and typically pay out better over time, while some artists will still benefit from scarcity.
“The smartest ones would be very service centered and try to create more enablement rather than sort of more control”
Patrick notes that we know more about what is going on inside every system in our lives than our own bodies, framing healthcare as an observability problem that streaming data could address.
“we know more about what's going on inside every system in our lives than our bodies”
Ek says the current cultural belief that founders can do no wrong is overstated, since founders are not always the right leaders, carry strain, and can be hubristic.
“I think right now, we are in this world where founders can do no wrong... I don't think founders are always the right thing”
Biotech is seeing picks-and-shovels infrastructure companies emerge, and machine-generated parallel experimentation (hundreds of concurrent experiments instead of one at a time over weeks or months) should materially accelerate innovation over the next decade.
“the common norm is to run hundreds of concurrent experiments. Some of them, the data scientists aren't even aware of because it was machine generated”
The main barrier to commercializing research from institutions is not the technology itself but the opaque technology-transfer process, which is a structural inhibitor that must be solved to speed up innovation.
“there are also a lot of structural inhibitors, maybe more so than the technology in itself”
The idea is not what matters most; roughly 97 percent of success is execution, and merely starting to execute places a founder in the top 1 percent because most people only talk about ideas.
“the idea isn't that important, what's really important, 97% of it is really all about how you execute against that idea”
The most important thing Ek has done is create an environment where failure is acceptable, so teams test things they expect may fail and pivot, with decision culture allowing mistakes without risking the whole business.
“I think the most important thing I have done is probably try to create an environment where it's okay to fail”
Persevering requires being stubborn enough to ignore people who try to pull a founder off course, yet wise enough to listen for what critics actually mean, since valid feedback often sits beneath their words.
“you need to be stubborn enough to not pay too much attention to them, but wise enough to really listen to what they're saying”
Most people fail to distinguish irreversible one-way door decisions from reversible two-way door decisions, and both mistakes, treating everything as irreversible or everything as reversible, are dangerous, so each decision type needs a different process.
“Most people cannot make the distinction of what's what. So they'll either treat everything as the one-way door decision or treat everything is a two-way door decision. Both are incredibly dangerous.”
Commercializing research from universities and institutions is opaque, and solving technology transfer needs standardized contracts comparable to seed-stage SAFEs, which atoms companies currently lack.
“how you commercialize research from institutions, it is a fairly opaque process to begin with”
Future drugs and therapies should be personalized to the individual, with the specific therapy targeting a person's particular condition, using the same kind of inference that now tailors Spotify experiences.
“the specific therapy that targets whatever you may have, should be completely personalized”
Within Europe, each country and even each city has its own mini-cluster at a different maturity stage, such as London and Stockholm, and these clusters are now growing rapidly as talent and capital disperse.
“even within Europe, it is a bunch of mini clusters in itself, even within a country”
Europe will likely play catch-up to the US and China in digital businesses but may lead in gigantic societal problems like climate and healthcare, where coordination problems and unknown playbooks reward its breadth of innovation.
“that's not where Europe will have a leadership position. It's just going to play catch up to the US and China”
Audio is even more nascent than video: fewer audio minutes have moved online than digital video minutes, which Ek sees as evidence of significant future innovation in audio formats.
“audio is in even more nascent formats than video is. There are even fewer minutes that have moved online”
Automating discovery could lift how often people dive into podcast rabbit holes by 10 to 20 times rather than doubling it, and the second-order effects of that scale are unknown.
“It's not like you're 2 X the discovery rate. You're looking at a 10 or 20 X”
Most of the world's attention goes to bits (digital companies), but almost all of the largest societal problems have a significant atoms component that requires real scientific breakthroughs, so the next generation of entrepreneurs should focus there.
“almost all of them are atoms problems or at least have a significant component, which is real atoms and not just entirely done by bits”
During the pandemic creators moved beyond ad-supported income to models such as Patreon, live audio rooms, and paid memberships, and successful creators build virtuous loops where media audiences feed other businesses.
“Up until very recently, ad supported was the primary mean through how many of these platforms were able to sustain themselves.”
Atoms companies today face an opaque playbook similar to early digital companies: it is unclear who to approach for capital, what product-market fit means, and how to structure the company, since standardized series A, B, and C norms do not yet exist for them.
“the whole playbook on how to do them is completely different, and we're back to when I started digital companies”
Solving a project like green steel, which may cost over $100 billion, requires coordinating bold ambition, the best talent in the world, and cheap, long-term capital, which is hard given competing opportunities.
“how do you create a coordination of a bold enough ambition, the best possible talent in the world, and a cheap enough and long-term enough capital source”
Ek learns new domains by traveling to meet people and spending the time needed to reach real insights, accepting a long period of being the dumbest person in the room, which human psychology resists.
“you will go from pretty much being the dumbest person in the room for a very, very, very long time”
Founder-led businesses create crystallized focus because a founder typically owns, sits on the board, and leads management and product toward one mission, but founders are often impatient and prone to hubris, so jumping into problems prematurely is a risk.
“what it often brings is this clarity of purpose with a founder, because not only did this person start the company, they're typically a big owner in the business”
Spotify's strategy is to return power to creators and enable each to choose its own business model, because the one-size-fits-all platform approach was a necessity for getting early users but is no longer ideal as the audience is global and models like $1,000 products can also work.
“I think the Spotify specific strategies, we want to enable more power back to the creator”
Healthcare reimbursement is based on hospital visits and billable actions, so the payment system does not align with preventive care and must be redesigned to fund proactive treatment.
“the whole system is about being reimbursed when, based on hospital visits, based on when someone actually does something?”
Strong companies build a decision culture that teaches people how to make decisions, allowing mistakes without risking the whole business, rather than routing everything through one central authority.
“some of the greatest cultures that we have today are ones that have created the great decision culture”
The creator base on Spotify has grown to about 8 million creators from about 3 million two years earlier, and it could reach around 50 million in the next few years, possibly five times more.
“Today, we're about 8 million creators. I believe when you and I did this, we were about three.”
Blockchain is early but could enable remix culture, global experimentation with music, and control for creators and labels, though many stakeholders in the value chain make it hard to design a system that works for everyone.
“there you have now technology like blockchain and other things, which are relatively early in its sort of maturity curve, but I think could play a very important role”
First-principles reasoning was a common concept in physics long before it became popular among investors and entrepreneurs, so Elon Musk did not invent it, and leaders should learn several frameworks per year across disciplines.
“it wasn't Elon Musk that invented it, it was a very common thing in the field and discipline of physics”
The macro tailwind of digitalization and globalization may have another decade or two of runway, since digital activity is still a small share of the overall economy.
“maybe we're looking at another decade or two where that has some more profound runway left in it”
Early in Spotify Ek fought against discarding data, a decision that seemed unnecessary at the time but later became standard because more data enables inference about complex systems.
“I fought pretty hard for us not throwing away any data, which today seems like a pretty ordinary decision”
Music is now part of most internet experiences as creators use songs as a remix-like expression on platforms such as TikTok, but copyright and licensing patchwork prevents clean sharing of those videos across platforms.
“If I would take one of those videos and try to share it on another platform, in many cases, you just can't for copyright reasons”
Spotify's platform has hosted the Mexican president's weekly national address as a podcast-style publication, illustrating how the platform now faces every type of content, including political content.
“we have the president of Mexico that now every week started publishing their national address on Spotify”
Climate discourse divides into people who think society must radically change lifestyles and techno-optimists who think innovation alone will solve it, and the truth is probably somewhere in between.
“The reality is not very surprisingly that the truth is probably somewhere in between.”
Healthcare is the largest employer in the US, and if its growth continues it is projected to absorb the entire working population, which is unsustainable.
“Healthcare, if I'm not mistaking it, in the US is the number one employer of people, and it's projected to overtake the entire working population”
A mission-driven company's value equals the sum of the problems it solves, and there remain many improvements to make day by day, which grounds Ek's continued building.
“the value of what you're building is the sum of all the problems that you solve”
Podcasts mostly fall on the nutritious side of Sheer's nutritious-versus-delicious framework, offering deep expert knowledge and rabbit-hole learning, which may expose a large new audience to subjects it never encountered before.
“podcast is one of those things that mostly errs on the side of being more nutritious than perhaps being delicious”
Graham Duncan's concept of working with source describes a flame carried by the founder who takes the first risk to create something new, and Patrick asks how the founder can spread this so they are not a bottleneck.
“There's this interesting concept our friend Graham Duncan calls working with source and source is described as carried by the founder that takes the first risk”
The healthcare system does not currently contain much tangible, usable data, so systems must be built to collect it, process it, and cross-reference it with symptoms before meaningful predictions can be made.
“just don't exist that much data in the healthcare systems that's real tangible data. But we have to build the systems to collect that data”
Patrick connects Ek's variable-reduction method to Frank Slootman's Snowflake principle of narrowing focus to increase quality, reducing variants toward a near-binary choice to make better decisions and outcomes.
“It reminds me of Frank Slootman, that Snowflake's concept of narrow the focus, increase the quality.”
Podcasting is bootstrapping a modern library of Alexandria by making deep, siloed expert knowledge accessible, a form of open source for knowledge that accelerates learning and building on prior knowledge.
“podcasting is bootstrapping the world's deep knowledge, not just surface level knowledge”
Patrick argues that the future is about running cheap experiments across the spectrum of business models, with Spotify pushing value to creators, even for free, to grow the ecosystem rather than maximize its take rate.
“what Spotify's focus is just to push more, even for free, down onto the creators and grow the ecosystem versus grow the take rate”
News was once thought impossible to monetize, but paid models work, and Ek expects further journalism business models to emerge even though current paywalls are frustrating to consumers.
“The common view a few years ago is that news had to be free. It was impossible to monetize”
Ek admits that he did not fully understand what he meant when he said scale had been redefined, and the speed of growth in digital companies has exceeded his expectations.
“I think even I didn't understand what I meant. I am looking at the world now”
Patrick's five-year mantra is learn, build, share, repeat, with building being hardest because teaching and making decisions are where difficulties surface, and decision-making is the core skill of building.
“learn, build, share, repeat... the most important in that because it's the hardest is the build concept”