3. Corporate Networks
When I was in college, I remember thinking to myself, this internet thing is awesome because you can look up anything you want, you can read news, you can download music, you can watch movies, you can find information on Google, you can get reference material on Wikipedia, except the thing that is most important to humans, which is other people, was not there.1
Skeuomorphic and Native Technologies
People use new technologies in one of two ways: (1) to do something they could already do but can now do faster, cheaper, easier, or higher quality; or (2) to do something brand-new that they simply couldn’t do before. Early in the development of new technologies, the first category of activities tends to be more popular, but it’s the second set that has more lasting effects on the world.
Improving existing activities happens first because it’s more straightforward. Discovering the real powers of a new medium takes time. When Johannes Gutenberg, inventor of the movable-type printing press, published his namesake Bible in the fifteenth century, he made it look like a handmade manuscript copy. Who could imagine a book as anything else? And yet, as Alan Kay, the computer scientist and Turing Award winner, once observed, “the real message of printing was not to imitate hand-written Bibles, but 150 years later to argue in new ways about science and political governance”—a catalyst for revolution.
Doing new things requires leaps of imagination. Early film directors shot films like plays. Effectively, they were making theatrical productions with better distribution models. Filmmaking changed only after true innovators realized the potential for a visual grammar native to the new form. The development of electricity followed a similar path. People originally switched from gas and candles to electric light for the convenience. Only decades later did they tap the electric grid to power all manner of appliances, from toasters to Teslas.
Technology that borrows from what came before is sometimes called skeuomorphic. The term originally referred to design elements in art that are intentional, albeit unnecessary, holdovers. During the Steve Jobs era, Apple popularized the idea2 as a description of digital graphics that look like familiar objects, like a wood-grained bookshelf as decoration in a reading app or a trash can icon to represent deleted files. Skeuomorphic design made it easier for people to adapt to interactions with computer screens. Now people in the tech industry use the term to describe technologies that mimic existing activities or experiences. Copying things that already exist makes new things feel familiar and helps people get comfortable with them.
The internet was skeuomorphic in the 1990s. At the time, it mostly consisted of digital adaptations of pre-internet things: websites imitating brochures and catalogs, email as an extension of letter writing, and shopping reminiscent of mail-order commerce. People called this the read era because, although people could send emails, submit data, and buy things, information typically flowed one way: from website to user. The analogy is to a read-only digital file, which can be opened and viewed but not edited. Website making was a specialized skill back then, and most activities didn’t involve publishing to wider audiences.
It’s hard to picture this now, but the internet of the 1990s and early 2000s3 was nothing like the always-on, high-speed mobile internet of today. People would sit in front of a bulky desktop PC and “log on” only intermittently, usually to check email, plan travel, or browse the web. Images loaded slowly, and video streaming, when it worked at all, was janky. Most people logged on via dial-up modems—slow landline connections that plunked and plodded along at speeds people today would regard as excruciating.
Even at the height of dot-com mania, enthusiasm for the internet went only so far. Just before the excitement peaked in March 2000, the National Academy of Engineering placed the internet thirteenth on its list of greatest engineering achievements of the twentieth century.4 The invention ranked below radio and telephones (sixth), air-conditioning and refrigeration (tenth), and space exploration (twelfth).
Then—pop!—the bubble burst. Stocks tanked across the board. In 2001, Amazon’s share price hit an all-time low.5 The retailer’s market cap reached a nadir of $2.2 billion (less than half a percent of what it is today). When the Pew Research Center, a prominent polling firm, asked Americans in October 2002 if they would adopt broadband,6 the majority said no. People mostly used the internet for email and web “surfing.” Did it really need to be faster? The mainstream consensus was that the internet was cool, sure, but it had limited uses and probably wasn’t a good place to build a livelihood. The market crash proved that.
Yet the internet was on the cusp of a renaissance—even as the industry smoldered, a small but growing movement was coalescing.
By the mid-2000s, technologists were beginning to explore internet-native product designs. If “skeuomorphic” means more of the same, then “native” indicates novelty. New services were cropping up that would take advantage of the internet’s unique capabilities, rather than merely retreading paths beaten by offline analogs. Key trends included blogging, social networking, online dating, public résumé building, and photo sharing. Technical innovations like APIs allowed seamless integrations between internet services. Websites became interoperable. They also became dynamic, able to refresh automatically. “Mash-ups” of applications and data were suddenly everywhere. The web had become fluid.
Richard MacManus put it best7 in the first post on his early, influential tech blog, ReadWriteWeb, in April 2003. “The web was never just supposed to be a one-way publishing system, but the first decade of the web has been dominated by a tool which has been read-only—the web browser,” he wrote. “The goal now is to convert the web into a two-way system. Ordinary people should be able to write to the web, just as easily as they can browse and read it.”
That the internet could be more than a read-only medium inspired and invigorated a new generation of builders and users. That this reimagination of the internet could allow anyone to easily create content and put it in front of large audiences—not only to take in information, but to broadcast it—opened whole new fields of possibilities. And so the web would begin its next phase, enabling people to consume and publish, freely, and at unprecedented scale: activities that had no precursor in the pre-internet world.
The read-write era, also known as Web 2.0, had arrived.
The Rise of Corporate Networks
The read-write era also marked a shift in network design. Some technologists stuck to the open protocol network architecture, building new protocols and, on top of that, apps. But the developers who would be most successful pursued a different tack: the corporate network model.
Corporate networks have a simple structure. In the middle, a company controls centralized services that power the network. This company has complete control. It can rewrite its terms of service, determine who has access, and redirect how money flows, at any time, for any reason. Corporate networks are centralized because there is ultimately one person, usually the chief executive officer, who makes all the rules.
Users, software developers, and other participants are pushed to the network’s edges, where they are subject to the central corporation’s whims.
The corporate network model allowed a new generation of builders to move faster. Developers could quickly ship features and iterate, instead of waiting to coordinate with standards groups and other stakeholders. They could create advanced, interactive experiences by centralizing services inside data centers. And crucially, because the prize of owning a network was irresistible to venture capital investors, they could raise capital to invest in growth.
In the 1990s, internet startups ran many experiments, but by the 2000s it became clear that the best business models involved owning a network. eBay showed the way.8 Founded in 1995 as AuctionWeb, the company quickly became a stock market darling,9 and people considered it a case study for how valuable networks could be. The company was more profitable than Amazon,10 its main competitor, and most people believed it had a better business model. eBay had a strong network effect and didn’t hold inventory, reducing its costs. Amazon had a weaker network effect and held inventory, resulting in higher costs. The success of eBay, along with other network-effect-harnessing businesses, like PayPal (which eBay bought in 2002 and would spin out thirteen years later), kicked off a wave of venture capital funding for startups trying to create networks.
YouTube’s story illustrates the rise of corporate networks. In the mid-2000s, broadband home internet started going mainstream11 as infrastructure improved and costs fell. High-quality video streaming became practical for ordinary users. Entrepreneurs took notice and started building internet video startups. Some of these startups enabled existing video providers, like TV networks, to stream online. Others supported open protocols, like Media RSS and RSS-TV, which were multimedia extensions of RSS. Still others built their own corporate networks around “social video,” making it easy for anyone with an internet connection to publish videos.
YouTube championed this last approach. The service started as a video-dating site12 before expanding its focus to video broadly. YouTube’s first hit feature enabled users to embed videos in their own websites. At the time, YouTube’s website had a small audience. To the extent video creators had followings, it was usually on their own websites. Hosting video was expensive and complex, and YouTube made it free and easy.
YouTube’s video-embedding product is an example of a tactic I call “come for the tool, stay for the network.”13 The idea is to attract users with a tool that piggybacks on existing networks, like the websites of video creators, and then to entice those users to participate in another network, like YouTube’s website and app. The tool helps a service reach a critical mass, at which point network effects can kick in. Over time, the alternative network becomes more valuable than the preexisting networks—and harder for competitors to replicate. The tool might get better as the company adds features, but the value of the network increases much more rapidly, at a compounding rate. Today, many services host videos for free, but YouTube stays ahead because the service has such a large audience—which is to say, a large network. The tool is the hook, but the network is what creates long-term value for users and the company.
Fledgling corporate networks regularly employ this tactic. Instagram’s initial hook was a free photo-filtering tool. Other apps offered photo filters at the time, but most of them cost money. Instagram made it easy to share touched-up photos on existing networks,14 like Facebook and Twitter, while simultaneously sharing them on Instagram’s network. Eventually, people stopped bothering to share photos anywhere but on Instagram.
YouTube demonstrated the power of this strategy. The service drew in content creators by subsidizing the storage and bandwidth costs of streaming video. Any videos could be uploaded to YouTube and played on any other website for free. The company reasoned that the upside to controlling the network for internet video distribution would exceed the cost of providing a video-embedding tool.
Still, YouTube needed someone to foot the bills. Running a business hosting so many videos was an expensive proposition, and raising outside funding wasn’t a surefire solution. Venture capital was a much smaller industry in the mid-2000s, and it was suffering from the aftereffects of the dot-com crash. As users uploaded copyright-infringing material, YouTube also faced existential legal challenges.15 So, in 2006, YouTube sold to Google, a business flush with ad money whose visionary founders recognized the network’s potential as well as its synergy with their existing business. The bet paid off. Today, YouTube contributes more than $160 billion to Google’s market cap,16 according to various Wall Street analyst estimates.
Subsidization helps explain why protocol networks have so much difficulty competing with corporate networks. Services tied to a community-supported network, like RSS, have no source of financing that can subsidize hosting costs to the extent that company-supported networks can. The treasuries of donation-based projects pale in comparison to the war chests of Big Tech giants. Giving away the tool makes financial sense only when the company doing the subsidizing—not the community—will own the ultimate prize: the network.
The Problem with Corporate Networks: The Attract-Extract Cycle
If you ask people to name a specific example of corporate competition, they will likely cite a rivalry between makers of similar products: Coke versus Pepsi. Nike versus Adidas. Mac versus PC. Products that are essentially interchangeable are, in business lingo, called substitutes.
Competition between substitutes is straightforward. A meal at either McDonald’s or Burger King will (probably) satisfy a person’s appetite, and you wouldn’t expect a typical customer to visit both restaurants during a lunchtime rush. Similarly, someone might buy a pickup truck from Ford or GM, but that same person probably won’t splurge on both at once. If a customer is going to buy a single product, businesses fight to make sure it’s their own.
Products that are bundled or used together are, in contrast, called complements. Coffee and cream are complements. So are spaghetti and meatballs, cars and gasoline, and computers and software. Social networks and content creators, like YouTube and MrBeast, the host of one of YouTube’s most popular channels, are complements too. The pairings reinforce the value of the parts: what’s a hot dog without its bun, or an iPhone without apps?
One might expect such couplings to be the best of friends, but complements are, in fact, the greatest of frenemies. If customers are willing to pay no more than a fixed amount for a given bundle, complements will fight to claim the greatest share of sales from that bundle. The battles between complements can be brutal, zero-sum affairs. Indeed, some of the most ferocious competition in business occurs between bedfellows.
Imagine a hypothetical clash between food truck suppliers, for instance. If a customer is willing to pay $5 for a hot dog, savvy sausage makers will maneuver to capture more of the $5 that might otherwise go to the bun baker next door. Perhaps the butchers will buy bread wholesale and underprice the competition by pairing cheaper buns, gratis, with wieners. Or maybe they’ll market trendy ways of eating hot dogs sans buns, as in organic, gluten-free wurst. Irate bakers might, in retaliation, raise livestock and flood the market with meat to lower the price of a sausage relative to bread. Or maybe they will introduce vegan franks to cut the butchers out entirely.
These examples are silly, but the point is more value for one complement means less value for the other. Both sides will jockey to get ahead as they seek to grow the market for hot dogs in what can be described only as a, yes, dog-eat-dog world.
Network effects complicate the competition between corporate network complements by setting up incentives that are in conflict. On the one hand, a corporate network’s complements help grow the network and strengthen its network effect. On the other hand, a corporate network’s complements can siphon away revenue that might otherwise have gone to the network owner. The tension between these goals almost always causes the relationship between corporate networks and their complements to snap.
In the 1990s, Microsoft provided a high-profile demonstration17 of this when it made strategic moves against complements of its operating system, Windows. Microsoft wanted third-party application developers to build on Windows, but it didn’t want individual applications to get too popular. When an app started doing well, Microsoft would bundle a free version with Windows, as it did with its Microsoft-branded media player, email client, or, most famously, internet browser. Most of the third-party apps that survived these attacks were too small for Microsoft to care about. From a profit maximization perspective, the best outcome for a platform like Windows would be to have lots of smaller complements with weak, fragmented power, but for the sum of those complements to make the platform more valuable as a whole. (Microsoft’s complement-crushing strategy was a major reason why the U.S. Department of Justice accused the company of antitrust violations in 1998.)18
Social networks also have a history of conflict with their primary complements, content creators. Consider modern ad-based social networks that seek to maximize profits. Most social networks have high fixed costs covering software development and infrastructure. The marginal costs are low: adding more servers and bandwidth generates more revenue than it costs. For the most part, raising profits boils down to raising revenue. Simple as that.
Social networks can maximize revenue in one of two ways. The first is to grow the network. The most effective way to do this is to create a positive feedback loop where more content leads to more users, and more users leads to more content. It’s a virtuous cycle. If people spend more time on the network, the company can make more advertising revenue.
The second way for a social network to maximize revenue is through promoted content. Social feeds generally consist of two kinds of content: organic and promoted. Organic content shows up in users’ feeds through the usual algorithmic processes. Promoted content appears because creators have paid to feature it. Social networks can juice revenues by getting more content creators to pay up. The networks can charge more per promotion, and they can also pad users’ feeds with more sponsored content. The risk is that at some point the strategy could degrade the user experience and exceed people’s tolerance for ads.
A common tactic social networks use to get content creators to promote more content is to let creators achieve a certain scale of audience, and then to adjust the algorithm so the creators no longer receive the same levels of attention organically. In other words, once creators are generating meaningful revenue and have become economically dependent on the network, the network owners dampen the creators’ reach so they are forced to buy sponsored posts to maintain or grow their audience. This makes growing one’s audience increasingly expensive over time. Content creators call the move a bait and switch, and if you talk to them, you’ll hear the complaint regularly.
Companies face the same problem. If you read the regulatory filings of public companies that advertise on social networks, you’ll see that the marketing costs for most are rising.19 Social networks are very good at extracting maximum profits from their most important complements: content creators (including advertisers). This doesn’t mean the bait and switch is a nefarious conspiracy by corporate management. It just means that corporate networks will end up behaving this way if they are smart about profit optimization. Why is the pattern so consistent and enduring? Because only the networks that are smart about profit optimization survive.
Independent or third-party software developers are the other important category of social network complements. Developers are valuable to networks because they outsource the production of new software. Social networks often encourage the growth of third-party apps at first.20 Later the networks identify the apps as a competitive risk and cut them off, just as Facebook once did to Vine and others.
Corporate networks that don’t crush complements will often copy or, sometimes, acquire them. When Twitter released its first iPhone app in 2010, it put out a rebranded version of Tweetie,21 a third-party app it had acquired that year. Soon after, Twitter deprecated features available to other third-party apps,22 including a variety of feed readers, dashboards, and filters. Developers felt betrayed.23 Andrew Stone, the founder of one affected app, Twittelator, told The Verge in 2012, “Whatever perceived gains that might be achieved by eliminating the third parties should be weighed against the lingering public perception that Twitter got greedy.”
Twitter was acting, Stone added, like “the mythological Greek Titan Cronos, [who] began eating each of his children as they were born.”
Building startups on top of social networks was widespread in the second half of the 2000s, before the reversal. Conventional wisdom among startups held that besides mobile phones social networks were the next big platform for entrepreneurs. Many of the hottest startups at the time, such as RockYou (ad network),24 Slide (social app maker),25 StockTwits (stock market tracker),26 and UberMedia (another social app maker),27 were built on top of social networks. Many founder friends of mine were building startups and applications on top of Facebook, Twitter, and other social networks back then. Even Netflix introduced an API in 200828 to encourage third-party development, until shutting it down six years later.
Building on Twitter was particularly popular. People considered it the most open of the corporate networks—that is, until the company changed its policies29 and killed its developer ecosystem. I worried at the time about startups depending too much on Twitter,30 a concern I expressed in a 2009 blog post, “The Inevitable Showdown Between Twitter and Twitter Apps.”
I should have heeded my own advice. My second startup, Hunch, an artificial intelligence company I co-founded in 2008, depended on Twitter’s API. Hunch learned users’ interests and offered product recommendations based on Twitter data. My co-founders and I sold the company to eBay in 2011, partly because so much of the open data we depended on was becoming unavailable. (eBay had its own data it could feed into our machine learning tech.)
The transition from open social networks to the closed versions people are familiar with today dates to 2010. One tell, as I noted at the time: Google started warning users31 who attempted to export their Google Contacts to Facebook, “Hold on a second. Are you super sure you want to import your contact information for your friends into a service that won’t let you get it out?” At the time Facebook let users download their personal information—photos, profile info, and so on—but only as an unwieldy.zip file. Facebook made no easy-to-use, interoperable API available. The company was clamping down on its social graph, preventing anyone from easily downloading friend lists. Google blasted Facebook’s policy as “data protectionism.”
As corporate networks clenched their fists, venture capital funding for applications built on top of social platforms dried up. If these networks wouldn’t let anyone building on top get too big, then why invest there? It was very different from the era of protocol networks, like the web and email, when everyone trusted the networks to remain accessible and rent-free in perpetuity and understood that they could get as big as a market allowed. The arrival of corporate networks ended those implicit promises. Building on a corporate network was like building on a broken foundation. The term of art to describe this hazard of the new era: platform risk.
Without third-party developers, corporate networks must rely solely on their own employees for new product development. Look no further than Twitter to see the consequences of misaligned network incentives in action. More than seventeen years after its founding, Twitter is still wrestling with nasty spam problems. Bill Joy, the co-founder of Sun Microsystems,32 once famously observed that no matter who you are, most of the smartest people work for someone else. When email had a spam problem, smart people who worked for someone else (or often for themselves) came to the rescue. No cavalry would come for Twitter. Platform risk scared everyone away.
Almost all new technologies follow an “S-curve,” a growth-over-time chart that resembles the letter S. The curve starts out flat in the first phase, as a technology’s developers search for a market and find early adopters. As the builders find product-market fit, the curve begins to tilt up quickly, reflecting mainstream uptake. The curve then flattens again as the product saturates the market.
Network adoption tends to follow an S-curve. As networks climb the curve, the relationship between corporate networks and their complements unfolds in a predictable pattern. The engagement starts out friendly. Networks do everything they can to recruit complements like software developers and content creators to make their services more compelling. The network effects are weak at this early stage. Users and complements have many choices, and they’re not locked in yet. The perks are flowing, people are happy, everything’s kumbaya.
Then the relationship sours. As the network moves up the S-curve, the platform begins to wield more power over users and third parties. Network effects strengthen, but growth slows. The relationship between the platform and its complements turns hostile. Positive sum becomes zero-sum. To keep profits coming, platforms start capturing more of the money that flows through the network. This is what happened when Facebook strangled Vine and other apps, and when Twitter swallowed its third party offspring whole. Platforms eventually cannibalize their complements.
An example helps explain why bigger networks often stop interoperating. Suppose you have two networks, a smaller one with ten nodes, A, and a larger one with twenty nodes, B. If the two networks interoperate, they will both have thirty nodes. There are different ways to approximate the value of a network. Let’s use Metcalfe’s law, which you may recall states that the value of a network varies with the square of the number of nodes. When interoperating, A’s value vaults to nine hundred (thirty nodes squared) from one hundred (ten nodes squared). B gains less. It also reaches a value of nine hundred (thirty nodes squared) but from a base of four hundred (twenty nodes squared). So A becomes 9 times as valuable, while B becomes only 2.25 times as valuable. A gets a much better deal.
This is a simple example, but it shows why, as a network grows, adding complements and interoperating with other networks become less attractive. The moment a platform has maximum leverage is the same moment it makes sense to do an about-face. Bigger networks have less to gain and more to lose by interoperating. Why boost potential competitors?
Facebook’s fraught relationship with a once-close partner, the game maker Zynga, demonstrates these concerns. For years after its founding in 2007, Zynga was the social network’s biggest sensation. Hits like Zynga Poker, Mafia Wars, and Words with Friends attracted tens of millions of players. Alluding to Zynga’s first major breakout game, FarmVille,33 in a 2011 post for New York magazine, one writer described the company’s popularity thus: “Pretty much anybody who has been on Facebook long enough, which these days is almost everybody, has at some point received a request to adopt a cow.”
For Zynga, virtual cows were a cash cow. By 2012, the company had grown to account for double-digit percentages of Facebook’s revenue,34 including by selling users digital livestock. Wall Street analysts called out Zynga’s outsized contribution to Facebook’s top line as a significant risk. The game maker could lure people away to a gaming platform of its own, after all. So Facebook diversified its revenue35 and ripped up its partnership with Zynga,36 nearly killing the company. (Zynga later restarted its business after a years-long turnaround effort and, in 2022, another gaming company, Take-Two Interactive, bought it for $12.7 billion.)37
The lesson: big networks can gain from interoperating under the right circumstances, but rivals may gain more. The trade-off favors cooperation early and competition later.
I call this the attract-extract cycle. Corporate networks obey its logic without fail. For complements, the transition from cooperation to competition feels like betrayal. Over time, the best entrepreneurs, developers, and investors become wary of building on top of corporate networks. Decades of evidence show that doing so will end in disappointment. It’s impossible to quantify how much innovation this has cost the world. The closest window into the alternative universe where corporate networks remain community owned is to look at the entrepreneurial activity that continues to build on email and the web, which remains considerable even after all these decades. Every year, entrepreneurs create millions of websites and newsletters alongside new software companies, media enterprises, small business e-commerce sites, and more.
Some startup founders and investors, feeling burned, have turned away from the corporate network model—myself included. I know many well-intentioned people who work for corporate networks. The problem is not the people. It’s the model. The interests of the company and network participants are simply misaligned, resulting in a worse experience for the user. A corporate network that doesn’t run the bait-and-switch strategy will be squashed by competitors that do.
Opacity is another downside of corporate networks. People lose trust when functions like algorithmic rankings, spam filtering, deplatforming, and other decisions are managed inside a black box by for-profit entities. Not sure why your account got suspended? Or why your app was rejected from an app store? Or why you no longer seem to have as much social clout as you once did? Corporate networks have become critical tools affecting people’s lives, and they are constant subjects of debate and frustration. Management might change, sometimes sharing your values and other times not. Again, the real problem is the model. Everyone is at the whim of corporate platforms.
Compare this with the transparency of protocol networks. Email and the web are governed by a coalition of entities that enforces laws, as well as communities of users and software developers that make technology decisions. Both processes are open and democratic. Client software is free to add moderation and filtering. If users don’t like the way the software works, they can switch to new software without losing their connections. The power is in the hands of the community. Expanding stakeholdership builds trust.
On the bright side, corporate networks like Facebook, Twitter, LinkedIn, and YouTube played a significant role in helping to grow the internet over the last twenty years. The iPhone’s introduction in 2007, and the App Store’s debut a year later, led to a wave of useful networks that included WhatsApp, Snap, Tinder, Instagram, and Venmo. These corporate networks helped bring advanced services to five billion internet users.38 They enabled anyone with internet access to become a publisher, build an audience, and potentially make a living. Corporate networks drastically lowered the barrier to entry for people to reach broad audiences in ways that were less specialized and labor-intensive than website making and much more effective than using email alone. In this way, corporate networks improved upon protocol networks. The web’s second era helped achieve the dream of the technologists in the early 2000s to upgrade the internet from “read-only” to “read-write.”
Corporate networks beat protocol networks because of superior features and sustainable funding. Only email and the web, legacies of the early internet, have resisted the centralizing forces of corporate networks, thanks to their unique history, longevity, and entrenched customs—an instance of the “Lindy effect,” where the longer something has been around, the likelier it is to stick around. (Although there is always the possibility that even these protocol networks could get subsumed by corporate networks, even if it’s hard to imagine.)
More recent protocol networks enjoy no such resiliency. No credible protocol network, after thirty years of attempts, has succeeded beyond niche adoption. Newer protocol networks are a rarity, and those that technologists do create invariably struggle to gain traction. Corporate networks colonize and overtake new protocol networks like kudzu. Successful networks succumb to the inescapable, profit-driven logic of the attract-extract cycle—just as happened in the case of Twitter versus RSS and so many other examples. The corporate model has simply become too effective.
But software is a creative medium with boundless room for exploration, and the internet is still early in its development. New network architectures can address the problems created by corporate networks. Specifically, networks built on blockchains can combine the best features of prior networks, benefiting builders, creators, and consumers and ushering in a third era of the internet.