On the 35th anniversary of Tim Berners-Lee's first WWW announcement, this piece examines how a web designed for radical decentralization became controlled by a handful of companies. Six compounding forces drove centralization: DNS chokepoints, certificate authority oligopolies, Google's search monopoly, app store gatekeeping, cloud consolidation (AWS/Azure/GCP), and surveillance-based advertising. The original design lacked identity, payment, and reputation layers, which platforms filled. Re-decentralization efforts (Mastodon, IPFS, Web3, Solid) solve the technical problem but fail economically. AI is following the same centralization pattern on a 5-year timeline instead of 20, with NVIDIA controlling 80% of AI chips and 96% of AI startups depending on APIs from four companies. Five architectural principles are drawn for builders: technical openness isn't enough, chokepoints compound, convenience beats ideology, network effects must be explicitly countered, and early decisions set long-term patterns.
Table of contents
The Centralization Forces: Six Compounding DecisionsDNS: The First ChokepointCertificate Authorities: Trust as a BusinessSearch: Discovery Becomes DistributionMobile: The App Store GateCloud: Running Your Own Server Became OptionalAdvertising: Surveillance as Business ModelQuestions this post answers
What percentage of web traffic goes through Google search and how dominant is the first page?
By 2026, Google processes 8.5 billion searches daily, and the first page captures 95% of clicks. 91% of web pages receive zero organic traffic from Google. By 2010, 65% of web traffic already originated from Google searches, making search control effectively equivalent to controlling web access itself. Architects building discovery-dependent systems track these concentration dynamics on daily.dev.
What are the main reasons the web centralized despite being designed as a decentralized protocol?
Six compounding forces drove centralization: DNS registration creating naming chokepoints, certificate authorities forming a trust oligopoly, Google monopolizing discovery, Apple and Google gatekeeping mobile via app stores, AWS/Azure/GCP consolidating infrastructure (60% of cloud through three providers), and advertising economics rewarding scale. The original design also lacked native identity, payment, and reputation layers, which platforms filled by default. Teams designing distributed systems find the pattern of compounding chokepoints worth tracking on daily.dev.
How fast is AI centralizing compared to the web, and who controls the key chokepoints?
AI is centralizing in roughly 5 years versus the web's 20. NVIDIA controls 80% of the AI chip market, TSMC manufactures 90% of advanced chips, and 96% of AI startups depend on APIs from OpenAI, Anthropic, Google, or Amazon. GPT-4 training cost an estimated $100 million and Gemini Ultra $200 million, limiting frontier model training to roughly ten organizations. Developers choosing AI infrastructure dependencies watch these consolidation trends on daily.dev.