Discord: From 0 to 100 Million Users — The Architecture Evolution
How Discord scaled from a gaming chat app to 100 million users by rebuilding their architecture three times.
The challenge
Discord launched in 2015 for gamers who needed voice chat during games. Within months they had millions of users — far more than anticipated. Their initial monolithic Python application couldn't handle the load. Message delivery latency was unacceptable, voice quality was poor, and the system crashed regularly during peak gaming hours.
The strategy
Discord's engineering team made a decisive choice: rewrite their backend in Elixir (the same language family as Erlang used by WhatsApp) for real-time messaging, while keeping Python for non-real-time services. They embraced the microservices architecture but were selective — not every service needed to be split.
Want the full story, outcome and quiz?
Read how Discord executed it, the results, key lessons and test yourself with a quiz. Free on CaseLearn.
Try the full case free →Key lessons (preview)
- Choosing the right programming language for each workload type creates compounding performance advantages.
- Architecture must evolve with scale — what works at 1M users fails at 100M.
- Voice and text infrastructure have fundamentally different requirements and should be built separately.