Growing the Internet Economy
What this covers
Patrick interviews John Collison, co-founder of Stripe, in a conversation that moves from payments infrastructure to corporate structure to accounting. It starts with history: by market cap the 2000 bubble was mostly a telecom bubble, and the fiber overbuild left cheap capacity that became the platform for the next decade. Collison uses that to make a point about Second-Order Effects Thinking. Analysts forecast faster banks and airlines but missed video games and Uber. He puts the internet economy at only 5 to 6 percent of activity, early on the Sigmoid Adoption Curve.
The payments material is the most concrete. Collison explains why it is a scale business earning pennies per transaction, and why going global raises fixed costs. He tours national systems: Alipay and WeChat in China, UPI in India, iDEAL in the Netherlands. He argues that investors who thought online payments were solved missed that the internet's exchange layer is fragmenting. Other threads include identity verification and AML as the next infrastructure gap, Stripe's growth outside Silicon Valley, its Lambda School investment, and Excel as an underrated programming environment for no-code builders.
The contested ground is about structure and measurement. Collison draws on Warren Buffett's Owner Earnings to separate maintenance spending from growth spending, and argues that balance sheets handle software and intangibles badly. He treats GAAP as a human invention that can be improved. He asks why tech rarely uses the Holding-Company Conglomerate Model, citing Constellation Software and Vail Resorts. He also says scale protects incumbents only until a paradigm shifts, and that big companies can move fast, as Facebook did with live video. Whether incumbents or digital natives win varies by industry, as in Netflix versus Disney.
Collison argues the internet economy is still at an early stage, so its growth will come from second-order effects and from infrastructure like payments that the conglomerate and accounting models were never built to capture. Speed and rigor, not scale or conglomerate structure, are what let internet businesses compound.
- Payments on the internet remain fragmented and unsolved, so infrastructure is still being built.
- Speed becomes decisive when technology paradigms shift, and Stripe must keep testing itself against startups.
- Standard accounting treats software and intangible capital poorly, obscuring the split between growth investment and profitable core operations.
Technology creates new wealth, even as platforms repeatedly confront local regulators.
- Late-1980s and early-1990s American cable companies used tactics like mobilizing residents to call mayors against local municipalities, a pattern that resembles Uber's fights with cities, showing that technology platforms repeatedly enter markets over local regulators.
“Exactly like the tactics Uber might've used during that period when they were getting into fights with local cities”
- By market cap the 2000 bubble was mainly a telecom bubble rather than an internet bubble, and the telecom overinvestment in fiber created a glut of cheap internet capacity that washed out in 2001 and 2002 but became the platform for the following decade of growth.
“by market cap, the 2000 bubble was really a telco bubble and not an internet bubble”
- Real estate and banking benefit from fixed supply, such as San Francisco's constrained construction that funnels wealth to landlords and US banking consolidation, whereas technology is not a fixed pie, so new technology can create wealth rather than only redistributing share.
“it is not a fixed pie that everyone's looking to find creative ways to redo the capital stack”
Internet commerce is early and widening globally, with mostly unanticipated effects.
- The internet economy is only about 5 to 6 percent of total economic activity, and since most activity is not yet internet-enabled, the world sits at a shockingly early point on the sigmoid adoption curve, with a majority of economic activity expected to become internet-enabled.
“the internet economy is a very small fraction of the overall economy. Five, 6%, something like that”
- Stripe has grown more outside Silicon Valley than inside it, tripling remote engineers last year, and COVID-driven shifts may make this permanent, which could weaken Silicon Valley's dominance.
“Over time, the company has been growing much more outside of Silicon Valley than within Silicon Valley”
- Making it easy for anyone to start a Delaware company and accept payments lets people in countries outside the US and Western Europe participate meaningfully in the internet economy, acting as a mobile equalizer of opportunity.
“commerce moving online as a mobile equalizer, delivering more opportunity”
- Over-the-top e-commerce lets a product that could never find enough local buyers, such as a niche item sold from Limerick, Ireland, reach a large total global audience, so specialization increases and speed becomes an advantage for upstarts.
“you're seeing more niches possible”
- Analysts forecast the first-order effects of technology, like computers making banks and airlines faster, but miss second-order effects such as video games and Uber, so the most important implications of shifting commerce online are the unanticipated ones.
“people spend lots of time thinking about first-order effects of technology changes”
Speed wins at technology transitions, and disciplined large companies can still move fast.
- Scale and lock-in protect incumbents while a technology paradigm holds, but when the paradigm shifts, as from desktop to browsers or mobile, the ability to quickly build the new platform becomes the defensible advantage, so speed matters most at transitions.
“when the paradigm changes, that's when speed is of the essence and speed is a really defensible trace in companies”
- Whether incumbents can get good at digital faster than digital natives can grow varies by industry, with Amazon as a digital native example and Netflix versus Disney as a more even battle, and Warby Parker's lower cost structure from direct-to-consumer selling being hard for incumbents to follow.
“in which industries do the incumbents get good at digital faster than the digital native companies can grow”
- Working at a company that moves quickly is more enjoyable for employees than working at a slow large company such as IBM, so Collison treats speed as a quality-of-life feature that helps recruit talent.
“employees find it much more enjoyable to work as a company that's moving quickly rather than working for IBM”
- The common narrative that small startups are fast and large companies are slow is wrong, as Facebook's rapid live-video build and Google's use of in-house technology show that large companies can move quickly if they are disciplined.
“the common narrative is just that it's a fairly simple regression with size for a small startups are fast and large companies are slow”
- New Stripe products almost always begin with teams of fewer than 10 people, often fewer than five at the core, because small teams execute faster and the classic big-company mistake is assigning 300 people to a problem for three years before any market feedback.
“the classic big company mistake is to throw 300 people at a problem and have them executing for three years before getting any market feedback”
Global products and payments infrastructure require local adaptation given high fixed costs and country-specific integration demands.
- Payments is a scale business: companies earn pennies per transaction and need enormous volume to be profitable, and moving online has raised fixed costs well above what a domestic-only payments business required, because global coverage requires local bank integrations and engineering.
“as things have moved online, the fixed costs have gone way up compared to what was needed to run a domestic only payments business”
- Earlier payment platforms were assembled from strings of acquisitions, so a merchant expanding to a new region had to rework integrations, while Stripe avoided this by designing from an ideal API backward and standardizing internally.
“the previous generation of companies that people have to use to accept online payments were often assembled through strings of acquisitions”
- Payment systems differ by country: the US relies mostly on credit cards and checks, China shifted from UnionPay to Alipay and WeChat, India built government-backed Aadhaar and UPI, and the Netherlands uses iDEAL bank transfers instead of Visa or Mastercard.
“the Netherlands, people online there don't actually buy with a visa or MasterCard. They predominantly use the thing called iDEAL”
- When Stripe was founded, investors widely assumed online payments were already solved, but Collison argues the internet's information layer is highly interoperable while its economic exchange layer remains heterogeneous and fragmenting, which is why the problem was still open.
“we have a highly complex heterogeneous and increasingly fractured and fracturing landscape”
- Stripe's engineers predicted Visa's hardware failover events and switched over in advance, saving a few transactions per blip; no individual business could afford this optimization, but aggregated across all merchants the improvements are meaningful.
“No business could ever kind of afford to do this optimization themselves. But in aggregate, when you add up all these optimizations, they're really meaningful”
- Products that succeed in the United States often fail in other markets, even the UK, Japan, or Indonesia, because commerce is culturally nuanced, so Stripe needs globally diverse teams to build products that account for local considerations.
“the product that succeeds in the United States is actually going to be very different from the product that succeeds, I mean, even in the UK, nevermind Japan, nevermind India”
- Collison and Patrick built an offline compressed Wikipedia iPhone app during the 2007 to 2008 jailbreak era, which sold well overseas where iPhones lacked good data plans, showing global demand for offline information.
“Patrick and I ran a business that was offline Wikipedia for your iPhone”
- Identity verification is still hard to do across countries, even though it is needed for paying out funds under AML rules and for age-restricted products, so this is an infrastructure gap similar to payments before Stripe.
“That's something that historically has been really, really hard”
- Small bottom-up infrastructure optimizations, such as predicting Visa hardware failovers to avoid dropped transactions, are individually too minor for any single business to fund but add up to meaningful payment success gains across the network.
“No business could ever kind of afford to do this optimization themselves. But in aggregate, when you add up all these optimizations, they're really meaningful”
Management practices are human-made tools that can be learned from others, kept concrete, and held accountable to shareholders.
- Venture firms like Sequoia provide best-practice sharing and in-house functions such as executive recruiting as a service, so an eight-person VC firm can give portfolio companies the operational expertise that a conglomerate like Danaher would have provided through ownership.
“you can be an eight-person firm and you actually have really good access to best practices”
- Accounting standards such as GAAP are invented by humans to serve specific jobs, including taxes, management, creditors, and equity holders, so they can be reasoned about and improved, and criticism of non-GAAP metrics is misplaced because those standards are also arbitrarily chosen and tweaked.
“accounting standards are invented by us humans to give us a view of a business. And they're up to us to choose”
- Silicon Valley's B2B sales playbook has been passed from Oracle to Salesforce and is now on roughly version 27, with each new B2B startup modifying it, showing that best practices spread through people moving between companies.
“Silicon Valley is on version 27 of the playbook and it's always being updated”
- Google did not invent OKRs but adopted them from Intel, illustrating how Silicon Valley companies borrow management systems from predecessors.
“Google is famous for OKRs, Google didn't invented OKRs, they took them from Intel”
- Stripe believes much of management can be studied rather than learned only through experience, so it examined how Apple, Google, and Amazon handled planning, and Silicon Valley's accumulated shared playbook is a collaborative body of best practices.
“we think a lot of this stuff can be studied”
- Investors giving operating advice often say intelligent-sounding but non-actionable things, such as telling a company to build deep moats, and the capital allocation framework is also hard to act on because time horizons are unclear and returns are uncertain.
“investors, when giving advice to operating companies, have a really good knack for saying intelligent ish sounding things that are competing non-actionable”
- Board members are the bosses of management and work for shareholders, and while helping shareholders usually means helping management, the governance role must be clearly understood, consistent with Buffett's criticism of cozy board-management relationships.
“board members need to realize that they work for the shareholders and they are the boss of management and not the other way around”
- Strong board members bring experience the company lacks, such as Jonathan Chadwick's public-company CFO experience at VMware and Skype while Stripe hires a CFO, and Mike Moritz's venture and hiring experience from Sequoia.
“what you're looking for in building a board, is a set of experiences that will be really valuable to the company”
- Good internal and external writing can be tested by asking how you would explain it to a friend, which strips out complex corporate jargon, and by having readers recall what they can and deleting everything they cannot remember.
“Another device I tend to use is, get people to read something and then just tell me everything they can remember for it, and then basically delete everything they can't remember.”
Honest capital allocation separates maintenance spending from growth investment and funds only productive bets.
- Amazon operates closest to a pure capital allocator, using a strict framework to fund new bets, pouring resources into initiatives that work, such as Alexa with thousands or tens of thousands of staff, and withdrawing resources from those that do not.
“Amazon... are probably the tech company that are closest to being pure capital allocators”
- Traditional balance sheets value tangible assets like an espresso machine that holds value and can be sold, but they fail to capture software and intangible capital, such as Google's search engine or Stripe radar, which are hard to value and carried at cost.
“it's really hard to reason about the value of intangible capital like Stripe radar”
- Buffett's owner earnings concept is most useful for splitting capital expenditure into maintenance spending required to keep competitive position and unit volume, and growth spending that expands the business, which companies should be honest with themselves about.
“splitting out CapEx into what is just needed to tread water, investment required for the company to keep its same competitive position”
- Buffett's early textile mill investments illustrate a terrible business in which constant capex is needed just to stay in place, which is the risk of confusing maintenance spending with productive investment in technology companies.
“the textile mills that Berkshire Hathaway got starts with and just being such an incredibly terrible business”
- Tech companies in growth phases reinvest nearly all cash flow, so their true profitability is hidden; separating a cash-generating core from growth investments is useful, but it is imperfect because each business segment is itself fractal, with cash-generating cores and expansion areas nested inside.
“the retail business is fractal, the retail business itself, as you look at us, they are expanding to new lines.”
Autonomous operating units with light central control allocate scarce resources best.
- Technology companies rarely use the holding-company model of one parent owning many independent businesses that share expertise and rotate managers, which is common in other industries; Constellation Software, with roughly 500 independent businesses, is the main counterexample.
“no one really has within technology done, what you actually see pretty commonly in the rest of the world, which is one holding company for a whole bunch of independent businesses”
- Vail Resorts and similar serial acquirers buy companies at good prices and then leave the acquired managers considerable operating latitude instead of tightly integrating them into one platform, yet still generate strong performance.
“they often give the managers of the acquired companies still a lot of latitudes in how they operate”
- Stripe combines top-down strategic guidance that sets rough effort levels, such as expansion in Europe and Southeast Asia versus software, with bottom-up team determination of what to build inside those constraints.
“you end up marrying together those tops down and bottoms up initiatives”
- As a company grows, its central problem becomes allocating scarce resources, such as capital, engineers, or management bandwidth, to the most productive uses, and this framework applies to internal decisions like whether to fund fraud prevention, lending, or international expansion.
“the name of the game is allocating scarce resources”
- Splitting Amazon into a cash-generating retail business and an investment-stage AWS does not work cleanly, because retail itself is fractal, with a cash-generating core plus new lines and geographies that it funds with that cash.
“the retail business is fractal”
Rigor in product and writing comes from questioning constraints and cutting excess.
- A useful test for clear writing is to imagine explaining the subject to a friend over Zoom, because people adopt filler corporate jargon as they rise in organizations, when a plain description like preventing fraud would do.
“how would you explain this to your friend over Zoom”
- A practical editing method is to have readers summarize everything they remember from a document and then cut what they could not remember, because most documents are too long and not tightly edited.
“get people to read something and then just tell me everything they can remember for it, and then basically delete everything they can't remember”
- The returns to writing well are very high, and in a 3,000-person global company with asynchronous communication, documents have far more readers than writers, so it is worth investing heavily in crisp written communication.
“generally documents have many more readers than they have writers”
- Stripe's product rigor comes from repeatedly asking why a constraint exists and tracing it to its root cause, which revealed that instant account setup was possible despite industry claims that compliance and underwriting made it impossible.
“Ask why is that the case? What's the underlying reason? Where is this written down”
Startups should win by validating products with early customers and building lean.
- Stripe's first version reached its first paying customer within three months of writing the first line of code, and that early customer feedback validated which features mattered and prevented building the wrong things.
“we had the first customer using us within three months of writing the first line of code”
- Launching a lean consumer startup is harder now than 10 years ago because a consumer company needs iPhone and Android apps and higher real-time and capability expectations, whereas a decade ago a simple web product with WordPress-style frameworks could reach people broadly.
“it's a little bit harder to do a really lean consumer startup than it was before”
- When Stripe was raising money, investors often assumed online payments was a solved problem, making fundraising hard until the founders repeatedly argued otherwise.
“One of the biggest obstacles we faced was people thinking this is just a solved problem.”
- Whether a company grows by acquisition depends on its stage: early companies compete on product speed and ease of setup that acquisitions cannot buy, while mature companies gain platform breadth that acquisitions can augment.
“Whereas those companies got more mature, it tends to be about platform breadth. I think you can then augment that platform breadth with acquisition. So I think it's probably a bit of a stage of life thing.”
Software engineering stays scarce and hard to measure, and no-code tools still require engineering skill.
- Excel is one of the most underappreciated programming environments because it executes continuously, lays out code and data spatially in cells, and lets many people learn light programming by copying others' models, so no-code tool builders should study it.
“Excel, I think is one of the most underappreciated programming environments in the world”
- Stripe's investment in Lambda School, which retrains people into software engineering in about nine months through remote courses, shows that software engineering remains highly under-supplied and that the skills market still has a large arbitrage.
“the fact that they have had such success, I think speaks to the fact that we're still really short of software engineers”
- Software engineering output is much harder to measure than sales productivity, which can be tracked per person and month, so Stripe relies on frequent engineer surveys as an imperfect but better-than-nothing measure of developer productivity.
“it's incredible to me how hard it is to measure the outputs of software engineering”
- No-code tools are not a full replacement for programmers because building with tools like Zapier still requires reasoning about relationships between objects and data flows, which is a form of engineering even without writing code.
“even when you're doing no code, you're still reasoning about the relations between different objects and data flows”
- Institutions tend to bring people back onto the normal track, but supporting people who go down unconventional rabbit holes, as a supportive school principal did, is the more valuable approach, and Stripe's investment in Pioneer aims at that goal.
“we're really interested in how you help people wander and go down those rabbit holes”
- Ireland's technology industry and scene have changed dramatically since Collison left in 2009, as more software engineering moves outside the Bay Area.
“when I go back to Ireland, it's crazy. It's night and day.”