How to build an advantage that feeds itself

1 The bottom line

A self-feeding advantage is a loop where each customer, product or partner makes the next one cheaper to win or worth more. Copiers can see any single piece, but they cannot copy the whole system.

2 Key takeaways

• Build a loop where each unit of growth lowers your cost or raises what each customer is worth. The gain compounds while rivals stay on a straight line.
  • Morris Chang, TSMC1:18:11: the flywheel started slowly, but TSMC went from behind to ahead within about ten years.
  • Jeff Bezos, Amazon37:10: a price check on 100 bestselling books showed 23% savings for the customer.
• Win one dense core completely before you spread. A new user must always find a live network, not an empty room.
  • Mark Zuckerberg, Meta (Facebook)58:20: Facebook reached over 100 schools within about four months of launching.
  • Bernie Marcus and Arthur Blank, The Home Depot1:06:50: California alone has 250 Home Depots.
• Make every product feed the same loop. Each new product then adds demand for the others instead of competing with them.
  • Michael Eisner, The Walt Disney Company45:09: The Lion King musical has earned over $11 billion in total revenue.
  • Mark Zuckerberg, Meta (Facebook)1:40:21: by November 2005 Facebook served 230 million page views a day with under 10 million users.
• Open the platform so partners and developers carry the growth. Their capital and effort grow your network faster than your own payroll could.
  • Asa Candler, The Coca-Cola Company39:49: within 10 years about 400 bottlers existed, with no capital from Coca-Cola.
  • Jensen Huang, Nvidia2:12:06: CUDA developers grew from 100,000 after four years to 4 million by May 2023.
• Fit the pieces into one system that rivals cannot copy one tactic at a time. Copying a single piece fails without the others.
  • Elon Musk and the SpaceX team, SpaceX37:37: in 2025 all other launchers worldwide did not reach even a fifth of what SpaceX put into orbit.
  • Jensen Huang, Nvidia1:46:42: gross margin rose from 24% in commodity graphics to 70% as software, networking and systems were layered on.

The answer to what's hard to copy isn't any single tactic. It's that the tactics are a system.

Founders, on SpaceX37:37

2.1 Start with a loop where growth improves your economics

The strongest loops make every extra customer cheaper to serve or worth more.

• Write the loop as a cycle and promise to repeat it. Jeff Bezos, Amazon37:10
  • Lower cost allows lower prices, which drives growth, which spreads fixed costs and allows further cuts.
  • Bezos built the strategy on the claim that what is good for customers is good for shareholders over the long term.
  • The host compares it to Henry Ford's "maximum service at minimum cost."
• Look for scale that raises revenue per unit, not only lowers cost. Larry Page and Sergey Brin, Google2:34:23
  • More bidders on each keyword push the winning price up.
  • A long tail of advertisers means more searches earn money at all.
  • So each search is worth more as Google grows, which lets it pay more than any rival to acquire a user.
  • Ranking ads by bid and click-through rate made relevant ads win at lower bids, and users saw better ads. Google2:08:25
• Pour the surplus back into the capability that pulls you further ahead. Morris Chang, TSMC1:18:11
  • Gross margins near 50% and operating margins near 40% funded more fabs and better process technology.
  • Better process let fabless customers build better chips and address more markets, which brought more revenue.
  • In the early 2000s 22 companies were at the leading edge. By the late 2000s it was 14, by the mid 2010s six, and at 5 nanometers only TSMC and Samsung.
• Build the loop into the product design itself. Ingvar Kamprad, IKEA53:00
  • Flat packing cut transport damage and freight.
  • It let customers take furniture home the same day and let IKEA design for machine production.
  • A small Stockholm store had already made knockdown furniture. Kamprad turned the idea into a flywheel that compounded for decades.
  • The IKEA effect, where customers who assemble furniture value it more, was discovered only afterward.

2.2 Seed a dense core so every new user finds a live network

Start where you can be number one fast, then expand from strength.

• Fill one small network completely before opening the next. Mark Zuckerberg, Meta (Facebook)53:20
  • 650 people joined in the first 24 hours, and over half of Harvard's undergraduates were active within two weeks.
  • Every later school was opened the same way, so a new user always found a live network. Friendster, open to everyone, felt empty.
  • Pick a first market where an incumbent exists, to learn at once whether you can displace it. Facebook opened at Columbia three weeks after Harvard, then added neighboring schools to wall in the win. Meta58:20
  • 70% of everyone who ever signed up was active that day, and that held for several years.
• Make trust part of the product from the first day. Mark Zuckerberg, Meta (Facebook)49:50
  • Registration needed a harvard.edu email, so profiles were real identities among classmates.
  • Users entered all their own content, which trained them to contribute.
  • Myspace and Friendster grew faster at first, but Facebook's users were verified people from the same school.
• Enter a market with force so one round of spending covers many units. Bernie Marcus and Arthur Blank, The Home Depot1:06:50
  • Home Depot opened two Atlanta stores the same day, then two more, so one metro's newspaper, radio and TV spend covered many stores.
  • The playbook for 45 years was to enter a city, become number one, then move on.
  • Density later underpinned delivery to 90% of US homes within 2 to 24 hours. Washington state has 48 stores.
• Do not clone a leader's product unless you can move its network. Jeff Bezos, Amazon2:20:20
  • Amazon Auctions copied eBay and had faster search, but eBay already had its buyers and sellers.
  • Amazon also kept auctions on a separate tab with no traffic of its own.
  • The failure led directly to Marketplace.

2.3 Make every product feed the same loop

Each new product should send demand back to the core, not just add revenue.

• Saturate the core hit, then push it into side channels that keep it in view. Walt Disney and Roy Disney, The Walt Disney Company1:11:46
  • Build characters people love, saturate their main medium, then extend into comics, merchandise, books and clubs.
  • The side channels keep the character visible without overexposing it in the core, so scarcity and quality stay there.
  • Design the core with the side products in mind. Each of the seven dwarves got a personality partly so Disney had seven characters to sell. Disney1:44:42
  • Mickey ran as a daily comic strip in 60 US papers and 20 more countries. Roughly 2,183 Snow White products were sold.
  • The Mickey Mouse Club grew to 800 clubs and over a million members, more than the Boy and Girl Scouts combined. Disney56:56
  • Walt hired Herman Kamen on 50% of what he brought in over a baseline. Within a year there were 40 Mickey licenses. Disney59:40
• Choose features that each pull people toward the next. Mark Zuckerberg, Meta (Facebook)1:40:21
  • Events led to photos, tagging and Newsfeed stories, which drew more people to the next event and onto Facebook.
  • Each feature was its own company elsewhere, such as Flickr, Eventbrite and Yelp, but here they reinforced one engagement loop.
  • Photos was written by one engineer in two weeks.
• Bundle an extra that a rival would have to copy as well. Jeff Bezos, Amazon51:52
  • Amazon's standalone video subscription was not working, so Bezos made video free for Prime members.
  • He recalled that Netflix started streaming as an extra to its DVD plan.
  • A competitor might copy Prime shopping or build a Netflix-type service, but was unlikely to do both.
  • Prime began in October 2004 as a moat around the best customers, with Bezos choosing "fast and free" before fulfillment was ready. Prime was live by February 2005. Amazon49:06
• Give each customer a higher rung to climb to. Enzo Ferrari, Ferrari3:00:00
  • Ferrari runs a range, special series, Icona models and supercars, plus clubs, track days and Classiche certification.
  • Growing the base and the height together lets it grow without diluting the brand.
  • Over 90% of the 330,000 Ferraris ever made are still on the road.
  • Eisner added hotels, a timeshare and more parks to turn Disney's parks into multi-day resorts. Spend per family rose from hundreds to thousands of dollars. Disney48:18

2.4 Open the platform so others carry the growth

Let partners, developers and sellers build on you, and your network grows faster than your headcount.

• Give partners ownership so they build the network for you. Asa Candler, The Coca-Cola Company39:49
  • In 1899 Candler let two men bottle Coke at their own risk, with syrup at $1 a gallon and no end date.
  • They sub-licensed territories, and local bottlers appeared with no capital from Coca-Cola.
  • Dee Hock told Bank of America's board that a network owned by all banks would be worth far more than 100% of a closed program. Dee Hock, Visa1:07:00
  • Visa let each bank brand the top band of the card, while the three-band mark stayed universal. Banks made affinity cards with airlines and sports teams. Visa1:34:00
  • Coke gets 27% of system revenue with about 10% of system employees.
  • All 200+ franchise banks signed up and none left. Visa's merchant network was about four times American Express's by the mid-1980s. Visa1:39:00
• Price entry so low that no one has a reason to use a rival. Bill Gates, Microsoft1:10:30
  • Microsoft licensed BASIC to Apple, Commodore and Radio Shack for cash up front, not the most each deal could bring.
  • Gates wanted every maker to find it a no-brainer, so it became the standard and the spiral reinforced itself.
  • Apple paid $31,000 for 8 years of access. Overhead was only about five employees, so low prices were affordable.
• Give the platform away free and sell what it runs on. Jensen Huang, Nvidia31:20
  • Nvidia has never charged for CUDA, but it runs only on Nvidia hardware.
  • From 2006 every GPU shipped was CUDA capable, even though it cost space on the chip and many inside Nvidia doubted it. Nvidia2:12:06
  • Keep every chip compatible. Jensen called that the only unnegotiable rule in the company. Nvidia52:25
  • There are now three million registered CUDA developers and 450 SDKs and models.
• Let outsiders build on your assets and let them surprise you. Jeff Bezos, Amazon Web Services27:45
  • After Tim O'Reilly's visit in 2002, Bezos created a team to open the product catalog to outside developers.
  • Amazon put third-party listings on the same product pages, leaning on its catalog of ASINs, and launched Marketplace in 2000. Amazon2:47:03
  • Marketplace was 15% of orders at launch and is now over 50% of what sells on amazon.com.
  • Google's free Maps API in 2006 started the mashup era and enabled startups such as Zillow and Uber. Google35:25
  • YouTube shared roughly 50% of ad revenue with creators. That took a decade longer to reach profit but built the base. Google1:18:19

2.5 Lock in customers and protect what is hard to copy

The loop pays only if customers stay and rivals cannot rebuild the pieces.

• Invest heavily to win customers you will keep for years. Larry Ellison, Oracle38:12
  • Ellison knew a database customer is locked in for roughly a decade, so winning new ones was worth heavy spending.
  • Bezos made the same bet in 1999, writing that Amazon would "invest heavily in introductions to new customers." Amazon26:30
  • Amazon.com itself did not finish moving off Oracle until 2019, 13 years after AWS launched. AWS1:51:00
• Bundle products so adopting one pulls in the rest. Steve Ballmer, Microsoft1:47:40
  • The Enterprise Agreement was a three-year deal priced per employee or device, covering Windows, Office and servers.
  • Customers adopted whatever came free in the bundle, and zero marginal cost made the bundle cheap to give.
  • Active Directory became the source of truth that pulled in the rest of the stack. Microsoft1:41:40
  • By 2007 about 55% of the $51 billion in revenue came from multi-year and single-year agreements.
• Own the pieces that money cannot buy. Morris Chang, TSMC1:57:47
  • Leading edge chips need ASML's EUV machines, which cost $200 million or more, and few people know how to run them.
  • Forty years of TSMC's people, IP and ASML relationships cannot be rebuilt from scratch.
  • Qualcomm patents enough core technology that no one can build these systems without paying, and keeps the rest as trade secrets. Qualcomm2:18:00
  • Nikon once competed with ASML on lithography and gave up because it was too hard.
  • Qualcomm bought Flarion for $600 million in 2005 to hold patents for 4G, since each new generation expires the old patents. Qualcomm1:42:30
• Build the full stack but let customers buy any layer. Jensen Huang, Nvidia1:45:36
  • Nvidia designs the GPU, CPU, networking and software as one system, yet clouds can buy bare chips while others buy a whole box or rent.
  • The integrated system carries far more margin than the chips inside it. Nvidia1:15:49
  • An eight GPU DGX box starts at $500,000 against $320,000 of GPUs, so about $180,000 is margin on the solution.

2.6 Spend the surplus on the next customer and the next advantage

Each gain should pay for the next step, so rivals fall further behind.

• Overpay for distribution once each user is worth more to you than to anyone else. Larry Page and Sergey Brin, Google2:36:54
  • Google earned more per search, so even an equal revenue share beat a rival's offer, and rivals could not afford to match it.
  • Being private let it overpay without worrying about a weak quarter.
  • Google paid Adobe, Real Networks, WinZip and Dell to bundle its toolbar. Google2:42:27
  • Revenue shares ran up to 100%, and one went above 100%.
  • Toolbar users searched seven times as much, lifting revenue per user from about $2 to $10 or more.
• Stack advantages so each one pays for the next. John D. Rockefeller, Standard Oil30:15
  • He borrowed heavily to grow, used size to win rail rebates, and used the extra profit to buy competitors.
  • Jay Gould offered rebates on all traffic to only the three biggest Cleveland refiners. Rockefeller would not have received that offer without the borrowing and growth.
  • Rebates added $50,000 a year in profit when most other refiners could not break even.
• Use scale to buy the best customers and spread trust features over a huge base. Phil Knight, Nike3:09:54
  • The biggest athletes go where the largest machine pays most, so Nike acquires the LeBrons of the world.
  • Amazon turned customer experience from a variable cost into a fixed cost. A warning that you already bought an item costs the same for 1 million or 70 million customers. Amazon3:21:00
  • Kobe reportedly got $8 million to join Nike. Nike holds 86% of performance basketball shoes.
• Fund loyalty on the supply side and traffic on the demand side. Ingvar Kamprad, IKEA47:27
  • IKEA paid suppliers within 10 days while other dealers took 3 or 4 months, and some suppliers delivered in unmarked vans during the dealers' boycott.
  • It builds extra inventory when a manufacturer has spare capacity, and offers net 30 terms when others offer 60 or 90. IKEA3:04:00
  • It capped restaurant margin at 10% to draw people to the store. IKEA1:25:10
  • IKEA has about 1,600 suppliers in 55 countries, and the average relationship lasts 11 years.
  • 30% of IKEA visitors come just to eat. IKEA also owns shopping centers around new stores to capture the traffic it creates. IKEA2:08:10

2.7 Where they differ

How tightly to control the system, and how much to share, depends on your volume and your product.

• Integrate only when volume pays for it. Elon Musk, SpaceX15:12
  • SpaceX needed first principles, vertical integration and standardization together. Without all three, the system breaks.
  • Vertical integration at the traditional cadence of two to four vehicles a year is a liability. Volume makes the math work.
  • Bezos built the Kindle in house even though it cost more and took longer, because outsourcing would leave Amazon without the skills. Amazon33:35
• Open networks scale faster, but with a worse experience. Dee Hock, Visa1:39:00
  • Visa signed one bank at a time and each brought its own customers and merchants. American Express, as a bank itself, had to become a bank in each country.
  • The open-loop tradeoff was a clunkier, lowest-common-denominator experience.
• Share the technology, but keep the edge. Irwin Jacobs and Andrew Viterbi, Qualcomm2:17:30
  • No phone company would depend on a single vendor, so Qualcomm helped other suppliers become just good enough to be credible.
  • It kept secrets about how to get the best performance, which preserved its lead.
  • Jensen Huang avoids the word moat and prefers a network of developers and customers. The host guesses a moat can breed complacency. Nvidia53:00
• Some content loops work better than others. Walt Disney, The Walt Disney Company1:12:08
  • Animated characters never age and take no cut, so the hosts argue the flywheel works best with animation.
  • Live action IP such as Davy Crockett or James Bond needs reboots, while Mickey stays current.