On a Thursday at the end of October, Roblox, a platform that sustains a virtual world for over 160 million monthly users and 8 million creators, went down. It stayed that way through most of the weekend.

Sure, Roblox is just a gaming platform for kids, but the outage offers a glimpse of what it might be like if things went down in the metaverse, a future version of the internet featuring 3D virtual spaces. The economic stakes are high already on Roblox — it supports its own internal economy worth hundreds of millions of dollars. Problems with a similar outage would be exponentially more damaging in the metaverse.

Even worse, we still don’t know what exactly caused the Roblox outage. Some observers have speculated that it was triggered by an overload of traffic from a Chipotle promotion that was running at the same time, though Roblox has denied it. This raises the specter of a similar confluence of issues arising in the metaverse, and helps make an argument in favor of building a stronger internet with Web3.

The Roblox outage was far from the only similar event affecting a major internet platform recently. Amazon Web Services, the cloud infrastructure arm of the e-commerce giant, went down three times in December alone. The AWS outages have brought with it interruptions to Slack, and Asana—important professional services for companies trying to get stuff done. 

There was also a massive Facebook outage in October, where all the services owned by the company then known as Facebook—that’s Instagram, WhatsApp, Oculus and Facebook itself—were inaccessible for over six hours. When you consider that Facebook has nearly 3 billion monthly active users, an almost unfathomable amount of stuff got interrupted during those six or seven hours.

Another problem outage occurred in June, when “edge cloud platform” Fastly went down for about an hour, taking down numerous highly-trafficked websites with it, including for Slack, Stripe and The New York Times.

The Facebook incident in particular highlights the risks of centralization. The firm’s engineering team explained that, essentially, one of the engineers sent the wrong command during routine maintenance, causing all of the company’s data centers to be disconnected. This led to a cascade of disconnections in other systems, including those that allow Facebook’s servers to be found on the internet at all—a core part of the internet known as the Border Gateway Protocol. 

Essentially, a single wrong command by an engineer at Facebook had rendered the massive platform invisible to everyone else on the internet. 

The more decentralized a system is, the less likely it will meet the fate of a Facebook or Roblox outage: taken out by a single erroneous command. While “Web3” remains a contested term—not least among the titans of Silicon Valley and Web2—its central tenet is decentralization. A more decentralized system is meant to be more robust and, in the words of the economist Nicholas Nassim Taleb, “anti-fragile.”

Indeed, blockchains are arguably the most anti-fragile structures on the internet. They’re designed to be permanent, immutable, and unstoppable. It’s appropriate, then, for Web3 to be built on blockchain foundations. 

But a blockchain’s durability, and its immutability, is only as robust as its degree of decentralization. If a blockchain is supported by a handful of nodes, it’s easy to get them to reorder a transaction, or roll back the chain. If it’s underpinned by thousands of independent nodes, it’s a lot harder to get them all to jump at the same time, thus strengthening its immutability. 

So, how decentralized, exactly, are blockchains today? We’ve seen outages or slowdowns with blockchain networks like the super-fast Ethereum competitor Solana in recent months, for instance. And during various past crises in Ethereum and Bitcoin – such as when The DAO was hacked in 2016 or when the “blocksize war” fiercely divided the Bitcoin community for years – numerous investors, startup founders, and community leaders were accused of undermining the ethos of decentralization behind those chains. 

But perhaps decentralization doesn’t exist as a static fixed point or goal. Instead of seeing a blockchain or platform as centralized or not, we can consider it part of a process of decentralization and recentralization. This is the framework offered by Web2 sage Tim O’Reilly in a recent essay on the viability of Web3

In O’Reilly’s teling, the personal computer offered a wave of decentralization: users didn’t have to go to large corporations or university labs to use computers anymore. But then the competition for operating systems on the PC was ultimately won by Microsoft—a wave of recentralization. The internet itself is a decentralized communications network that bested centralised alternatives of the day, so-called intranets. But it enabled the rise of massive data monopolies like Facebook and Google—another wave of recentralization. 

Applying O’Reilly’s lens, the question facing blockchains today is whether each layer of their technology stacks are decentralizing or recentralizing, and the impact of each. 

For instance, bitcoin mining has been dominated by a single entity at various points in time, from Ghash to Bitmain. More recently, hashrate fragmented when it left mainland China and is currently in the process of reconsolidating. On Ethereum, much has been said about the centralising forces of cloud infrastructure such as Infura, which overwhelmingly power the myriad decentralized applications that live on the Ethereum blockchain. When Infura goes down, so do lots of dApps—although Ethereum itself isn’t down. 

What would it mean, then, to build a better internet? If Web3 is going to really mean something, it will be because platforms and applications won’t collapse because of a fat-finger error by an engineer at one corporation. Blockchains hold the promise of introducing new layers of technology atop the decentralized foundation of the internet. But fulfilling this promise requires both users and builders to understand that each layer is always in the process of decentralizing—or doing the opposite.

Ad placeholder