The Illusion of Zero Marginal Costs

No free lunch.

Hi. This is your host Can.

8chan, Be Gone

If you are a network provider and one of your customers is using your services to host content for murderers, should you keep providing service to that customer? Unfortunately, Cloudflare found itself asking this question once again this week, following 2 vile mass shootings in less than 24 hours. It was a whirlwind. First, they said they would keep on keeping on, but then quickly changed course and decided to kick 8chan off of their network.

Does this mean 8chan is kicked off The Internet?

Probably not, and even the CEO conceded on this. Cloudflare might be the flashiest company in the block, but there are many others that provide similar services. As the Chief Executive Matthew Prince argues, other companies that Cloudflare severed its ties still exist.

The real reason why this is worth discussing is, though, Cloudflare’s critical position on the global internet. The company itself touts the same in its blog post explaining its 8chan decision

As a result of that, a huge portion of the Internet now sits behind our network. 10% of the top million, 17% of the top 100,000, and 19% of the top 10,000 Internet properties use us today. 10% of the Fortune 1,000 are paying Cloudflare customers.

It might feel natural to think of The Internet as a conglomeration of few systemically important companies. It certainly feels that way. Google, for example, owns a giant share of the search market. Facebook dominates the social networking space. Amazon owns a huge chunk of the online retail, and a big chunk of the internet runs on its cloud services. It’s hard to think of the internet as we know it today without any of these companies.

Yet, this is an aberration. Or at least, it does go against the initial design of the internet, which was supposed to be decentralized. In fact, decentralization was not just a feature of the internet. Rather, internet itself was born out of a necessity to build a decentralized communication network.

Packet Switching

Somewhat unsurprisingly, this specific requirement was borne out of military needs. US government needed a way to build a network that can withstand a major disruption, say, a nuclear attack that takes out many of its pieces. The main innovation that enabled this feat is called “packet switching”. Let’s talk about it, because, why not?

The idea is simple. Whenever you want to send a message, you first divide the contents into small chunks, called “packets”. Each packet contains both part of the message (i.e. payload) and some extra metadata that helps put the packets together (i.e. header). Then, sending the message becomes an act of releasing the packets into the network. Each node makes its best attempt at routing the packets into their destination.

One way to conceptualize this is to contrast it to a phone call. Instead of the source and the destination establishing a dedicated connection like a phone call, in a packet switched network, you just spread your message and “hope” it makes to the end, without a direct connection.

This “connectionless” network is essentially how the internet runs today. There are no hubs, but instead tons of machines running in tandem for the greater good of the network. You can, in theory, take out most of the nodes and as long as there’s a single path from your source to the destination, your packets will arrive and they will be correctly ordered. It is an interesting collectivist dream in action, enabled by technology.

Yet, as Internet evolved from a military network to nerds in college campuses to AOL chatrooms to essentially the backbone of modern economies, it became subject to and of not just technological forces but also economical ones. More than anything, ever diminishing marginal costs are one of the big reasons why we have a centralized internet.

Zero Marginal Costs

In order to understand the effect of diminishing margins on centralization, consider the two types of costs associated, namely fixed costs and marginal costs. Grossly simplified, in order to turn a profit, a company first needs to cover its fixed costs, and then make a small margin to cover its marginal costs on each sale.

In an industry where there are big marginal costs but small fixed costs, a competitor can enter your market by offering the same product at a smaller profit margin by simply lowering its marginal cost. In other words, the entry point for a competitor is the marginal cost.

This presents a challenge, however, when there are no marginal costs. In the digital era, where bits are practically free to move around and increasingly free to store, there are increasingly smaller marginal costs associated with any product or service. In this environment, it’s the fixed costs that dominate the competitive environment.

From the lens of competition, the only entry point for a rival is to offer a product at a smaller fixed cost per user. Now, you can do this in of two ways; you can either lower your fixed costs by…lowering your fixed costs such as salaries and capital expenditures. Of course, the you can instead spread those fixed costs on more and more customers. Thus, your fixed cost per customer goes lower as you add more customers, making you more profitable compared to the competition.

Obviously, this is easier said than done, but you can see the aesthetic appeal of this kind of business. Ben Thompson’s Aggregation Theory, for example, is largely build on this notion of diminishing marginal costs. People in Silicon Valley love to talk about network effects, where the currency is data but a low marginal cost business actually gives you a monetary network effect where your sheer size gives you a competitive advantage. Once you pull ahead, it is much harder for your competitors to catch up.

Zero Competition

Of course, companies come and go and industries change. Companies get complacent and miss the ground shifting under them; technological changes make industries irrelevant overnight, or sometimes consumers simply change their preferences. However, all things being equal, in digital economies where fixed cost dominate, centralization is a natural force.

Whether you think this is a problem, and how big of a problem it is depends heavily on your point of view. It might make sense to have a few competitors to keep the regulators at bay, but have too many and your profits are driven to zero. If you are a consumer, then the opposite holds true. More competition means better and cheaper stuff.

If you are interested in the health of the internet as a global, interconnected communication medium however, you’d generally want more competition and less centralization, largely for technical reasons. As we go back to the packet switching discussion earlier, more competition (i.e. less centralization) means a healthier network. The more computers owned by different people that are loosely connected to each other means harder for internet to break.

In the context of Cloudflare and internet infrastructure , this could not be more obvious. Just a few weeks ago, the same company had a 30 minute downtime where significant portions of the internet simply went down. Cloudflare’s public post mortem on the issue traces it to a single misconfigured regular expression (a way to filter data) that increased CPU load. Cloudflare, to its credit, identified the problem and resolved it relatively quickly. Yet, it’s likely that such outages will only get bigger in impact in the future.

Hidden Costs of Zero Margins

We often talk about hidden liabilities in data on The Margins; the main idea isn’t that data is not without value but it also comes with risks that aren’t immediately accounted but surface later. There’s probably a similar point to be made about the seemingly zero marginal costs of doing business online.

As more sales channels become self-serve, more and more businesses move to automate more of their moderation and compliance efforts, more of these risks will surface. At pixel time, not just Cloudflare, but many other suppliers of 8chan were slowly cutting their ties with the company. For most of those companies, the idea of being associated with such a vile, sick website was simply not worth the downsides, be it the PR backlash, losing employees who do not work at there, or simply the moral baggage of being complicit in senseless violence.

It will be up to all of us, as consumers, investors, inventors and builders of the future economy to fully grasp the impact of our work. Internet was designed to be decentralized, and it flourished as such. We might need more structure on it now, as it became much more than what its initial creators thought it’d become. Similarly, we might need more regulation for the industry, and ideally some globally accepted standards. The worst thing to do, however, would be to do nothing.

Personal Note📝

After a whirlwind of travel and short-term leases spread over 3 countries and 5 cities, I am back in San Francisco. It’s where I lived longest, where I have most of my friends, and whatever problems I have with it, where I feel home. I am also looking for employment now as a product manager. If you wanted to reach out and meet in person, this is as good as a time as any. My personal email is

What I’m Reading

Universal Basic Income: How and Why: OK, technically listened to this one. Capitalisn’t is my all-time favorite podcast. In this episode, Luigi Zingales and Kate Woldack discuss a Silicon Valley favorite: Universal Basic Income. I especially found the argument that how UBI is financed is much more important than what it would do for people fascinating and thought-provoking. You really owe yourself to listen to this episode, and probably catch up on the back catalogue too.

The disinformation age: a revolution in propaganda: Peter Pomerantsev is a Soviet-born British journalist who writes on information warfare. His new book discusses how in this age, suppression not comes in the form of hiding information, but rather drowning it. This is an argument made often by the likes of Tim Wu also. This excerpt piece in the Guardian is sharp in its writing and depressing in its content. I am looking forward to reading the book entirely.

Gloria remembers how in Marcos’s time you could see the enemy. There was a sort of predictability: they could kill you, or you could skip town, contact a lawyer, write to a human rights group, take up arms. You knew who the agents were, who was coming for you, who your enemy was. But now? You couldn’t tell who you were up against. They were anonymous, everywhere and nowhere. How could you fight an online mob? You couldn’t even tell how many of them were real. And of course this allowed the government to claim they had nothing to do with these campaigns. Wasn’t it just a question of concerned citizens exercising their right to free speech?