The Operating System: How Three Decades Rewrote the Economics of Companies
What Technological Innovation From the Web Browser to ChatGPT Can Teach Us About How Companies Compete to Win Today
Check out these reader favorites:
PAID SUBSCRIBER PREVIEW EXCLUSIVE
A Reader’s Guide
Over thirty years, the digital revolution didn’t just create new products, it forced fundamental rewriting and adaptation of the economics companies operate and compete within. This article traces six forces that shaped this transformation:
We’ll trace these six forces chronologically through five eras, showing how each period validated some classical principles while shattering others.
The Five Eras
Part I: The Commercial Internet Era (1995-2000) - Netscape’s IPO ignites the dot-com boom as transaction costs collapse and network effects emerge, culminating in a bubble that proves both classical economics and new platform dynamics matter.
Part II: Web 2.0 and Platform Economics (2001-2007) - Google perfects algorithmic advertising while economists formalize two-sided market theory, turning platforms from market phenomenon into academic discipline and investment strategy.
Part III: Mobile, Marketplaces, and the Gig Economy (2008-2019) - The iPhone and App Store create new operating systems for commerce and labor, as marketplace design becomes engineering and unit economics become standardized metrics.
Part IV: COVID-19 and Platform Acceleration (2020-2021) - The pandemic compresses years of digital transformation into months, validating all three convergences and entrenching all three divergences as platforms become essential infrastructure.
Part V: AI as Operating System (2022-Present) - ChatGPT triggers unprecedented infrastructure investment as artificial intelligence amplifies every force to extreme levels, raising the question of whether we need entirely new economic frameworks.
Hurricane Sandy and the Price of a Ride Home
“Technology changes. Economic laws do not.” Carl Shapiro + Hal Varian
On October 29, 2012, Hurricane Sandy hit New York City. Subway tunnels flooded. Taxis vanished. And Uber’s prices surged to 8.5 times normal rates. The backlash was immediate and fierce. New York’s attorney general threatened investigation. Social media erupted with accusations of price gouging. Uber initially defended the surge, citing Economics 101: higher prices ration scarce supply and incentivize more drivers.
This single moment captures the tension at the heart of the past three decades. Classical economics said Uber was right, price signals allocate resources efficiently. But something felt different. The algorithm wasn’t a market, it was a system with clear human implications. The drivers were participants in a network they didn’t control. The data flowing through the transaction accumulated to Uber, creating competitive advantage no rival could match.
In 1998, economists Carl Shapiro and Hal Varian wrote confidently in Information Rules: “Technology changes. Economic laws do not.” They were both right and catastrophically wrong. The Internet era would prove that transaction costs, price signals, and economies of scale still mattered, but it would also reveal three forces classical theory struggled to explain: network effects that create permanent monopolies, winner-take-all dynamics driven by path dependence, and data as a self-reinforcing asset that accumulates rather than depreciates.
This is the story of how markets evolved, how theory caught up (and sometimes didn’t), and how we arrived at an economic operating system that layers new rules atop old ones. Across this article, we’ll follow the aforementioned economic convergences and divergences across five eras, from Netscape’s 1995 IPO to today’s AI transformation.
The Netscape Moment: August 9, 1995
Netscape went public with a company that had never turned a profit. The IPO price of $28 doubled to $58 by day’s end, valuing the firm above $2 billion. Only sixteen million people worldwide used the Internet. The browser market Netscape dominated would soon be free, as Microsoft bundled Internet Explorer with Windows. Yet investors saw something unprecedented: a product with near-zero marginal cost reaching millions of users.
Marc Andreessen, Netscape’s co-founder, had created Mosaic at the University of Illinois. His browser made the Web visual and accessible. Together with Jim Clark, Andreessen commercialized the technology, betting that if they built the on-ramp to the Internet, revenue would follow. They were right about the infrastructure value, wrong about the business model. But Netscape’s IPO opened the floodgates.
Within months, Amazon (founded July 1994) and eBay (founded September 1995) went live, establishing the foundational architecture for e-commerce. Jeff Bezos chose books because they were low-weight, non-perishable, and had a long tail…millions of titles no physical bookstore could stock. Pierre Omidyar built eBay to connect buyers and sellers directly, taking a small fee per transaction. Both models depended on the Internet collapsing transaction costs.
In 1937, Ronald Coase asked a simple question: Why do firms exist? His answer: to minimize transaction costs, the expense of searching, negotiating, and enforcing agreements in open markets. When transaction costs are high, hierarchy beats markets. When they fall, markets beat hierarchy.
The Internet represented the most dramatic transaction cost reduction in economic history:
Search Costs: Amazon put two million book titles online, searchable in seconds. A physical bookstore might stock 100,000 titles. The difference wasn’t incremental; it was categorical. Buyers no longer needed to visit multiple stores or order from distant catalogs with lengthy delivery times.
Bargaining Costs: eBay eliminated price negotiation through auction mechanisms. Buyers bid, sellers set reserves, and the platform enforces rules. No haggling, no uncertainty, no need for trusted intermediaries beyond eBay itself.
Enforcement Costs: Payment processing, fraud prevention, and dispute resolution have been shifted from bilateral negotiations to platform governance. Early mechanisms were crude, eBay’s feedback system, Amazon’s A-to-Z guarantee—but they worked.
Shapiro and Varian’s 1998 book, Information Rules, argued that these dynamics were classical economics in new clothing. Lock-in, switching costs, and network effects had existed in telephone networks and railroad standards for a century. The Internet simply amplified them. For a few years, this seemed right.
Classical economics recognized supply-side scale advantages: spreading fixed costs over larger output. Steel mills, automobile factories, and telecommunications networks all exhibited this pattern.
Software took it to an extreme. The first copy of Netscape Navigator cost millions to develop. The millionth copy cost nothing—no raw materials, no incremental labor, no distribution expense beyond negligible bandwidth. This was the economics of “non-rival goods” on steroids.
Venture capitalists understood immediately. John Doerr at Kleiner Perkins backed Netscape, Amazon, and Google with a simple thesis: get big fast. Market share would create defensibility through brand, data, and scale economies. Companies that reached critical mass first would win permanently.
This logic drove the bubble. If scale economics were winner-take-all, then current losses didn’t matter. Whoever controlled the most customers would eventually monetize them. The mistake wasn’t recognizing scale economies—it was ignoring unit economics entirely.
While transaction costs and scale economies aligned with classical theory, something stranger was happening with network effects.
The concept wasn’t new. In 1974, Jeffrey Rohlfs published “A Theory of Interdependent Demand for a Communications Service,” analyzing how telephone network value grows with users. In the 1980s, Bob Metcalfe articulated his famous law: network value grows proportional to the square of connected users.
But the foundational academic work came from economists Michael Katz and Carl Shapiro. Their 1985 paper “Network Externalities, Competition, and Compatibility” showed that markets with network effects behave differently from classical markets. Instead of diminishing returns, they exhibit increasing returns. Small early advantages compound. Multiple equilibria become possible. History and luck matter as much as efficiency.
The most important work came from Brian Arthur. His 1994 book Increasing Returns and Path Dependence in the Economy, developed at the Santa Fe Institute, made the break with neoclassical economics explicit. In manufacturing, each additional unit of input yields less output (diminishing returns). In technology markets with network effects, each additional user makes the product more valuable to all existing users. “That which is ahead gets further ahead,” Arthur wrote. Markets don’t converge to competitive equilibrium—they lock in to a dominant provider.
This wasn’t just theory. By 1999, eBay’s value proposition was obvious: most sellers meant most buyers, which meant more sellers. Microsoft Windows dominated because software developers coded for the largest user base. Positive feedback loops replaced competitive equilibration.
Classical economists resisted this framework. It suggested markets could settle into inefficient equilibria, that first movers could maintain permanent advantages, and that competition might not discipline monopolies. But the data kept confirming Arthur’s predictions.
The Dot-Com Bubble
By 1999, the Nasdaq had tripled in three years. Companies with no revenue commanded billion-dollar valuations. Investors justified this with network effects: if Metcalfe’s Law held, early leaders had unlimited potential.
The unit economics were catastrophic. According to research, the average dot-com spent 94 cents on sales and marketing for every dollar of revenue. Pets.com spent $11.8 million on a Super Bowl ad while losing money on every transaction. Webvan raised $800 million to build out grocery delivery infrastructure, then collapsed when customer acquisition costs exceeded lifetime value by an order of magnitude.
The Nasdaq peaked at 5,048 on March 10, 2000. By October 2002, it had fallen 77 percent to 1,108. Companies with neither profits nor paths to profitability went bankrupt. The Wall Street Journal coverage of the crash emphasized a simple lesson: network effects are real, but they don’t suspend basic economics.
The survivors, Amazon, eBay, Google (founded 1998, though not yet public) were companies that mastered both. They understood transaction cost reduction, scale economies, and network effects, but they also built businesses with plausible unit economics. Amazon lost money for years while building infrastructure, but Bezos had a clear model: drive down costs through scale and eventually achieve profitability. eBay was profitable from early on, taking a percentage of each transaction with minimal overhead. Google discovered the most efficient advertising market in history.
Investor Lessons Learned
Sequoia Capital’s Michael Moritz led Google’s Series A, investing $12.5 million for 10 percent of the company in 1999. The return at Google’s 2004 IPO was 400x. Moritz’s philosophy, articulated in interviews, emphasized the marriage of product and distribution: technology alone isn’t enough, nor is market access. Companies need both.
Benchmark Capital’s investment in eBay—$6.7 million for 22 percent in 1997—returned billions. Partner Bill Gurley became one of the most articulate voices on marketplace dynamics and network effects, publishing extensively on platform economics.
Peter Thiel, who co-founded PayPal in 1998, drew a different lesson. Competition is for losers, he later argued. PayPal succeeded by identifying an underserved niche (20,000 eBay power sellers) and establishing a network before its competitors caught up. This became Thiel’s investment thesis: seek monopoly, not competition.
The 1990s proved classical economics still mattered. Companies that ignored transaction costs, unit economics, or basic profitability failed. But the era also revealed that network effects created dynamics classical theory couldn’t fully explain. Markets with increasing returns didn’t equilibrate, they concentrated. First movers gained compounding advantages. And pure efficiency didn’t guarantee survival; timing and network position mattered at least as much.
The stage was set for the 2000s, when economists would build formal theories around the discoveries that practitioners had made.
Operating by John Brewton goes deep on what it takes to build, scale, and optimize modern companies. This breakdown analysis is just the starting point. Check out last week’s analysis of Amazon’s current operating strategy to see more.
What Paid Subscribers Get:
📬 3x Weekly Operating Notes – Short, actionable insights on corporate strategy, operating excellence, and business history
📊 Epic Operating Resource Articles – Deep-dive case studies, historic reviews like this and actionable framework lessons.
💬 Paid Subs-Only Chat & Operating Working Group – Direct access to John and a community of operators tackling real challenges
🎙️ Quarterly Live Operating Q&A – Community working sessions where we break down current strategies and answer your questions
Operating Founders get even more:
🤝 Quarterly 1:1 Operating Sessions with John – Four 30-minute private sessions per year to work through your specific operating challenges
Plans
Monthly: $17/month
Annual: $95/year (save 54%)
Operating Founder: $550/year (includes 1:1 sessions)
Most importantly, thank you for reading! I’d love for you to join our growing community and help you solve your biggest operating conundrums.
When Theory Caught Up to Practice
The Pattern in This Era:
Convergence in focus: Algorithmic price signals created unprecedented efficiency
Divergence emerging: Two-sided markets required new economic frameworks
The result: Platform economics became a formal academic discipline
The Post-Crash Reset
The dot-com crash created a crucible. Companies surviving had to prove business models, not just growth rates. Metrics shifted from traffic and eyeballs to monetization and customer lifetime value. Google, which had been founded in 1998 but remained private through the crash, exemplified this transition.
Google’s breakthrough came not from better search, though its PageRank algorithm was superior, but from better advertising economics. In 2000, Google introduced AdWords, transitioning from a CPM (cost per thousand impressions) model to a CPC (cost per click) model. By 2003, they perfected the model: ads ranked by both bid price and relevance, charged per click, with continuous auction-based pricing.
This created the most efficient advertising market in history. Advertisers paid only for engaged users. Google optimized revenue by matching ads to search intent. Users saw relevant ads rather than spam. Every incentive aligned.
The academic architect was Hal Varian, who joined Google as chief economist in 2002. Varian applied auction theory, traditionally used in the context of Treasury bonds and spectrum licenses, to digital advertising. His work demonstrated how mechanism design could optimize marketplace outcomes in real-time.
This validated classical economics in the strongest possible way. Price signals allocated resources with unprecedented precision. Markets cleared continuously. Information asymmetries decreased as data accumulated. By the mid-2000s, Google achieved 44 percent operating margins in search advertising, one of the most profitable businesses ever created.
Google’s August 2004 IPO, conducted via Dutch auction to democratize access, priced at $85 per share and raised $1.2 billion. The IPO prospectus detailed Google’s business model openly: give away search to capture advertiser revenue. Wall Street initially worried about the unorthodox structure, but the numbers spoke for themselves.
But Google’s model didn’t fit classical supply and demand curves. The company served two distinct customer groups (searchers and advertisers) with interdependent demand. More searchers attracted more advertisers, which funded better search, which attracted more searchers. This wasn’t a linear market; it was a feedback loop.
The theoretical breakthrough came from French economists Jean-Charles Rochet and Jean Tirole. Their 2003 paper “Platform Competition in Two-Sided Markets” established that platforms must choose a price structure, not just a price level. They could subsidize one side (free search) to capture value from another (paid advertising). This was economically rational in ways classical models didn’t accommodate.
Tirole won the 2014 Nobel Prize in Economics “for his analysis of market power and regulation,” with two-sided markets as a central contribution. His work explained not just Google, but credit card networks (subsidize merchants to reach consumers), operating systems (subsidize users to reach developers), and eventually mobile app stores.
The implications were profound. In classical markets, prices equalize supply and demand. In two-sided markets, platforms could subsidize one side indefinitely, cross-subsidizing from the other. This meant:
Free could be a sustainable business model, not a customer acquisition tactic
Market power on one side could create market power on the other
Platform owners gained bargaining leverage over both sides
Traditional antitrust frameworks (focused on consumer prices) missed the dynamics entirely
By 2006, this pattern was everywhere. Facebook, launched in 2004, offered free social networking funded by advertising. YouTube, purchased by Google in 2006 for $1.65 billion, gave away video hosting and sharing. Skype offered free calls funded by premium features. Each platform subsidized users to capture advertiser revenue, data value, or network effects.
By 2007, market concentration was undeniable. Google controlled over 60 percent of search (rising to 90 percent globally within a few years). Facebook dominated social networking with 100 million users by August 2008. Amazon led e-commerce with 30 percent market share and growing.
Brian Arthur’s path dependence predictions had proven correct, but even he might have been surprised by the durability of these positions. Classical economics suggested that monopolies invite competition, high margins attract entrants, innovation disrupts incumbents, and markets self-correct. Instead, network effects created moats that widened over time.
The academic explanation involved multiple reinforcing advantages:
Direct Network Effects: Each new Facebook user made the platform more valuable to existing users. Switching to a rival meant losing your social graph, your photos, your history. The switching cost wasn’t monetary; it was social capital.
Indirect Network Effects: Windows dominated because developers coded for the largest user base, which made Windows more valuable, which attracted more developers. Apple would eventually break this cycle with the iPhone, but it took a decade and a completely new computing paradigm.
Data Accumulation: Google’s search quality improved with scale. More queries meant more data about what people clicked, which improved ranking algorithms, which attracted more users. This feedback loop gave incumbents permanent learning advantages.
Economies of Scope: Platforms that captured users in one domain expanded into adjacent markets. Google moved from search to email (Gmail, 2004), maps (Google Maps, 2005), video (YouTube, 2006), and browsers (Chrome, 2008). Amazon expanded from books to everything. Facebook would later absorb Instagram and WhatsApp.
Research on winner-takes-all dynamics has shown that while some markets support multiple platforms, concentration is the norm. Even where competitors survived, they typically split markets geographically (e.g., Uber and Lyft in the U.S., Didi in China) or demographically, rather than truly competing head-to-head.
The Investor Thesis: Monopoly as Strategy
Peter Thiel, who sold PayPal to eBay for $1.5 billion in 2002, turned his experience into an investment philosophy. His Stanford lectures, later published as Zero to One (2014), made the argument explicit: monopoly should be the goal.
Thiel’s logic: Competition drives profits to zero. Monopoly allows companies to invest in long-term R&D, deliver superior products, and earn sustainable returns. The path to monopoly runs through proprietary technology, network effects, economies of scale, and branding. Companies should start with small markets they can dominate, then expand.
This thinking shaped the era’s biggest bets. Thiel’s Founders Fund backed Facebook in 2004 (500,000 for 10 percent), recognizing social networking’s network effects. He invested in LinkedIn, Yelp, and SpaceX, seeking businesses with natural monopoly characteristics.
Marc Andreessen, who had co-founded Netscape and later Opsware, launched Andreessen Horowitz in 2009 with Ben Horowitz. Their investment thesis, articulated in Andreessen’s 2011 Wall Street Journal essay “Why Software Is Eating the World,” held that software’s economics—high fixed costs, zero marginal costs—would infiltrate every industry. Companies that built platforms rather than products would capture outsize returns.
What This Era Taught Us
The 2000s demonstrated that classical economics still applied—Google’s advertising auctions proved that price signals work brilliantly when designed well. But the era also showed that platform economics required new frameworks. Tirole’s two-sided market theory, Arthur’s path dependence, and the accumulated evidence of winner-take-all outcomes meant economists had to expand their toolkit.
By 2007, platform economics existed as a formal subdiscipline. MIT Sloan Management Review, Harvard Business Review, and business schools worldwide began teaching these concepts. The National Bureau of Economic Research published dozens of working papers on network effects, platform competition, and digital markets.
But if the 2000s were about theory catching up to practice, the 2010s would be about turning theory into engineering—explicitly designing markets using economic principles.
If you’d like to work together, I’ve carved out some time to work 1:1 each month with a few top-notch Founders and Operators. Send me a message. :)
The iPhone Changes Everything
January 9, 2007
Steve Jobs took the stage at Macworld and unveiled “three revolutionary products”: a widescreen iPod, a mobile phone, and an internet communicator. Then he revealed they were one device. The iPhone removed the stylus, removed the physical keyboard, and replaced them with a touchscreen interface. Desktop-class applications and networking capabilities in your pocket.
But the true revolution came 18 months later. On July 10, 2008, Apple opened the App Store with 500 applications. Within a year, 50,000 apps existed. By 2018, the App Store had distributed $100 billion to developers from 2.1 million iOS apps.
The App Store created a two-sided market where Apple controlled both sides. Developers paid $99 annually plus 30 percent of transaction value. Users paid nothing for the store itself. Apple provided distribution, payment processing, and discovery in exchange for its take rate. The model minted billions while creating an entire sector: the app economy.
Studies estimated the app economy generated 466,000 U.S. jobs by 2012, growth from essentially zero in 2007. Mobile developers, designers, product managers, and user acquisition specialists became distinct professions. The OECD recognized the app economy as a new economic category with its own metrics: Daily Active Users (DAU), retention curves, in-app purchase conversion rates, and lifetime value calculations.
Mobile’s always-on, location-aware nature enabled real-time matching and pricing. Uber, founded in 2009 by Travis Kalanick and Garrett Camp, created the defining example: surge pricing.
The mechanic was simple: When demand exceeded supply, prices rose automatically. Higher prices rationed demand (some riders chose alternatives) and increased supply (more drivers logged on). When supply and demand rebalanced, prices fell. This was Economics 101, textbook price signaling.
Economists loved it. Dynamic pricing allocated scarce resources efficiently. Riders who valued trips most paid premium prices. Riders with flexibility waited or used substitutes. Drivers responded to incentives. The market cleared continuously.
The public hated it. Surge pricing during Hurricane Sandy (October 2012) sparked outrage. Prices rising 8x during New Year’s Eve felt exploitative. Snowstorms, concerts, and emergencies all triggered backlash. Wired covered the controversy, noting the tension between economic efficiency and moral intuition.
Research by Harvard Business School professors Alexander Mackay and Samuel Weinstein documented an additional concern: algorithmic pricing could raise market-wide prices even without collusion. Superior pricing algorithms extracted consumer surplus more efficiently than human pricing ever could. This was legal but unsettling, markets were becoming designed systems optimized to capture value.
The Gig Economy: Labor Markets as Platforms
Uber’s success spawned imitators and variations. Lyft launched in 2012. DoorDash entered food delivery in 2013. Instacart started grocery delivery in 2012. TaskRabbit, founded in 2008, connected “Taskers” with one-off jobs. Airbnb, founded in 2008, let homeowners rent spare rooms or entire properties.
Each platform restructured labor markets:
Uber transformed full-time employment into contract work. Drivers used their own vehicles, worked when they wanted, and received per-ride payment. Uber provided the app, payment processing, matching algorithm, and insurance. The company classified drivers as independent contractors, avoiding employment overhead like benefits, payroll taxes, and minimum wage guarantees.
Airbnb enabled asset monetization. Homeowners and renters could generate income from spare capacity—a guest room, a vacation home, an entire apartment during a work trip. The platform handled transactions, reputation systems, and dispute resolution.
TaskRabbit and others created spot labor markets. Need furniture moved? Someone to wait in line? Handyman services? Workers bid on tasks, completed them, and got reviewed. The platform guaranteed payment and managed reputation.
The New York Times (August 2022) asked: “If the Job Market Is So Good, Why Is Gig Work Thriving?” The answer involved both push and pull. Pull factors included flexibility, autonomy, and supplementary income. Push factors included unstable traditional employment, credential requirements, and geographic mismatch between workers and jobs.
By 2019, research estimated 36 percent of U.S. workers participated in gig work to some degree. Wired extensively covered the gig economy, documenting both empowerment narratives (flexible work, entrepreneurial opportunity) and precarity concerns (no benefits, algorithmic management, income volatility, lack of labor protections).
The Economist warned that China’s 200 million gig workers represented a cautionary tale: low wages, no social safety net, and algorithmic control that could penalize workers for factors beyond their control (like bad traffic making them late to pickups).
Marketplace Design as Applied Economics
The gig economy’s explosion created a new discipline: marketplace design. If platforms could architect labor markets, they needed economic principles to optimize them.
Susan Athey, at Stanford Graduate School of Business, pioneered this field. Athey combined economic theory, machine learning, and causal inference to help companies design two-sided markets. She served as consulting chief economist at Microsoft, applying auction theory to Bing advertising and Xbox marketplace pricing. Her research demonstrated how data and algorithms could optimize matching, pricing, and incentives in real-time.
Geoffrey Parker (MIT), Marshall Van Alstyne (Boston University), and Sangeet Paul Choudary synthesized academic insights in their 2016 book Platform Revolution. They provided playbooks for building and operating two-sided markets: how to attract both sides simultaneously, how to price, how to manage competition and governance, and how to scale across markets.
MIT launched its Platform Strategy Summit in 2013, bringing together academics and practitioners to share research on marketplace dynamics. Harvard Business Review published case studies showing that network effects alone weren’t sufficient; platforms needed careful orchestration of supply, demand, pricing, and trust.
SaaS Unit Economics: The Metrics That Matter
While consumer platforms captured headlines, enterprise software was undergoing its own transformation. Software-as-a-service replaced perpetual licenses with subscriptions. This changed everything about how companies managed growth and profitability.
The critical metrics became:
Customer Acquisition Cost (CAC): What does it cost to land a customer?
Lifetime Value (LTV): How much revenue does a customer generate before churning?
CAC Payback Period: How long until revenue recoups acquisition cost?
Annual Recurring Revenue (ARR): Predictable subscription revenue
Net Revenue Retention: Do existing customers expand spending?
Venture capital firms adopted these metrics as funding criteria. Companies needed LTV:CAC ratios of 3:1 or higher. They needed CAC payback under 12 months. The “Rule of 40”—growth rate plus profit margin should exceed 40 percent—became the bar for top performers.
David Skok’s blog “For Entrepreneurs” codified SaaS metrics for a generation of founders. Companies like HubSpot published how they optimized unit economics, reducing monthly churn from 3.5 percent to 1.5 percent and dramatically improving LTV.
Platform Gatekeepers
The 30 Percent Tax
Apple and Google’s app stores created unprecedented developer access to global markets. They also created choke points. Both companies took 30 percent of in-app purchases. For many developers, this was acceptable—the platforms provided discovery, distribution, payment processing, and security. For larger companies, it felt like extraction.
The New York Times reported (April 2021) that Spotify, Epic Games, and others complained to Congress about app store fees. Epic sued both Apple and Google over the commissions, arguing they constituted anticompetitive behavior. The 30 percent take rate had been standard since 2008, but as apps became primary business channels, the toll became material.
OECD research examined how platform concentration affected productivity. While platforms increased efficiency for many participants, they also shifted bargaining power toward platform owners. The data showed “higher persistence among largest platforms weakens positive association with firm-level productivity growth”—in other words, once platforms dominated, they extracted more value and enabled less dynamism.
The Investor Perspective
Software Eats Everything
In his August 2011 Wall Street Journal essay, Marc Andreessen argued that software was infiltrating every industry. Newspapers were being disrupted by web aggregators. Video rental died with Netflix streaming. Retail moved to Amazon. Transportation became Uber and Lyft. Lodging shifted to Airbnb.
Andreessen’s thesis was rooted in economics: software’s zero marginal cost and platform network effects gave it natural advantages. Industries with physical assets, geographic constraints, and regulatory overhead would slowly be unbundled by digital intermediaries capturing the high-value coordination layer.
A16z’s investment strategy reflected this. They backed marketplaces (Airbnb, Lyft, Instacart), SaaS companies (Okta, GitHub, Databricks), and consumer platforms (Instagram, Pinterest, Clubhouse). The fund published extensively on marketplace design, network effects, and platform strategy, becoming a knowledge platform itself.
Ben Horowitz emphasized founder quality and cultural fit. His book The Hard Thing About Hard Things documented the operational realities of scaling platforms: managing two-sided liquidity, balancing supply and demand, navigating regulation, and making layoffs when markets shifted.
What This Era Taught Us
The 2010s proved that all six forces—three convergences, three divergences—operated simultaneously. Transaction costs approached zero (convergence). Price signals worked instantaneously (convergence). Economies of scale reached unprecedented levels (convergence). Network effects created dominant positions (divergence). Winner-take-all markets concentrated power (divergence). Data accumulation became self-reinforcing capital (divergence).
But most importantly, this era showed that markets weren’t emerging organically—they were being designed. Economists weren’t observing; they were building. Platform architecture decisions determined who captured value, how participants interacted, and which equilibria emerged.
By 2019, the question wasn’t whether platforms would dominate, but whether they could be governed, regulated, or constrained. That question became urgent in 2020.
March 2020: The Great Compression
In early March 2020, companies worldwide sent employees home. Offices closed. Business travel ceased. In-person meetings moved to Zoom. Slack usage doubled. Microsoft Teams grew from 44 million daily users in March to 115 million by October.
The shift wasn’t a gradual adoption of new tools; it was a forced, immediate, total transformation. MIT Sloan Management Review documented how digital strategy, not just technology, determined which organizations adapted successfully. Companies with mature digital capabilities, cloud infrastructure, remote work policies, and distributed decision-making thrived. Those dependent on physical presence struggled.
This validated all three convergences at once:
Transaction Costs: Zoom eliminated commute costs, travel expenses, and geographic friction. Global teams collaborated in real time. The cost of coordination dropped to near zero.
Price Signals: Delivery platforms used dynamic pricing to manage demand surges. DoorDash, Instacart, and Amazon Flex increased delivery fees during peak periods, balancing customer orders with available drivers.
Scale Economies: Cloud infrastructure scaled to absorb simultaneous demand shocks. AWS, Azure, and Google Cloud enabled companies to provision capacity instantly. Netflix, Zoom, and streaming services handled unprecedented concurrent usage without service degradation.
The Platform Windfall
The New York Times documented how platform companies captured disproportionate pandemic gains. Amazon’s revenue grew 38 percent in 2020. Facebook’s daily active users increased 12 percent. Google’s search advertising rebounded after a brief Q2 2020 dip. Microsoft’s market cap grew $500 billion during the pandemic.
Meanwhile, small businesses suffered. Restaurants, retail stores, gyms, and personal services either closed permanently or survived on dramatically reduced capacity. The divergence wasn’t about competence—it was about structure. Platforms benefited from network effects: more users attracted more merchants, which attracted more users. Physical businesses faced capacity constraints, geographic limits, and regulatory restrictions.
This entrenched all three divergences:
Network Effects: Dominant platforms became essential infrastructure. Zoom became synonymous with video calls. Amazon became the default for online retail. Slack and Microsoft Teams became collaboration standards. Network effects turned temporary necessity into permanent habit.
Winner-Take-All: Market concentration increased. Amazon’s e-commerce share grew. DoorDash consolidated restaurant delivery. Instacart dominated grocery delivery. The pandemic accelerated trends already in motion, but compressed them dramatically.
Data Accumulation: Platforms accumulated behavioral data at unprecedented scale. Amazon learned purchasing patterns across categories. Zoom gathered insights on meeting behaviors. Netflix refined recommendation algorithms. This data compounded competitive advantages—incumbents learned faster than entrants.
Remote Work: Permanent Shift or Temporary Accommodation?
The Bureau of Labor Statistics tracked remote work’s persistence. Before COVID, roughly 5 percent of U.S. workers worked from home full-time. By April 2020, that figure hit 60 percent for workers whose jobs allowed it. Even as offices reopened, remote work stabilized at 15-20 percent, triple pre-pandemic levels.
Research documented productivity impacts. Knowledge workers in focused roles (programming, writing, analysis) often maintained or improved productivity. Collaborative roles (product development, strategy, innovation) showed mixed results. Managers struggled with coordination and mentorship. New employees faced onboarding challenges.
The shift concentrated benefits among credentialed knowledge workers. Service workers, manufacturing employees, and essential workers couldn’t work remotely. This widened inequality—high earners saved commute time and gained flexibility while lower earners faced health risks and economic precarity.
Regulatory Scrutiny Intensifies
The pandemic highlighted platform power’s social implications. As Amazon and tech giants thrived, antitrust conversations intensified. The House Judiciary Committee’s Antitrust Subcommittee published a 449-page report documenting anticompetitive behavior by Amazon, Apple, Facebook, and Google. The EU accelerated investigations and fines. State attorneys general filed suits.
The critique: platforms had become infrastructure, but operated as private companies pursuing profit maximization. They controlled which businesses succeeded, what information circulated, and how markets functioned. This power exceeded historical monopolies because platforms operated across sectors and geographies simultaneously.
Defenders noted that platforms provided enormous consumer value—free search, low-cost goods, convenient services. But critics argued that value extraction from merchants, data collection, and market foreclosure represented anticompetitive harm not captured by traditional consumer welfare standards.
What This Era Taught Us
COVID-19 compressed digital transformation timelines and validated both the promises and perils of platform economics. The convergences proved real: transaction costs can approach zero, price signals can optimize allocation, and scale economies can enable remarkable productivity. The divergences proved durable: network effects entrench incumbents, winner-take-all markets concentrate power, and data becomes self-reinforcing capital.
The pandemic showed that platforms weren’t just companies—they were infrastructure. This raised questions: Should infrastructure be privately owned? How should society govern essential digital platforms? Can markets remain competitive when network effects create permanent advantages?
These questions would become more urgent as the next technological wave emerged: artificial intelligence.
The Pattern in This Era:
Convergences amplified or redefined?: AI further reduces transaction costs, optimizes pricing, and scales knowledge work—but in ways that may require new frameworks
Divergences intensified or transformed?: AI exhibits network effects, enables winner-take-all concentration, and turns data into even more powerful capital—but does this create new economic laws?
The open question: Is AI classical platform economics at extreme scale, or something genuinely new?
November 30, 2022
ChatGPT and the Paradigm Shift
OpenAI released ChatGPT to the public as a research preview. Within five days, it had one million users. Within two months, 100 million. No product in history had scaled that fast, not Facebook, not TikTok, not Instagram.
Unlike prior enterprise software that required months to evaluate and deploy, ChatGPT was instantly useful. It wrote emails, debugged code, drafted essays, summarized documents, and answered questions in natural language. The interface was a text box. The learning curve was zero.
Within weeks, Wired reported that tech giants were committing unprecedented capital to AI infrastructure. Microsoft invested $10 billion in OpenAI. Google accelerated its AI roadmap. Meta allocated $50 billion to AI research and infrastructure. Amazon Web Services expanded data center capacity. By 2025, the top tech companies were investing $370 billion annually in AI—nearly 45 percent of Microsoft’s revenue alone.
The Infrastructure Race
The Financial Times documented Virginia’s explosive data center growth. Data centers consumed 9.6 gigawatts of capacity, with construction sprawling across Virginia and Maryland. Jensen Huang, Nvidia’s CEO, declared: “Intelligence is essential infrastructure. We are clearly in the beginning stages of build-out.”
This was the 1990s internet infrastructure boom again, but compressed and intensified. Nvidia became the critical supplier, with its GPUs powering AI training and inference. NVIDIA’s market cap grew from $360 billion in January 2023 to over $3 trillion by mid-2025. Data center REITs became growth stocks. Power grid capacity became a constraint—utilities struggled to supply enough electricity for training runs that consumed megawatts continuously.
The capital requirements meant only a handful of companies could compete at the frontier. Tr aining GPT-4 reportedly cost over $100 million. Future models would cost more. This created a natural oligopoly: OpenAI (backed by Microsoft), Google DeepMind, Anthropic (backed by Amazon and Google), and Meta could afford frontier research. Everyone else built on their models.
AI further collapsed coordination costs. McKinsey research documented how AI agents could automate tasks previously requiring human judgment:
Customer service bots handled routine inquiries, escalating only complex issues
Code generation tools wrote functions from natural language descriptions
Legal document review automated contract analysis
Financial modeling generated scenarios from text prompts
Supply chain optimization adjusted to real-time constraints
This raised classic Coasean questions. If AI agents can coordinate work automatically, why do firms need middle managers? Why have hierarchies at all? Berkeley research titled “From Coase to AI Agents: Why Economics of the Firm Still Matters” examined whether Coase’s framework still applies when intelligent software replicates internal coordination functions.
The tentative answer: yes, but transformed. Firms still exist to minimize transaction costs, but AI shifts which transactions happen inside firms versus across markets. Tasks requiring trust, context, and tacit knowledge remain in-house. Standardized work moves to AI-mediated spot markets. The boundary moves, but Coase’s logic holds.
AI-driven dynamic pricing became ubiquitous. Airlines, hotels, ride-sharing, e-commerce, and SaaS companies used machine learning to optimize prices continuously. The algorithms considered:
Individual customer willingness-to-pay (inferred from browsing, purchase history, demographics)
Real-time supply and demand
Competitor pricing
Time-based patterns
External factors (weather, events, seasonality)
Research by Harvard Business School and UCLA documented concerning implications. AI pricing could extract consumer surplus more efficiently than any prior method. Personalized pricing—charging different customers different amounts for identical goods—became standard practice, hidden behind “dynamic” pricing labels.
The social tension: this was economically efficient (resources allocated to highest-value uses), but felt unfair. Two customers buying the same flight could pay wildly different prices. Surge pricing for essential goods during emergencies provoked moral outrage. Yet, paradoxically, these same mechanisms enabled services that couldn’t otherwise exist at scale.
AI promised unprecedented leverage: automate work without hiring proportionally. Bain & Company research (June 2025) found that early AI adopters reported “doubling EBIT margins by reimagining workflows with AI-native processes.” These weren’t incremental efficiency gains—they were architectural changes.
Examples included:
Law firms using AI to draft contracts, reducing junior associate needs
Consulting firms automating slide decks and analysis
Software companies using AI coding assistants, reducing engineering team sizes
Customer service operations replacing human agents with AI
Content creation (marketing, journalism, entertainment) generated partially or fully by AI
The pattern: knowledge work that involved pattern matching, summarization, synthesis, or routine analysis could be automated or augmented. Work requiring genuine creativity, complex judgment, or human relationship remained human-performed but AI-assisted.
This created a productivity surge for companies that adapted. It also raised employment questions: if AI enables ten people to do the work of fifty, what happens to the forty?
AI amplified data network effects to unprecedented levels:
More data → Better models → Better products → More users → More data
The flywheel accelerated. Companies with large existing datasets (Google’s search queries, Amazon’s purchase histories, Meta’s social graphs) could train better models. Better models attracted more users. More users generated more training data.
This created compounding advantages. Google’s AI search results improved faster than rivals because it had more query data to learn from. Amazon’s product recommendations got better because it observed more purchases. Meta’s content ranking optimized engagement because it saw more social interactions.
A16z research (April 2023) on “Beyond Metcalfe’s Law” noted that AI’s learning curves meant network effects strengthened over time, not plateaued. Traditional networks reached saturation—there’s limited value to adding the billionth Facebook user. AI networks kept improving with more data, creating “learning effects” that compounded indefinitely.
The capital costs, data advantages, and engineering talent required to build frontier AI models meant only a handful of companies could compete. Unlike prior platform markets, where dozens of companies launched (remember Friendster, MySpace, and Orkut before Facebook won?), AI started out concentrated.
The Wall Street Journal asked: “Is the AI Boom a Bubble?” Comparisons to the 1990s dot-com bubble emerged. Investment bank research showed AI infrastructure spending rivaled or exceeded the late-1990s telecom buildout. The Nasdaq’s AI-driven surge in 2023-2024 echoed the patterns of 1998-1999.
But key differences suggested this wasn’t a bubble in the same way:
AI models demonstrated immediate, tangible value (unlike dot-coms with revenue-free business plans)
The infrastructure being built (data centers, power capacity, networking) had clear use cases
Major customers (enterprises, governments, consumers) were adopting rapidly
Unit economics, while unclear for some use cases, showed profitability paths
Still, questions remained. Would AI economics resemble cloud computing (few hyperscale winners: AWS, Azure, Google Cloud) or software (thousands of profitable companies building on shared infrastructure)? Early evidence suggested the former—foundation models would concentrate, applications would proliferate.
Forbes and Schroders asked: “Does the winner take all, or do we all win?” The answer seemed to be: both. A few companies would dominate foundation models, but thousands of companies would build AI-powered applications. The question was how value would split between layers.
If data was important in the 2010s, it became essential in the AI era. The quality, quantity, and diversity of the training data determined the model's capability.
This created several new dynamics:
Data as Competitive Moat: Companies with proprietary datasets—medical records, financial transactions, customer interactions, industrial sensor data—had advantages competitors couldn’t easily replicate. Reddit’s data licensing deals, Stack Overflow’s AI partnerships, and publishers’ content licensing negotiations reflected data’s elevated value.
Data Scarcity: By 2024, AI companies had exhausted high-quality public internet data. Future scaling required synthetic data, private data licensing, or multimodal data. This put data owners (publishers, platforms, individuals) in stronger bargaining positions.
Regulatory Capture: Data privacy regulations—GDPR in Europe, CCPA in California, emerging federal legislation—created compliance costs that incumbents could absorb but startups struggled with. This inadvertently entrenched large platforms with existing data stockpiles.
Data as Non-Depreciating Asset: Unlike physical capital that depreciates, data accumulated value over time. A decade of Amazon purchase history became more valuable with AI, not less. This meant incumbents’ historical data advantage widened, not narrowed.
Analysis emphasized that data “has been catapulted into the status of traditional assets like real property and commodities, but is very different in nature”—non-depreciating, combinatorial, and self-amplifying in value.
Academic and Research Frontier:
Do We Need New Theories?
The National Bureau of Economic Research, Stanford, Berkeley, MIT, and other institutions grappled with whether AI required new economic frameworks.
Option A: Classical Economics + Network Effects + Data Capital = AI is these existing forces at extreme scale. No fundamentally new laws needed, just adjustment of parameters.
Option B: AI Requires New Theory = Autonomous agents, recursive self-improvement, and general intelligence create economic dynamics that existing frameworks can’t fully capture.
Early evidence suggested Option A, but with important nuances. Chinese economists developed “digital economy theory” recognizing data as a distinct factor of production alongside capital, labor, and land. This wasn’t a new law, but an expansion of accounting frameworks to capture intangible assets.
Berkeley research asked: “As AI agents automate coordination, how do firm boundaries shift?” The answer seemed to be that Coase’s framework still applied—firms exist to minimize transaction costs—but AI changed which transactions were cheapest to internalize versus buy in markets.
Media and Policy Response
The New York Times published opinion pieces arguing “Big Tech’s predatory platform model doesn’t have to be our future,” advocating for regulatory intervention. Tim Wu’s work suggested structural separation—breaking apart platforms’ multiple roles as marketplace, participant, and regulator.
The Financial Times examined Europe’s tech dependency and infrastructure vulnerabilities. The Economist analyzed Amazon’s evolution and questioned sustainability of current business models as AI changed e-commerce economics.
A16z maintained advocacy for founder-friendly policy, arguing that innovation required regulatory light-touches. Their “Big Ideas in Tech” and “Little Tech Agenda” publications emphasized that smaller companies, not just giants, were building AI applications.
The policy debate centered on three questions:
Market Structure: Should AI foundation models be regulated as utilities? Should they be required to interoperate?
Data Governance: Who owns training data? Should individuals receive compensation when their data trains models?
Labor: If AI automates knowledge work at scale, what policies enable economic security and opportunity?
These debates remained unresolved as of late 2025, with legislation pending in the EU, U.S. Congress, and state legislatures worldwide.
What This Era Is Teaching Us
The open question: Is this classical economics plus platform dynamics at maximum amplitude, or are we witnessing the emergence of genuinely new economic phenomena that require expanded theory?
The answer will determine how we govern, regulate, and adapt to an AI-transformed economy.
Conclusion: A New Operating System
The Pattern Across 30 Years
We opened with Hurricane Sandy and Uber’s surge pricing—a moment when classical economics (efficient resource allocation) collided with new realities (algorithmic control, platform power, data extraction). That collision wasn’t unique to 2012. It has recurred across three decades as technology forced economics to evolve.
The three convergences with classical theory and the three divergences from it explain the pattern:
Transaction Costs (Convergence #1)
Coase was more right than he knew. The Internet, mobile, cloud, and now AI have progressively driven coordination costs toward zero. When transaction costs fall, markets replace hierarchies. Platforms emerged because they minimized friction better than any alternative. This validated classical economics—firms exist to minimize transaction costs, and technology keeps redefining what’s possible.Price Signals (Convergence #2)
Classical theory held that prices allocate resources efficiently. Digital markets proved this spectacularly. Google’s ad auctions, Uber’s surge pricing, and AI-driven dynamic pricing optimize allocation in real time. Markets clear continuously. Information asymmetries shrink. The machinery of price signals works better than ever—though whether the outcomes feel fair is a separate question.Scale Economies (Convergence #3)
Zero marginal cost amplified classical supply-side advantages to an extreme. Software, platforms, and AI exhibit unprecedented scale economies. The first copy costs millions or billions. The millionth costs nothing. This was always true in economics—steel mills, railroads, utilities all showed scale effects. Technology just removed the physical constraints that eventually created diseconomies.Network Effects (Divergence #1)
Demand-side increasing returns broke classical assumptions about diminishing returns and equilibrium. Brian Arthur showed that small advantages compound in technology markets. Facebook didn’t just get bigger—it got more valuable per user as it grew. This created positive feedback loops classical theory couldn’t easily accommodate. Markets don’t converge to competition; they lock in to dominance.Winner-Take-All (Divergence #2)
Path dependence meant history and luck mattered as much as efficiency. Multiple equilibria became possible. First movers gained permanent advantages. Peter Thiel was right: monopoly, not competition, became the rational goal. This violated neoclassical assumptions about markets self-correcting. Instead, concentration intensified over time.Data as Capital (Divergence #3)
The most novel force—data as non-depreciating, self-reinforcing, combinatorial capital—emerged gradually across the three decades. Data didn’t fit classical asset categories. It didn’t depreciate. It became more valuable when combined. It created learning advantages that compounded indefinitely. This required frameworks beyond traditional capital-labor-land models.
The past thirty years weren’t a contest between classical and platform economics. They demonstrated that economic theory is living practice, refined by reality:
The title of this piece—”The Operating System”—captures the layered reality. Modern companies don’t operate according to either classical or platform economics. They operate according to both, simultaneously:
Layer 1 (Classical Foundation): Minimize transaction costs, respond to price signals, achieve scale economies. These principles remain true and essential.
Layer 2 (Network Effects): Capture demand-side increasing returns, build positive feedback loops, create platform architectures that connect producers and consumers.
Layer 3 (Winner-Take-All): Pursue defensible positions through proprietary technology, network effects, and switching costs. Accept that some markets will concentrate rather than fragment.
Layer 4 (Data Capital): Accumulate proprietary data assets, build learning flywheels, create compounding informational advantages.
Layer 5 (AI Automation): Automate coordination, optimize decisions, and scale knowledge work—amplifying all prior layers.
Successful companies master all five layers. Failure at any level compromises the entire stack. This is why building and operating companies has become more complex, more specialized, and more dependent on economic sophistication.
The Unanswered Questions
Three decades of evolution leave critical questions unresolved:
For Operators: How do you compete when incumbents have permanent advantages from network effects, massive data assets, and overwhelming capital? Worker cooperatives, protocol-based alternatives, and niche positioning offer partial answers, but the challenge remains. Most markets concentrate toward a few winners.
For Economists: Does artificial intelligence require entirely new theoretical frameworks, or is it classical economics (plus network effects, plus data capital) at extreme scale? If AI agents coordinate like firms, what does that mean for organizational boundaries? If data becomes essential capital, how do we measure it, tax it, and regulate it?
For Society: How do we govern essential infrastructure owned by profit-maximizing private companies? Traditional antitrust focuses on consumer prices, but platform power manifests through data control, merchant fees, and market foreclosure. Should platforms be regulated as utilities? Should they be required to interoperate? Should data rights belong to individuals or companies that collect it?
Economics didn’t stop evolving in 2025. Each technological wave—from Internet to mobile to AI—expanded what theory must explain. The convergences remind us that classical principles remain foundational. The divergences show that new phenomena require new frameworks.
The companies that thrive will be those that master both the old rules and the new ones: minimize transaction costs while capturing network effects, optimize price signals while accumulating data capital, and achieve scale economies while navigating winner-take-all dynamics.
The economists who shape the next chapter will be those who design markets, not just describe them—who build tools for platform operators while expanding theory to accommodate AI, data, and whatever comes next.
And the societies that flourish will be those that govern these forces wisely—preserving the efficiencies of platform economics while preventing the extraction, concentration, and power imbalances that harm competition and opportunity.
We’ve spent thirty years watching economic laws both hold and evolve. The next thirty will show whether we can govern the operating system we’ve built.
- john -
John Brewton documents the history and future of operating companies at Operating by John Brewton. He is a graduate of Harvard University and began his career as a Phd. student in economics at the University of Chicago. After selling his family’s B2B industrial distribution company in 2021, he has been helping business owners, founders and investors optimize their operations ever since. He is the founder of 6A East Partners, a research and advisory firm asking the question: What is the future of companies? He still cringes at his early LinkedIn posts and loves making content each and everyday, despite the protestations of his beloved wife, Fabiola, at times.
Appendix: Sources & References
This article draws on research from leading academic institutions, business publications, and primary sources spanning three decades of digital transformation. Below are the key sources organized by category and topic.
Academic Economics & Theory
Ronald Coase - Transaction Cost Economics
Coase, R. H. (1937). “The Nature of the Firm”
Brian Arthur - Increasing Returns & Path Dependence
Arthur, W. B. (1994). Increasing Returns and Path Dependence in the Economy
Santa Fe Institute research on complexity economics
Jean Tirole - Two-Sided Markets
Nobel Prize in Economics 2014 - “for his analysis of market power and regulation”
Rochet, J-C. & Tirole, J. (2003). “Platform Competition in Two-Sided Markets”
Michael Katz & Carl Shapiro - Network Effects
Katz, M. & Shapiro, C. (1985). “Network Externalities, Competition, and Compatibility” - American Economic Review
Hal Varian - Information Economics & Mechanism Design
Shapiro, C. & Varian, H. (1998). Information Rules: A Strategic Guide to the Network Economy
Susan Athey - Market Design
Business Press & Analysis
The Wall Street Journal
Andreessen, M. (August 2011). “Why Software Is Eating the World”
The New York Times
Platform Companies Coverage (March 2021)
“If the Job Market Is So Good, Why Is Gig Work Thriving?” (August 2022)
App Store Antitrust Coverage (April 2021)
The Economist
Platform Competition Coverage (February 2021)
Wired
Netscape IPO Anniversary (August 2011)
AI Data Center Infrastructure (November 2025)
Financial Times
Platform Economics Coverage
Academic Institutions & Research Centers
MIT Sloan Management Review
Harvard Business Review
“Network Effects Aren’t Enough” (March 2016)
Platform strategy case studies
Stanford Graduate School of Business
Susan Athey’s research on marketplace design
Berkeley, Haas School of Business
Technology Companies & Milestones
Netscape
AdWords and advertising auction mechanism
Apple
App Store 10th Anniversary (July 2018)
Venture Capital & Investor Perspectives
Andreessen Horowitz (a16z)
Benchmark Capital
Sequoia Capital
Founders Fund
Government & Policy Sources
Bureau of Labor Statistics
Remote work tracking (2020-2024)
OECD
App Economy research (2013)
Platform productivity analysis
House Judiciary Committee
Antitrust Subcommittee Report on Digital Markets (2020)
Books & Monographs
Arthur, W. B. (1994). Increasing Returns and Path Dependence in the Economy
Shapiro, C. & Varian, H. (1998). Information Rules: A Strategic Guide to the Network Economy
Thiel, P. (2014). Zero to One: Notes on Startups, or How to Build the Future
Parker, G., Van Alstyne, M., & Choudary, S. P. (2016). Platform Revolution
Horowitz, B. (2014). The Hard Thing About Hard Things
SaaS Metrics & Unit Economics
For Entrepreneurs (David Skok)
SaaS metrics codification and best practices
Key Metrics Covered:
Customer Acquisition Cost (CAC)
Lifetime Value (LTV)
CAC Payback Period
Annual Recurring Revenue (ARR)
Net Revenue Retention
Rule of 40
Contemporary AI Research & Coverage
Bain & Company
AI Transformation Research (June 2025)
McKinsey & Company
AI adoption and impact studies
OpenAI
ChatGPT launch and development
Nvidia
AI infrastructure and GPU development
Note on Sources: This article synthesizes research from academic papers, business journalism, company reports, and investor commentary spanning 1995-2025. All working hyperlinks are embedded in the article text. Where specific papers or historical coverage is no longer available online, citations reference the original publication details and dates for verification through academic databases and archives.
This is a masterful breakdown...
The tension between classic theory and the digital reality of increasing returns, network effects, and data accumulation is the core tension of the last 30 years.
I think the move from the "Post-Crash Reset" to formalizing platform economics is the most critical shift.
It makes you wonder: with AI accelerating every one of these forces to an "extreme level," what new economic framework will we need to invent in the next decade?
Reading through the eras, I couldn't help but think of the Buffett Indicator. This history feels like the perfect explanation for why that market-cap-to-GDP ratio has been so elevated for much of the last decade.
The "increasing returns" and winner-take-all dynamics you describe are the very engine that powered valuations to levels that seem disconnected from the underlying economy. A great lens for understanding the modern market.