Stripe and Solid-State Economics
Cars, Excel spreadsheets, vacuum tube-based computers, poorly-implemented recursive programs, and attempts to win at real-time strategy games all break for approximately the same reason: they have lots of moving pieces, and the more moving pieces there are, the easier it is for something to stop working. Over the last half-century or more, across a variety of physical products, mechanical products have been replaced by more efficient, more compact, and much more reliable solid-state ones.
This is a literal change in the physical world; the computer I'm using now has an SSD rather than an HDD; reading data off the disk is zippier than it used to be. Tesla talks about how their drivetrain has about a tenth of the moving parts of a typical ICE drivetrain (17 compared to around 200), which means that fewer things break and they're easier to fix (and that managing inventory is easier, too). But this trend is also a useful analogy, and it gets more powerful in software. Solid-state is more of a qualitative trait for software, but it applies when a system is:
- Useful for some real-world function;
- Deterministic, or, since that's impossible, graceful at handling the occasional errors that do crop up; and
- Composable, such that one system can take inputs from another and pass them on to a third.
This is how software eats the world: by taking business functions one at a time, turning them into well-documented API calls with useful error messages (but infrequent errors) that can be chained together with arbitrary complexity and then run with minimal human involvement. The long-term dream is that, when you mouseover the "order now" button on an e-commerce site, a seamless chain of events updates the entire supply chain, and this tiny indication of incremental demand actually affects inventory restocking, hiring, and financial planning all the way down the line. We're not there yet, but we're moving in that direction. And Stripe is supplying the crucial payments layer, and a growing set of adjacent products, to make it happen.
Stripe is part of an interesting category of value-creating companies whose offering is to make some process work the way you'd imagine it worked if you had never actually tried to do it yourself. Early in my career, when I was analyzing data mostly in Excel, I operated on a general assumption that if a business collected data from multiple sources, it would be fairly simple to filter from data source A based on criteria from source B, pipe the output into C, and have this summarized and updated in source D. It's never this easy, but part of the value prop of Snowflake ($) is that it should be. Cloudflare is another example. As their line goes, "The Internet was never actually designed to do what we have asked it to do."
In payments, you might naïvely imagine that if you are located in, say, Texas, and you wish to sell goods or services to someone located in, for example, India or the Netherlands or Japan, that you could throw up a quick payment form on your site and then get some money from them. As it turns out, though, there really isn't "a" global payments system. There are countries, they each have multiple payments systems, some of those systems overlap in some ways, and participating in those systems requires some combination of government approval, banking approval, technical overhead, and ongoing compliance and maintenance costs. Your German customer may prefer to pay through Giropay-enabled bank transfer, while your American customer is optimizing rewards points and plans to use a credit card, and your Japanese customer, naturally, would like to settle their bill at a convenience store. Also, sometimes the rules change. The dream is to transfer value electronically, but it's hard to achieve this in a seamless way. Payments systems long predate electronic communications, and, even worse, they computerized before other industries. In many banks, the core system was written by someone who a) only knew COBOL, and b) has been dead for several decades.
Stripe is an effort to build a seamless layer on top of these systems. This means two kinds of sustained effort:
Designing an API that handles a variety of different nuances, expectations, exceptional cases, widely-believed falsehoods, etc. Just about the only constants are that there are units of currency transmitted from (at least) one party to (at least one) other party, at some time, UTC.1
They need to design this API so that, as much as possible, it never changes. The economy is not solid-state when some components get end-of-life announcements, and nobody knows if somewhere deep inside their software stack, or the stack of one of the companies they interface with, is an API that's going to stop functioning. This forces a lot of creativity: what payments systems will arise later? What entities will want to transact in the future? What kinds of transactions might get banned in the future, and should those bans be implemented at the transaction level or somewhere else?
Stripe has long prided itself on quickly approving new users (in one early interview with them, the interviewer notes that it took a lot longer to get approved by their email provider than Stripe). And the actual process of integrating Stripe is also quite straightforward. But that encapsulates a lot of backend effort, edge case analysis, growth-vs-fraud tradeoffs, and integrating across systems that were never designed with integration in mind. As another developer once said of a similarly involved but otherwise very different project: "You know, a lot of effort went into making this effortless."
Stripe says "Our mission is to increase the GDP of the internet." This is one of those goals that is a bad idea to assign to most people, but a good idea to assign to high-agency people, since there's so much it could mean. What comports with Stripe's missions statement:
- Finding businesses that exist online and helping them grow faster
- Identifying businesses that could exist online, and giving them tools to grow revenue. (As they note later on their about page: "Stripe powers the kinds of companies that couldn’t exist even ten years ago")
- Accelerating the shift from offline to online
- Working to blur the online/offline distinction
The focus on Internet transactions superficially means writing off about 97% of global economic activity. For a payments company that monetizes based on transaction volume, beginning the process by giving up most of the TAM sounds like a strategic misstep. But there's another way to look at this. From Neal Stephenson's recent novel Fall, set in the near future:
In old movies sometimes you could see apparently sophisticated characters saying things like "I'm going online" or "I'm surfing the Internet,' which must have seemed cool at the time, but now it was a non sequitur, as if someone, in the middle of an otherwise normal conversation, suddenly announced "I'm breathing air."
If you assume that every transaction eventually gets mediated through the Internet, and that the process accelerates as the tooling gets better, then Stripe's real goal is to grow global GDP, and the intermediate short-term goal is to increase the share of it happening online by a modest 30x.
Targeting something fairly abstract, but still measurable, gives the company plenty of flexibility to decide what to launch next. And since their payment model is to take a share of transactions—sticker price of $0.30 + 2.9% for US, €0.25 + 1.4% for cards in Europe—they're pursuing a mission of growing the Internet economy, and monetizing it by operating an efficient tollbooth for access to that economy.
Warren Buffett has sung the praises of tollbooth economics for a long time. (Long enough that, as far back as the 1970s, lawyers in an antitrust case against him were asking him precisely what he liked about tollbooths.) Buffett owned ad agency stocks in the 70s, for example, because they generally took a straight 15% cut of client spend. CPI +10% meant TV ad spots were 10% more expensive, which meant, like magic, that Ogilvy & Mather grew 10%. It was a built-in inflation hedge and a way to ride along with economic growth in a capital-light way. Buffett knows what he's talking about here. Before he was a big investor in ad agencies, he owned about a quarter of Detroit International Bridge Co., a company that owned a literal toll road between the US and Canada.
The toll booth comparison is apt, because toll booths are a way to monetize infrastructure, and Stripe describes itself as infrastructure: building roads, not cars. But existing payments systems often get described as "rails": they're infrastructure, too! Is it redundant to build infrastructure on top of infrastructure? Not necessarily. First, different kinds of infrastructure are often complementary. When trucks were introduced, they might have appeared to compete with trains. Not so: freight volumes went up as trucks got more common, because trucks made it easier to move goods from trains to warehouses and stores. Cargo planes replace ships for some tasks, especially moving goods with a high price/weight ratio or a short expiration date (or, in the case of chips and smartphones, a bit of both). But since fuel cost per mile rises with speed, this actually means ships can slow down, making them more cost-effective for products that don't need to move in a hurry.
And there's a lot of value to be created in building meta-infrastructure that improves compatibility across existing systems. When the same box can be pulled off a ship and placed directly on a train or truck without being unpacked, the entire system gets more efficient. It also gets more information-sensitive, though: if you're going to pack up a box in Shenzhen and not open it until it arrives at a Walmart in Dubuque, you need to be extra sure that it contains exactly what it was supposed to, because nobody is going to open it up and check along the way. And if you're managing inventory at that store, you need to know well in advance what you'll need, because the most efficient trip it can take is a long but linear one; the cost of a cheap high-scale system is that changes are murderously expensive.
Infrastructure is not cheap, and, unlike more amorphous economic products, it has to actually be in specific places. A "bridge to nowhere" shows up as infrastructure on the balance sheet, but it doesn't make anyone better off. In Stripe's case, the analogy applies to where people work: a company can't solve payment problems for a given set of customers without spending time with them, and this has encouraged Stripe to open offices around the world (from the Bay Area to Singapore to Mexico to France). In a nicely-timed move, they announced in mid-2019 that their fifth engineering hub would be remote. Part of moving the world online means recognizing which parts of it happen offline, and the more deeply embedded they are in offline norms, the less visible they are online. Buy all the metadata you want, it will be hard to detect Brazil's cash-on-delivery boats to remote villages along the Amazon ($).
Since payments a) have strong network effects within a currency, but b) have much weaker network effects between currencies, there are a variety of payment methods around the world. And they don't just have cosmetic differences—they have significant variance in pricing, timing, and underlying economic model. Payment can be loosely-integrated with the rest of the business; get a merchant account and you're set. Or it can be a core competency; Grab had to render a significant population un-unbanked in order to offer ride-sharing services. In some places, payments are tied to customer expectations—for German shoppers, payment can happen after goods are received, and only once the buyer expects not to return them. Restrictions on interchange fees mean that European cards don't produce as much revenue for issuers, so they can't create the US's wild ecosystem of rewards points, tiers, special offers, arbitrages involving free shipping for 2.4 million one-dollar coins, etc. While there's variance in cost, it's not necessarily driven by market power; US consumers like credit cards (because of rewards, and because of the "float" from getting goods now and only paying a month later), while consumers in higher-savings countries prefer debit.
All this variety puts Stripe in an interesting position in the office-or-remote debate: while they don't necessarily need people in offices, they do benefit from having people in particular places. But this also leads to an interesting end state for the company: for Stripe to provide the world's meta-infrastructure for payments, and to build the solid-state economy, it will end up having people everywhere. There are some global companies that globalize in search of cheap labor, or globalize their selling but not their building; Stripe is implicitly long global talent, because everywhere that people buy and sell creates local knowledge that Stripe can turn into global value.
Bill Gates has a justly famous definition of a platform: "A platform is when the economic value of everybody that uses it, exceeds the value of the company that creates it." When I wrote about AWS last week, I argued that AWS' value stems from who it grows with. It's the second derivative of Gates' definition: a business or product gets more platformy if, as users grow, the share of the platform's value creation they capture also rises. Stripe falls into the second-derivative category, generally creating an expanding amount of value for its users. While the incremental costs are fairly fixed (with the exception of enterprise deals, whose economics tend not to get disclosed), most of the businesses that use Stripe deal with some form of fixed costs, so their incremental margins expand when they sell more.
Stripe's mission is to grow the GDP of the Internet. The means is to provide better payments, and payment-adjacent services, like fraud detection, intelligent card updates, dispute updates, adaptive acceptance (using machine learning to find the best approach to retrying cards for recurring payments), working capital loans, make payouts for marketplaces, and more.
But this is not without tradeoffs. Some features are unambiguous wins for all parties involved, like anything that makes an intentional payment more likely to go through. And some can shift value from one side to another. Stripe's customers are the merchants who use it, not the customers who buy from those merchants, even though Stripe's function is to intermediate between the two. Other payment platforms have struggled to balance competing interests. A credit card company, for example, has to deal with merchants, cardholders, merchant acquirers, card issuers, and rewards partners, and sometimes the party with the weaker hand is the one that gets some economic value distributed away from it.
The history of payments is full of tradeoffs between three trends:
- Universal acceptance for merchants, customers, and financial intermediaries.
- Lower latency between when someone starts a transaction and when they can stop thinking about it.
- Real-world implementation of the "consumption smoothing" theory that economic actors use saving and borrowing to spend money at a steady pace rather than being constrained by their cash on hand.
These always face tradeoffs, and new technologies tend to push the frontier of them forward. The early history of the credit card industry is full case studies in this. For example: before card authorization was automated, it could take several minutes to confirm a transaction, so many merchants wouldn't bother to verify that a card was good below a certain transaction threshold. After automation cut the approval time to under a minute, those floors were lowered—reducing latency indirectly led to reduced fraud, and lower fraud risk meant that more banks were willing to sign up cardholders, which made card use more universal.
Stripe actually has a way to avoid the tradeoff-heavy mature state, or at least to delay it for a long time. Patrick Collison points to Peter Kauffman's observation that the best investments have good returns, low risk, and long duration. One can think of redistributive decisions as trading a little less duration for a bit more returns; if one party is getting the worst deal it can stand, then as soon as circumstances change that party will be the first to defect. Whereas an infrastructure company has structurally high duration. Promoting positive-sum interactions actually increases duration and lowers risk, and if those interactions involve making incremental payments, they earn Stripe another 30 cents + 2.9% or so. That pattern is self-reinforcing, because a sufficiently popular API is a de facto standard, and the longer-term the outlook, the more of a standard it is.
Every sufficiently successful company eventually reaches the point where it can only sustain growth by raising its rate of value capture rather than value creation. Some companies have a long runway for doing this, but for many it's surprisingly short; successor executives tend to operate more conventionally than founders, and it's conventional to let the quality of a business degrade a bit in order for a high growth rate not to degrade at all.2 Stripe has a long time to think about this problem, and to inculcate a culture that will push back against it. But it's a ubiquitous form of corporate entropy. Jeff Bezos recently quoted Richard Dawkins on this:
Staving off death is a thing that you have to work at. Left to itself – and that is what it is when it dies – the body tends to revert to a state of equilibrium with its environment. If you measure some quantity such as the temperature, the acidity, the water content or the electrical potential in a living body, you will typically find that it is markedly different from the corresponding measure in the surroundings. Our bodies, for instance, are usually hotter than our surroundings, and in cold climates they have to work hard to maintain the differential. When we die the work stops, the temperature differential starts to disappear, and we end up the same temperature as our surroundings.
Corporate life is also a struggle against corporate entropy. But while companies typically succumb, there are other human institutions that have remained vibrant and true to their roots a lot longer. Harvard, for example, has been a respected educational institution through multiple generations, although it's had to adapt as respectability got more merit-based and less centered on providing a smooth transition from elite prep schools to the workplace. Some religions keep traditions alive for an extraordinarily long time, and there seems to be a positive correlation between how distinctive those traditions are and how durable the community is. (There may be selection bias, here; look back far enough, and everybody's traditions were weird.) So there are examples of long-lived institutions, but they're hard to copy.
Hiring People Who Will Battle for Basis Points
Even if an institution lasts longer than the lifetimes of the people who build it, it is, at any given moment, still made out of people. And Stripe spends a lot of time on recruiting. A Forbes profile from 2016 says:
They spent months agonizing over their first ten hires, knowing those would in turn attract the next wave of employees. They reserved 10% of Stripe's equity for this first cadre--an unusually large amount--and worked alongside many of them for a week or more before giving them the nod.
And in a more recent interview, Patrick Collison says:
[W]hen we hired our COO—she was previously a senior leader in Google’s sales organization—people were worried that she would change the culture. 1 We had to be explicit and clear about the fact that she would. That was the job. And I think the changes she has helped bring have been very healthy and beneficial for the company.
Hiring is not just a way to maintain a culture, but a way to help it grow in the right direction. There are other levers:
- Stripe Press (a little bit more on them below) produces a great selection of important books. It's hard to make a great return in the publishing business, but it is possible to create positive associations—if the most important book you've read this year and the one that looks nicest on your bookshelf come from the same organization, you'll remember it. One book Stripe Press republished, The Dream Machine, tells the story of the Internet from before the beginning, and shows just how important it was to have the right people working together on the right project at the right time. (Stripe used to give copies to all new employees; they can now choose from a selection of books, but Stripe keeps plenty of Dream Machine copies around the office.)
- Good design is a recruiting and motivational tool. It's a recruiting tool because working at a company that values clean design means everyone has a good impression of your work, and since most people don't get past a first impression, this has an outsized impact on external social status. Outside of tech, and more recently the investing community, it's easy to forget that Stripe is a fairly obscure company, in part because it's doing its work behind the scenes. The tradeoff is that requiring universally good design means embracing perfectionism. If the quality bar is high, it takes longer to get there, and it's a more uncertain process. First drafts take just one attempt, but you never know how many tries you'll need to get a good final copy.
Previous payment companies have also faced complicated constraints, some of which were the abstract sort that an MBA class might cover, and some of which were a bit more down-to-earth. Consider the counter. Counterspace was a major limit for early card systems. A clunky point-of-sale system or a bulky product for producing machine-readable receipts might take up more space than a merchant was willing to sacrifice. The modern equivalent of this is pixel-space on the checkout page. Once a customer has decided to make a purchase and is at the point of actually entering their payment details and completing the transaction, it's a high-stakes interaction: any improvement in conversion rate on this page drops directly to the bottom line, which means every pixel needs to be justified.
If displaying a logo like "Paypal" increases transaction volume, it stays; omitting a logo needs to be an option. Adding choices can raise the conversion rate if they're meaningful (so paying once upfront or paying over time through Affirm is a conversion-increasing option), but adding a large number of meaningless choices lowers it. Stripe saves this pixel-counterspace in a few ways, by doing simple things like automatically detecting the card type (why choose "Visa" or "MasterCard" from a drop-down when the digits of the card number literally encode this information?!) to minimizing the amount of personal data collected in order to verify a cardholder's identity (why choose a state from a dropdown, when the digits in the zip code literally encode...!?)
- The company uses referrals as a major recruiting tool, though not an exclusive one. (Relying entirely on referrals means that you're stuck in whatever bubble you started in. And, since it's a bubble, you won't notice.) Recruiting through referrals makes a lot of sense for a company that was founded by Y Combinator alumni who dropped out of Harvard and MIT; if you were trying to speedrun networking, that's what you would do. But, as it turns out, this scales: new hires know new people who might also be good new hires. And referrals are particularly effective for people whose résumés undersell them, or who undersell themselves in interviews. A short summary of why hiring is hard is that you know a lot more about someone after spending a few years with them than you learn with a few hours of interviews, and referral-based hiring taps into the fact that that knowledge exists somewhere.
- Stripe's culture seems to lean more towards writing and reading than other companies'. Stripe Press is a hint about that; it's hard to imagine getting too excited about the latest book by, say, American Express Press. Focusing on reading and writing is a complement to the relative deemphasis on interview skills; some people express their thoughts best in long-form, with editing, than they do extemporaneously in a conference room or on a Zoom call. In recruiting terms, this one is more of a tradeoff: making a company much more appealing to some people while making it a relatively worse fit for people who do their thinking out loud.
One benefit of a writing-centric, perfectionist culture is that it encourages honesty. It's easy to contradict yourself if you talk fast, or to be vague about topics that make you uncomfortable. But if your internal memo goes through five rounds of "Now, what do you mean by that?" you're probably going to have to make yourself clear. Some companies highlight values rather than tradeoffs, but values only make sense as tradeoffs. And writing-centricity doesn't mean valuing artistry over substance. It means being clear and direct about the substance. I asked Patrick about this, and his response (in part) was:
Clear thinkers -- i.e. those with good models -- are, as a result, usually good writers. We don’t care about the curlicues of someone’s language and poetic turns of phrase. We care about effective reasoning, clear and specific prose, and general avoidance of corporate euphemism, passive exonerative, and all of the other mumbo jumbo that people so often subconsciously think they’re supposed to use. “Signups are going down because we don’t understand how to target likely buyers in Canada” is a lot better than “We’re seeing softness in our acquisition metrics.” As it happens, I find that a lot of engineers -- people for whom “MFA” is “multi-factor authentication” -- are truly excellent writers.
(No, this was not an extemporaneous verbal reply.)
The paradox of having an explicit, well-documented, tradeoffs-embracing company culture is that it can mean new employees buy into the culture more than the old ones do, which makes the business feel cult-like. This is not necessarily a problem (I talked about cults as a good pattern for businesses to pursue in this Palantir writeup). This is probably true of any kind of tradition or cultural formation: we invent compact narratives to explain the recent past, and those narratives necessarily strip away some details. People who try to achieve compliance with the narrative rather than the history that led to that narrative will embody a somewhat caricatured version of what they're copying. But as long as the caricature captures the important details, it's probably better to have everyone following the same exaggerated norms than to let the culture melt away.3
Stripe has not exactly made a secret of valuing hard work over work-life balance. Some of their early job ads and interviews talk about the "Sunday test":
If this person were alone in the office on a Sunday, would that make you more likely to come in and want to work with them? If the answer is not a clear yes, then don't make the hire.
This is a pretty strict standard! Among other things, it presupposes that rolling into work on a random Sunday is normal, and imagines a scenario where someone in the office on Sunday would ping a coworker, who is, naturally, checking their work email or Slack. Other organizations have a related Sunday test ("Would you regularly work on Saturdays and/or Sundays if we paid you a lot of money?"), so it's not quite as blunt. But still, it's a way to indicate that the default expectation is that you'll like your work enough to do a lot of it, and like your coworkers enough to see a lot of them.
There are some organizations that like balance, but many of the most effective lean towards all-out effort, even at the risk of burnout.4 If the output of your work is a tiny, simple tip on top of a vast iceberg of complexity, you need all the institutional knowledge you can get. Some of it will be preserved in writing, some of it will be folklore, and some of it will be pure metis: the process knowledge you develop but can't express except through your work. Working 80 hours a week, as workaholics like to point out, means you can get a decade of experience in five years. (This may require two decades worth of coffee.) And if a business is changing fast enough, that means the hardest workers are the most experienced people on earth at their chosen task.
And Stripe has one more recruiting lever: impact. Consider the Twilio case study. Twilio increased its conversion rate by 10% by using Stripe. This was a combination of several smaller lifts; 5.5% here, 1% there, it adds up. As of their latest earnings report, Twilio's net dollar retention rate was 137%, so working with Stripe allowed them to make 10% more investments that each compound at an average rate of about 37% annualized. This is a material contribution to Twilio's enterprise value—it made a lot of Twilio employees and shareholders significantly better-off. Postmates added $70m in revenue, and saved $3m in authorization fees. Postmates isn't public, but its parent company trades at 8x sales, so that's about $560m in incremental enterprise value.
Since the payments space is complicated, and everyone's stack and approach will be different, you can't know in advance how much Stripe would benefit a big company until the models start spitting out results and the A/B tests hit statistical significance. And those stories are selected because they're impressive. But still, it gives Stripe some of the appeal of consulting and banking: solve big problems with dollars attached, for all sorts of companies! Bankers collect deal toys, Stripes collect case studies. (Yet another benefit of a writing-centric instead of offices-and-seating-charts-centric culture: pixels have a lower marginal cost than lucite, but to a discerning viewer they're more impressive.)
A Controlled Burn Approach to Hype
One of the rituals people go through whenever a tech company hits a valuation milestone is to go back and look at the first thread about the company on Hacker News. Dropbox, as we all know, could be built in a weekend—doesn't everyone know how rsync works? Since Coinbase was looking for a cofounder when Bitcoin was down, it was obviously an attempt to pump the price back up to $25 (of course, if Bitcoin had been at a record high then, Coinbase would have been a cynical attempt to raise venture funding, since $25 per Bitcoin was obviously unsustainable).
Sadly, we don't get to do that with the Hacker News thread about Stripe's launch. Instead, there are comments like:
Stripe pretty much takes payment processing kicking and screaming into 2011.
Stripe is a game changer.
That was my first 'aha! OMG...this API is awesome' moment.
Stripe recently saved me.
I wish them luck, and can't wait to use it.
I've been using Stripe since early June and can say that it is the best payment platform to develop on, period. It's just lightyears ahead in terms of development time, PCI compliance issues, and overhead. Not only that, but the guys at Stripe are extremely responsive to customer service questions and issues.
Best API ever for a payments processor - period.
We've been using Stripe for Picplum.com since the middle of the summer. These guys are insane -- they saw an error before we even did and emailed us saying they were working on fixing it.
This is an incredible product built by a team of amazing developers, solving a serious problem.
Is this real? Am I dreaming? This sounds too good to be true. Someone pinch me, please.
Stripe is simply amazing.
There was really only one complaint: that the service was US-only. A Stripe employee promised that they'd expand to other markets soon, leading to this:
And this is how it's been from the beginning. When Stripe was in beta, it was a cool beta, and people recommended it to their friends. Its early rounds look low in retrospect, but a rumored valuation in the hundreds of millions in 2012, when they had 28 employees was justifiably described as "white-hot." For comparison, Instagram had raised money at a similar valuation a few months before, just before getting bought out by Facebook, and it was priced based on its ability to supplant Facebook, which was then the hottest tech company in the universe. Stripe was able to hit the same kind of valuation milestone without a glamorous comp.
Hype is a blessing if you need to hire people, raise money, or close deals. Sometimes it's a synthesis of those: early checks into Stripe came from Peter Thiel and Elon Musk, who had each co-founded the companies that merged to create Paypal. Getting capital from them is a good way to get introductions to the right financial institutions (and to get a heads-up about which ones are not worth talking to).
The trouble with hype is that it has to keep growing. And if it diverges from reality, going from hyped to accurately assessed is a painful letdown. Right now, Stripe's revenue is a reported $7.4bn, growing 70% last year ($, WSJ). Its last reported valuation was $95bn. So it's trading at 13x sales, and this is a business with some serious ongoing costs. Those variable prices for different payment methods reflect the various prices that underlying payments systems charge. Stripe is in the lucky position to be the default choice for a large number of ambitious engineers, product managers, designers, and salespeople, but it's also valued in a way that fully bakes in many years of flawless execution.
The company could be splashier. It could market itself more aggressively. It could focus less on books and more on vivid ad campaigns in airports. But since Stripe's product was originally designed to make payments easier for developers, it's leaned in to marketing that appeals to people who a) write lots of code, and b) want to solve business problems cost-effectively. API docs are a necessity, but the cost of producing Stripe's API docs should arguably be booked as sales & marketing expense: they are a long form sales letter, with some inline source code for illustration, making a pitch to the target audience.
Hype can be engineered, temporarily, but it eventually fades. In the meantime, it can be channeled, and aiming it at long-duration business assets like developer trust is a useful way to keep it going.
Response Code 402
Here is a picture of the Internet, as it's experienced every day:
18 words of content on a 6.1-inch display! Two ads, and a banner—about ads. It wasn't supposed to be like hiking through a jungle during an earthquake while being stalked by hundreds of hungry info-parasites. This makes Stripe a form of retro-futurism, trying to make the Internet a bit more like what it was supposed to be. "We wanted human-computer symbiosis, instead we got 140 tracking pixels on every page."
The Internet is good at transmitting every kind of information from anyone to anyone else, except money, and money is the high-order bit. Since information can be so readily converted into money, much of the Internet is designed around a) gathering lots of data about people, and b) using it to convince them to buy things. But this creates a race to the bottom. If charging for money isn't a first-class function of the standard protocols everyone uses, anyone who assumes it's a default is failing to profit from some of their users. Whereas ads, since they target everyone, don't fail to monetize anyone.
There have always been businesses that used ads, and there are good businesses that rely on them. But their status as the default way to create revenue online has created an equilibrium that no one is happy with. Writers and publishers don't like the way their sites look, and don't like opting into an equilibrium where they're lead-gen for data-ingesting ad companies that capture so much of the economic upside. And because of the compounding benefits of more and more data, the ad ecosystem tends towards centralization in some high-value layers, and towards messy and redundant decentralization everywhere else.
When Stripe was founded, it wasn't "Stripe," it was /dev/payments. Payments were supposed to be a system resource that programmers could easily add to a script. This turned out to be complicated, for all sorts of reasons (it turns out that Delaware corporations can't be named with a leading slash, for one thing). But it was the right idea. Some software utilities are so generally useful that they get reinvented again and again, and eventually standardized into powerful, composable tools. A world where money is accessible through just another function call is a world where more of the economy fits together better and runs a lot faster.
Disclosure: I have friends who work at Stripe. I have friends who used to work at Stripe. I have friends who are almost certainly going to end up working at Stripe. Stripe processes the transactions when readers pay for The Diff. I'm pleased to announce that Stripe Press will be publishing the book Tobias Huber and I are working on. In addition to expanding the GDP of the Internet, Stripe has done a great job expanding the GDP of me.
Thanks to Patrick Collison for answering my Stripe-related questions, and for joining today's subscriber call.
Some sources I particularly recommend:
- Stratechery: Stripe: Platform of Platforms
- Not Boring: APIs All the Way Down
- Colossus' interview with John Collison
- Elad Gil's interview with Patrick Collison
- Patio11's four years at Stripe
Hulu's Regulatory Arbitrage
Usually, "regulatory arbitrage" means locating something legally in one jurisdiction in order to derive economic benefits somewhere else, or finding a new legal form with the same underlying economics as something that's banned, taxed, or otherwise restricted. It also applies to platforms, many of which perform government-like services (setting basic rules, enforcing contracts) with a government-like revenue model (collecting a 30% income tax). A few years ago, Hulu discovered that an expanded API they had access to allowed them to unsubscribe people through the app store and resubscribe them directly, avoiding the 30% Apple Tax. Apple found out about it, after a tweet, and closed the loophole.
This is very close to how typical aggressive tax strategies get banned. Someone discovers the strategy and quietly uses it to defer or avoid taxes. Eventually, they get good enough at this that their success is conspicuous. Someone writes a news story, and suddenly—the tax strategy is gone.
After the collapse of Archegos, banks are reducing how much they lend to hedge funds ($, WSJ). Some of this is prudent; as has been widely reported by now, Credit Suisse, which lost $5.4bn in the Archegos collapse, earned just $17.5m for their trouble ($, FT). I've suspected that there will be two big effects from Archegos:
- What's happening now, which is that banks reassess their risk and cut back hedge funds' leverage, and
- What will happen over the next year or two: banks will figure out how to avoid this specific scenario (a fund that is a) highly levered, and b) owns a substantial portion of the economic value of a few companies, through swaps) and will fix those bugs and get back to offering similar leverage. Meanwhile, hedge fund managers who read the story will realize that they could run a lot more leverage personally than they can through managing outside money, and will decide that the ultra-levered family office model is the right one for them.
Raising outside money is, from the manager's perspective, already a form of leverage. Instead of deriving gains from your own net worth, you can also earn fees from gains on somebody else's. But it's not the only source, and it has two important costs: it takes a lot of time to raise and keep outside capital, and for capacity-constrained strategies it caps the ultimate upside.
The US has endorsed waiving patent protections for Covid-19 vaccines. This is an interesting illustration of the difference between aggregate analysis and more detailed models. There are always obstacles to achieving some level of production for a given good, and in theory every obstacle removed should increase supply. But for these vaccines, IP protection does not seem to be the biggest barrier. This is a brand-new supply chain, built in the last year, and scaled up to unprecedented levels; everything we know about scaling mRNA vaccine production has been learned since the start of 2020. And scaling is hard, often in surprising ways! Moderna's CEO says ($, FT):
There is no idle mRNA manufacturing capacity in the world. This is a new technology, you cannot go hire people who know how to make mRNA — those people don’t exist... If you were to start today, you’re going to have to start by hiring people. Those vaccines don’t fall from the sky... There is no mRNA industry... When we hire people that come from traditional pharma, we have to train them in the art of mRNA.
This doesn't mean that waiving patent protection is completely ineffective. Other manufacturers can start to produce, and perhaps work out the same production bottlenecks at the same pace. So, in a few years—when it's not urgent—we'll have more supply. But the longer-term effect is on incentives. There are lots of dedicated, public-serving people who work in medical research and who will want to tackle pandemics no matter what. But there are also shareholders, boards of directors, and CEOs who are going to run the numbers and decide just how flat-out a company ought to run towards the goal of creating necessary novel drugs.
One of the questions I've thought about a lot in recent months is whether the pandemic made the world more resistant to existential risk. In some ways, the answer is yes; we're all going to pay a lot more attention to news stories about mysterious disease outbreaks! But a big part of solving existential problems is mobilizing resources, and a lot of those resources are currently controlled by companies that think very hard about their earnings per share. The smart thing to do, if we're already spending trillions, is to spend a bit more to ensure that next time, big companies know that the profitable thing to do is also the right thing to do. Alex Tabarrok is right: Patents are not the Problem!
Emissions: a Global Problem
China's greenhouse gas emissions are now higher than aggregate emissions for the OECD. Reducing emissions is partly a financial engineering problem. Rich countries can afford to pursue climate mitigation strategies, whether they're renewable energy sources, more efficient buildings and vehicles, or geoengineering. But the only backtested model for becoming a rich country involves burning lots of fossil fuels along the way. So one model for human civilization is that it has its own "burn rate." Importantly, this pattern has shown up before: escaping Malthusian traps, depleting food supplies or topsoil, avoiding institutional decay—all of these are existential problems that tend to get worse over time but that often get solved. Which is not an argument for passivity. On existential problems, the optimists always turn out to be right because optimists and pessimists only exist in a world where the risk doesn't pan out. In the meantime, there's a complicated negotiation: the rich world doesn't want the consequences of China (and India, and Africa, and South America) getting rich, but those parts of the world are understandably resistant to the argument that they ought to accept an even lower relative standard of living.
When Visa started batch-processing transactions in the 70s, the process started at 5pm Pacific Time. This was not because they expected to have some downtime in the evening; it was because they expected to run a global system, and 5pm PT is 12am UTC. ↩
Perversely, the longer a company keeps up its growth, by whatever means necessary, the higher a premium it trades at. This was a recurring theme in the "compounder graveyard" series: the companies in question were great, for a while, but eventually they had a deep reservoir of institutional and reputational capital, and chose to convert it from intangible assets to GAAP profits, until there were no intangibles left. ↩
This probably applies to countries, religions, and families. The people who make the rituals don't follow them as strictly as the people who codify them; the first Easter Mass, Passover celebration, Eid al-Fitr, etc. was necessarily observed by people who were figuring it out as they went along. Since the simplified, caricatured version of something can be passed down with higher fidelity than the nuanced one, adhering to that is really the only way to be sure you're getting things right.
Stay tuned for a less business-focused and more depressing corollary to this in a future piece. ↩
Part of what distorts this is that everyone has a wildly inaccurate view of how long they work, in part because this is colored by how often they feel like they've put in a lot of time, which itself is related to what their set-point is for hours of focused work per week. There are people who code all day and then go home to unwind by coding on their side projects; I know hedge fund analysts who spend all day researching large-cap stocks for their fund, and then go home to put in time on their hobby of researching small-cap stocks for themselves. Finding these sorts of people, and giving them enough meaningful work to soak up all that restless energy, doesn't make them any worse off, but it does mean they're focusing their energies towards shared goals, which will have a compounding effect. ↩