“If your blockchain goes down because AWS goes down, you’re not decentralized.”
— Ben Schiller, CoinDesk, Oct 2025
🧭 The Outage That Shook the Backbone
Bla bla bla, we've all heard the news. Maybe your Alexa broke. Whatever. For context:
On October 20–21, 2025, AWS suffered a major outage centered in the us-east-1 region.
Snapchat, Fortnite, Roblox, and even parts of Amazon itself went dark. Bla bla bla.
For nearly fifteen hours, the internet’s most “redundant” infrastructure was anything but.
If AWS sneezes, the internet catches a cold.
(I don't know who said that FYI).
This time the fallout went deeper — blockchains, RPC endpoints, and validator APIs started failing too.
The world’s “decentralized” systems suddenly felt very centralized.
🔍 The Tesla Generator Paradox
Web3 today is a bit like running a Tesla off a petrol generator. Which apparently people do, someone on Twitter even fact checked this image so it must be true.
Everyone is banging on about decentralisation but peel back the layers and you’ll find the same old fossil infrastructure humming underneath. I've been saying this for years, including during my tenure as a CTO at a DeFi startup. So, I must be right.
📊 What the Data Says
Because I haz got the datas, I did look at them datas!
If you've been reading my thrilling articles about the fact Sui is probably going to collapse quite soon, you'll know what's coming. I did some analysis of 122 Sui validators and public nodes - it reveals moderate decentralisation on paper — but concentration in practice.
- Latitude.sh: 18.0 %
- OVH: 18.0 %
- The Constant Company: 7.4 %
- Top 3 providers = 43 % of all validators
- US jurisdiction: 31 %
- Western Europe (UK + DE + FR + NL + IE): 41 %
- AWS + GCP + Azure combined: 10.7 %
The Herfindahl–Hirschman Index (HHI) is 859 — “unconcentrated” by regulatory standards but that’s misleading when dozens of nodes share the same upstream providers, fibre routes, and power grids.
BTW - I learned myself something today - HHI - fancy new term for Simon!
You can see all the data here:
https://github.com/pgdn-oss/pgdn-research/blob/main/reports/2025-10-sui-decentralisation.md
Oh and here is something fun - Sui once told me that their infra runs in this super private network in Switzerland but actually they have their stuff in Google Cloud. Sweet.
"On paper it’s diverse. In practice, it’s the same house with many doors."
🧩 Web3 Built on Web2 Foundations
The irony is painful: even networks designed for fault tolerance often rely on a few hyperscalers for uptime. Hyperscalers! (ChatGPT put this word in here, I decided it sounded cool).
When AWS stumbles, so do RPC providers like Infura, Alchemy, and QuickNode.
When Cloudflare misconfigures, half of Solana RPC endpoints vanish.
When OVH catches fire (literally, in 2021), validators go dark.
It’s decentralisation in code, centralisation in practice. Fact. I think it's the damn thought leaders again saying words.
⚙️ What Real Infrastructure Decentralisation Requires
- Jurisdictional diversity — not just different countries, but different regulators and risk domains.
- Bare-metal or sovereign hosting — own or colocate hardware. Don’t rent from hyperscalers. Actually lattitude.sh claim to be be bare metal but the point is we need diversity!
- Multi-provider topology — AWS + OVH + local datacenters + community nodes.
- Transparent infrastructure maps — disclose where nodes live and how they’re connected.
- Resilience testing — simulate region loss, BGP leaks, and power faults.
- Agentic monitoring — track correlated risk across clouds, not just node count. I've put this in because this is what I am building, hint hint.
🧠 The Hard Truth
We call it Web3, but until our nodes can survive an AWS outage,
it’s still Web2 in disguise — centralised scaffolding painted in decentralised colours.
Decentralisation isn’t about the number of validators.
It’s about how many can survive when the lights go out.
The good news?
Tools like agentic scanners, peer diversity metrics, and open telemetry are starting to expose these weak points. The next step is acting on them — before the next outage does it for us.
Top comments (0)